Integrations
zenml.integrations
special
The ZenML integrations module contains sub-modules for each integration that we
support. This includes orchestrators like Apache Airflow, visualization tools
like the facets library, as well as deep learning libraries like PyTorch.
airflow
special
The Airflow integration sub-module powers an alternative to the local
orchestrator. You can enable it by registering the Airflow orchestrator with
the CLI tool, then bootstrap using the zenml orchestrator up command.
AirflowIntegration (Integration)
Definition of Airflow Integration for ZenML.
Source code in zenml/integrations/airflow/__init__.py
class AirflowIntegration(Integration):
"""Definition of Airflow Integration for ZenML."""
NAME = AIRFLOW
REQUIREMENTS = ["apache-airflow==2.2.0"]
@classmethod
def activate(cls):
"""Activates all classes required for the airflow integration."""
from zenml.integrations.airflow import orchestrators # noqa
activate()
classmethod
Activates all classes required for the airflow integration.
Source code in zenml/integrations/airflow/__init__.py
@classmethod
def activate(cls):
"""Activates all classes required for the airflow integration."""
from zenml.integrations.airflow import orchestrators # noqa
orchestrators
special
The Airflow integration enables the use of Airflow as a pipeline orchestrator.
airflow_component
Definition for Airflow component for TFX.
AirflowComponent (PythonOperator)
Airflow-specific TFX Component. This class wrap a component run into its own PythonOperator in Airflow.
Source code in zenml/integrations/airflow/orchestrators/airflow_component.py
class AirflowComponent(python.PythonOperator):
"""Airflow-specific TFX Component.
This class wrap a component run into its own PythonOperator in Airflow.
"""
def __init__(
self,
*,
parent_dag: airflow.DAG,
pipeline_node: pipeline_pb2.PipelineNode,
mlmd_connection: metadata.Metadata,
pipeline_info: pipeline_pb2.PipelineInfo,
pipeline_runtime_spec: pipeline_pb2.PipelineRuntimeSpec,
executor_spec: Optional[message.Message] = None,
custom_driver_spec: Optional[message.Message] = None,
custom_executor_operators: Optional[
Dict[Any, Type[launcher.ExecutorOperator]]
] = None,
) -> None:
"""Constructs an Airflow implementation of TFX component.
Args:
parent_dag: The airflow DAG that this component is contained in.
pipeline_node: The specification of the node to launch.
mlmd_connection: ML metadata connection info.
pipeline_info: The information of the pipeline that this node
runs in.
pipeline_runtime_spec: The runtime information of the pipeline
that this node runs in.
executor_spec: Specification for the executor of the node.
custom_driver_spec: Specification for custom driver.
custom_executor_operators: Map of executable specs to executor
operators.
"""
launcher_callable = functools.partial(
_airflow_component_launcher,
pipeline_node=pipeline_node,
mlmd_connection=mlmd_connection,
pipeline_info=pipeline_info,
pipeline_runtime_spec=pipeline_runtime_spec,
executor_spec=executor_spec,
custom_driver_spec=custom_driver_spec,
custom_executor_operators=custom_executor_operators,
)
super().__init__(
task_id=pipeline_node.node_info.id,
provide_context=True,
python_callable=launcher_callable,
dag=parent_dag,
)
__init__(self, *, parent_dag, pipeline_node, mlmd_connection, pipeline_info, pipeline_runtime_spec, executor_spec=None, custom_driver_spec=None, custom_executor_operators=None)
special
Constructs an Airflow implementation of TFX component.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
parent_dag |
DAG |
The airflow DAG that this component is contained in. |
required |
pipeline_node |
PipelineNode |
The specification of the node to launch. |
required |
mlmd_connection |
Metadata |
ML metadata connection info. |
required |
pipeline_info |
PipelineInfo |
The information of the pipeline that this node runs in. |
required |
pipeline_runtime_spec |
PipelineRuntimeSpec |
The runtime information of the pipeline that this node runs in. |
required |
executor_spec |
Optional[google.protobuf.message.Message] |
Specification for the executor of the node. |
None |
custom_driver_spec |
Optional[google.protobuf.message.Message] |
Specification for custom driver. |
None |
custom_executor_operators |
Optional[Dict[Any, Type[~ExecutorOperator]]] |
Map of executable specs to executor operators. |
None |
Source code in zenml/integrations/airflow/orchestrators/airflow_component.py
def __init__(
self,
*,
parent_dag: airflow.DAG,
pipeline_node: pipeline_pb2.PipelineNode,
mlmd_connection: metadata.Metadata,
pipeline_info: pipeline_pb2.PipelineInfo,
pipeline_runtime_spec: pipeline_pb2.PipelineRuntimeSpec,
executor_spec: Optional[message.Message] = None,
custom_driver_spec: Optional[message.Message] = None,
custom_executor_operators: Optional[
Dict[Any, Type[launcher.ExecutorOperator]]
] = None,
) -> None:
"""Constructs an Airflow implementation of TFX component.
Args:
parent_dag: The airflow DAG that this component is contained in.
pipeline_node: The specification of the node to launch.
mlmd_connection: ML metadata connection info.
pipeline_info: The information of the pipeline that this node
runs in.
pipeline_runtime_spec: The runtime information of the pipeline
that this node runs in.
executor_spec: Specification for the executor of the node.
custom_driver_spec: Specification for custom driver.
custom_executor_operators: Map of executable specs to executor
operators.
"""
launcher_callable = functools.partial(
_airflow_component_launcher,
pipeline_node=pipeline_node,
mlmd_connection=mlmd_connection,
pipeline_info=pipeline_info,
pipeline_runtime_spec=pipeline_runtime_spec,
executor_spec=executor_spec,
custom_driver_spec=custom_driver_spec,
custom_executor_operators=custom_executor_operators,
)
super().__init__(
task_id=pipeline_node.node_info.id,
provide_context=True,
python_callable=launcher_callable,
dag=parent_dag,
)
airflow_dag_runner
Definition of Airflow TFX runner. This is an unmodified copy from the TFX source code (outside of superficial, stylistic changes)
AirflowDagRunner
Tfx runner on Airflow.
Source code in zenml/integrations/airflow/orchestrators/airflow_dag_runner.py
class AirflowDagRunner:
"""Tfx runner on Airflow."""
def __init__(
self,
config: Optional[Union[Dict[str, Any], AirflowPipelineConfig]] = None,
):
"""Creates an instance of AirflowDagRunner.
Args:
config: Optional Airflow pipeline config for customizing the
launching of each component.
"""
self._config = config or pipeline_config.PipelineConfig()
if isinstance(config, dict):
warnings.warn(
"Pass config as a dict type is going to deprecated in 0.1.16. "
"Use AirflowPipelineConfig type instead.",
PendingDeprecationWarning,
)
self._config = AirflowPipelineConfig(airflow_dag_config=config)
@property
def config(self) -> pipeline_config.PipelineConfig:
return self._config
def run(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> "airflow.DAG":
"""Deploys given logical pipeline on Airflow.
Args:
pipeline: Logical pipeline containing pipeline args and comps.
stack: The current stack that ZenML is running on
runtime_configuration: The configuration of the run
Returns:
An Airflow DAG.
"""
# Only import these when needed.
import airflow # noqa
from zenml.integrations.airflow.orchestrators import airflow_component
# Merge airflow-specific configs with pipeline args
tfx_pipeline = create_tfx_pipeline(pipeline, stack=stack)
if runtime_configuration.schedule:
catchup = runtime_configuration.schedule.catchup
else:
catchup = False
airflow_dag = airflow.DAG(
dag_id=tfx_pipeline.pipeline_info.pipeline_name,
**(
typing.cast(
AirflowPipelineConfig, self._config
).airflow_dag_config
),
is_paused_upon_creation=False,
catchup=catchup,
)
pipeline_root = tfx_pipeline.pipeline_info.pipeline_root
if "tmp_dir" not in tfx_pipeline.additional_pipeline_args:
tmp_dir = os.path.join(pipeline_root, ".temp", "")
tfx_pipeline.additional_pipeline_args["tmp_dir"] = tmp_dir
for component in tfx_pipeline.components:
if isinstance(component, base_component.BaseComponent):
component._resolve_pip_dependencies(pipeline_root)
self._replace_runtime_params(component)
pb2_pipeline: Pb2Pipeline = compiler.Compiler().compile(tfx_pipeline)
# Substitute the runtime parameter to be a concrete run_id
runtime_parameter_utils.substitute_runtime_parameter(
pb2_pipeline,
{
"pipeline-run-id": runtime_configuration.run_name,
},
)
deployment_config = runner_utils.extract_local_deployment_config(
pb2_pipeline
)
connection_config = (
Repository().active_stack.metadata_store.get_tfx_metadata_config()
)
component_impl_map = {}
for node in pb2_pipeline.nodes:
pipeline_node: PipelineNode = node.pipeline_node # type: ignore[valid-type]
# Add the stack as context to each pipeline node:
context_utils.add_context_to_node(
pipeline_node,
type_=MetadataContextTypes.STACK.value,
name=str(hash(json.dumps(stack.dict(), sort_keys=True))),
properties=stack.dict(),
)
# Add all pydantic objects from runtime_configuration to the context
context_utils.add_runtime_configuration_to_node(
pipeline_node, runtime_configuration
)
# Add pipeline requirements as a context
requirements = " ".join(sorted(pipeline.requirements))
context_utils.add_context_to_node(
pipeline_node,
type_=MetadataContextTypes.PIPELINE_REQUIREMENTS.value,
name=str(hash(requirements)),
properties={"pipeline_requirements": requirements},
)
node_id = pipeline_node.node_info.id
executor_spec = runner_utils.extract_executor_spec(
deployment_config, node_id
)
custom_driver_spec = runner_utils.extract_custom_driver_spec(
deployment_config, node_id
)
step = get_step_for_node(
pipeline_node, steps=list(pipeline.steps.values())
)
custom_executor_operators = {
executable_spec_pb2.PythonClassExecutableSpec: step.executor_operator
}
current_airflow_component = airflow_component.AirflowComponent(
parent_dag=airflow_dag,
pipeline_node=pipeline_node,
mlmd_connection=connection_config,
pipeline_info=pb2_pipeline.pipeline_info,
pipeline_runtime_spec=pb2_pipeline.runtime_spec,
executor_spec=executor_spec,
custom_driver_spec=custom_driver_spec,
custom_executor_operators=custom_executor_operators,
)
component_impl_map[node_id] = current_airflow_component
for upstream_node in node.pipeline_node.upstream_nodes:
assert (
upstream_node in component_impl_map
), "Components is not in topological order"
current_airflow_component.set_upstream(
component_impl_map[upstream_node]
)
return airflow_dag
def _replace_runtime_params(
self, comp: base_node.BaseNode
) -> base_node.BaseNode:
"""Replaces runtime params for dynamic Airflow parameter execution.
Args:
comp: TFX component to be parsed.
Returns:
Returns edited component.
"""
for k, prop in comp.exec_properties.copy().items():
if isinstance(prop, RuntimeParameter):
# Airflow only supports string parameters.
if prop.ptype != str:
raise RuntimeError(
f"RuntimeParameter in Airflow does not support "
f"{prop.ptype}. The only ptype supported is string."
)
# If the default is a template, drop the template markers
# when inserting it into the .get() default argument below.
# Otherwise, provide the default as a quoted string.
default = cast(str, prop.default)
if default.startswith("{{") and default.endswith("}}"):
default = default[2:-2]
else:
default = json.dumps(default)
template_field = '{{ dag_run.conf.get("%s", %s) }}' % (
prop.name,
default,
)
comp.exec_properties[k] = template_field
return comp
__init__(self, config=None)
special
Creates an instance of AirflowDagRunner.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
config |
Union[Dict[str, Any], zenml.integrations.airflow.orchestrators.airflow_dag_runner.AirflowPipelineConfig] |
Optional Airflow pipeline config for customizing the launching of each component. |
None |
Source code in zenml/integrations/airflow/orchestrators/airflow_dag_runner.py
def __init__(
self,
config: Optional[Union[Dict[str, Any], AirflowPipelineConfig]] = None,
):
"""Creates an instance of AirflowDagRunner.
Args:
config: Optional Airflow pipeline config for customizing the
launching of each component.
"""
self._config = config or pipeline_config.PipelineConfig()
if isinstance(config, dict):
warnings.warn(
"Pass config as a dict type is going to deprecated in 0.1.16. "
"Use AirflowPipelineConfig type instead.",
PendingDeprecationWarning,
)
self._config = AirflowPipelineConfig(airflow_dag_config=config)
run(self, pipeline, stack, runtime_configuration)
Deploys given logical pipeline on Airflow.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline |
BasePipeline |
Logical pipeline containing pipeline args and comps. |
required |
stack |
Stack |
The current stack that ZenML is running on |
required |
runtime_configuration |
RuntimeConfiguration |
The configuration of the run |
required |
Returns:
| Type | Description |
|---|---|
airflow.DAG |
An Airflow DAG. |
Source code in zenml/integrations/airflow/orchestrators/airflow_dag_runner.py
def run(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> "airflow.DAG":
"""Deploys given logical pipeline on Airflow.
Args:
pipeline: Logical pipeline containing pipeline args and comps.
stack: The current stack that ZenML is running on
runtime_configuration: The configuration of the run
Returns:
An Airflow DAG.
"""
# Only import these when needed.
import airflow # noqa
from zenml.integrations.airflow.orchestrators import airflow_component
# Merge airflow-specific configs with pipeline args
tfx_pipeline = create_tfx_pipeline(pipeline, stack=stack)
if runtime_configuration.schedule:
catchup = runtime_configuration.schedule.catchup
else:
catchup = False
airflow_dag = airflow.DAG(
dag_id=tfx_pipeline.pipeline_info.pipeline_name,
**(
typing.cast(
AirflowPipelineConfig, self._config
).airflow_dag_config
),
is_paused_upon_creation=False,
catchup=catchup,
)
pipeline_root = tfx_pipeline.pipeline_info.pipeline_root
if "tmp_dir" not in tfx_pipeline.additional_pipeline_args:
tmp_dir = os.path.join(pipeline_root, ".temp", "")
tfx_pipeline.additional_pipeline_args["tmp_dir"] = tmp_dir
for component in tfx_pipeline.components:
if isinstance(component, base_component.BaseComponent):
component._resolve_pip_dependencies(pipeline_root)
self._replace_runtime_params(component)
pb2_pipeline: Pb2Pipeline = compiler.Compiler().compile(tfx_pipeline)
# Substitute the runtime parameter to be a concrete run_id
runtime_parameter_utils.substitute_runtime_parameter(
pb2_pipeline,
{
"pipeline-run-id": runtime_configuration.run_name,
},
)
deployment_config = runner_utils.extract_local_deployment_config(
pb2_pipeline
)
connection_config = (
Repository().active_stack.metadata_store.get_tfx_metadata_config()
)
component_impl_map = {}
for node in pb2_pipeline.nodes:
pipeline_node: PipelineNode = node.pipeline_node # type: ignore[valid-type]
# Add the stack as context to each pipeline node:
context_utils.add_context_to_node(
pipeline_node,
type_=MetadataContextTypes.STACK.value,
name=str(hash(json.dumps(stack.dict(), sort_keys=True))),
properties=stack.dict(),
)
# Add all pydantic objects from runtime_configuration to the context
context_utils.add_runtime_configuration_to_node(
pipeline_node, runtime_configuration
)
# Add pipeline requirements as a context
requirements = " ".join(sorted(pipeline.requirements))
context_utils.add_context_to_node(
pipeline_node,
type_=MetadataContextTypes.PIPELINE_REQUIREMENTS.value,
name=str(hash(requirements)),
properties={"pipeline_requirements": requirements},
)
node_id = pipeline_node.node_info.id
executor_spec = runner_utils.extract_executor_spec(
deployment_config, node_id
)
custom_driver_spec = runner_utils.extract_custom_driver_spec(
deployment_config, node_id
)
step = get_step_for_node(
pipeline_node, steps=list(pipeline.steps.values())
)
custom_executor_operators = {
executable_spec_pb2.PythonClassExecutableSpec: step.executor_operator
}
current_airflow_component = airflow_component.AirflowComponent(
parent_dag=airflow_dag,
pipeline_node=pipeline_node,
mlmd_connection=connection_config,
pipeline_info=pb2_pipeline.pipeline_info,
pipeline_runtime_spec=pb2_pipeline.runtime_spec,
executor_spec=executor_spec,
custom_driver_spec=custom_driver_spec,
custom_executor_operators=custom_executor_operators,
)
component_impl_map[node_id] = current_airflow_component
for upstream_node in node.pipeline_node.upstream_nodes:
assert (
upstream_node in component_impl_map
), "Components is not in topological order"
current_airflow_component.set_upstream(
component_impl_map[upstream_node]
)
return airflow_dag
AirflowPipelineConfig (PipelineConfig)
Pipeline config for AirflowDagRunner.
Source code in zenml/integrations/airflow/orchestrators/airflow_dag_runner.py
class AirflowPipelineConfig(pipeline_config.PipelineConfig):
"""Pipeline config for AirflowDagRunner."""
def __init__(
self,
airflow_dag_config: Optional[Dict[str, Any]] = None,
**kwargs: Any,
):
"""Creates an instance of AirflowPipelineConfig.
Args:
airflow_dag_config: Configs of Airflow DAG model. See
https://airflow.apache.org/_api/airflow/models/dag/index.html#airflow.models.dag.DAG
for the full spec.
**kwargs: keyword args for PipelineConfig.
"""
super().__init__(**kwargs)
self.airflow_dag_config = airflow_dag_config or {}
__init__(self, airflow_dag_config=None, **kwargs)
special
Creates an instance of AirflowPipelineConfig.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
airflow_dag_config |
Optional[Dict[str, Any]] |
Configs of Airflow DAG model. See https://airflow.apache.org/_api/airflow/models/dag/index.html#airflow.models.dag.DAG for the full spec. |
None |
**kwargs |
Any |
keyword args for PipelineConfig. |
{} |
Source code in zenml/integrations/airflow/orchestrators/airflow_dag_runner.py
def __init__(
self,
airflow_dag_config: Optional[Dict[str, Any]] = None,
**kwargs: Any,
):
"""Creates an instance of AirflowPipelineConfig.
Args:
airflow_dag_config: Configs of Airflow DAG model. See
https://airflow.apache.org/_api/airflow/models/dag/index.html#airflow.models.dag.DAG
for the full spec.
**kwargs: keyword args for PipelineConfig.
"""
super().__init__(**kwargs)
self.airflow_dag_config = airflow_dag_config or {}
airflow_orchestrator
AirflowOrchestrator (BaseOrchestrator)
pydantic-model
Orchestrator responsible for running pipelines using Airflow.
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
class AirflowOrchestrator(BaseOrchestrator):
"""Orchestrator responsible for running pipelines using Airflow."""
airflow_home: str = ""
# Class Configuration
FLAVOR: ClassVar[str] = "airflow"
def __init__(self, **values: Any):
"""Sets environment variables to configure airflow."""
super().__init__(**values)
self._set_env()
@root_validator
def set_airflow_home(cls, values: Dict[str, Any]) -> Dict[str, Any]:
"""Sets airflow home according to orchestrator UUID."""
if "uuid" not in values:
raise ValueError("`uuid` needs to exist for AirflowOrchestrator.")
values["airflow_home"] = os.path.join(
zenml.io.utils.get_global_config_directory(),
AIRFLOW_ROOT_DIR,
str(values["uuid"]),
)
return values
@property
def dags_directory(self) -> str:
"""Returns path to the airflow dags directory."""
return os.path.join(self.airflow_home, "dags")
@property
def pid_file(self) -> str:
"""Returns path to the daemon PID file."""
return os.path.join(self.airflow_home, "airflow_daemon.pid")
@property
def log_file(self) -> str:
"""Returns path to the airflow log file."""
return os.path.join(self.airflow_home, "airflow_orchestrator.log")
@property
def password_file(self) -> str:
"""Returns path to the webserver password file."""
return os.path.join(self.airflow_home, "standalone_admin_password.txt")
def _set_env(self) -> None:
"""Sets environment variables to configure airflow."""
os.environ["AIRFLOW_HOME"] = self.airflow_home
os.environ["AIRFLOW__CORE__DAGS_FOLDER"] = self.dags_directory
os.environ["AIRFLOW__CORE__DAG_DISCOVERY_SAFE_MODE"] = "false"
os.environ["AIRFLOW__CORE__LOAD_EXAMPLES"] = "false"
# check the DAG folder every 10 seconds for new files
os.environ["AIRFLOW__SCHEDULER__DAG_DIR_LIST_INTERVAL"] = "10"
def _copy_to_dag_directory_if_necessary(self, dag_filepath: str) -> None:
"""Copies the DAG module to the airflow DAGs directory if it's not
already located there.
Args:
dag_filepath: Path to the file in which the DAG is defined.
"""
dags_directory = zenml.io.utils.resolve_relative_path(
self.dags_directory
)
if dags_directory == os.path.dirname(dag_filepath):
logger.debug("File is already in airflow DAGs directory.")
else:
logger.debug(
"Copying dag file '%s' to DAGs directory.", dag_filepath
)
destination_path = os.path.join(
dags_directory, os.path.basename(dag_filepath)
)
if fileio.exists(destination_path):
logger.info(
"File '%s' already exists, overwriting with new DAG file",
destination_path,
)
fileio.copy(dag_filepath, destination_path, overwrite=True)
def _log_webserver_credentials(self) -> None:
"""Logs URL and credentials to log in to the airflow webserver.
Raises:
FileNotFoundError: If the password file does not exist.
"""
if fileio.exists(self.password_file):
with open(self.password_file) as file:
password = file.read().strip()
else:
raise FileNotFoundError(
f"Can't find password file '{self.password_file}'"
)
logger.info(
"To inspect your DAGs, login to http://0.0.0.0:8080 "
"with username: admin password: %s",
password,
)
def runtime_options(self) -> Dict[str, Any]:
"""Runtime options for the airflow orchestrator."""
return {DAG_FILEPATH_OPTION_KEY: None}
def prepare_pipeline_deployment(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> None:
"""Checks whether airflow is running and copies the DAG file to the
airflow DAGs directory.
Raises:
RuntimeError: If airflow is not running or no DAG filepath runtime
option is provided.
"""
if not self.is_running:
raise RuntimeError(
"Airflow orchestrator is currently not running. Run `zenml "
"stack up` to provision resources for the active stack."
)
try:
dag_filepath = runtime_configuration[DAG_FILEPATH_OPTION_KEY]
except KeyError:
raise RuntimeError(
f"No DAG filepath found in runtime configuration. Make sure "
f"to add the filepath to your airflow DAG file as a runtime "
f"option (key: '{DAG_FILEPATH_OPTION_KEY}')."
)
self._copy_to_dag_directory_if_necessary(dag_filepath=dag_filepath)
@property
def is_running(self) -> bool:
"""Returns whether the airflow daemon is currently running."""
from airflow.cli.commands.standalone_command import StandaloneCommand
from airflow.jobs.triggerer_job import TriggererJob
daemon_running = daemon.check_if_daemon_is_running(self.pid_file)
command = StandaloneCommand()
webserver_port_open = command.port_open(8080)
if not daemon_running:
if webserver_port_open:
raise RuntimeError(
"The airflow daemon does not seem to be running but "
"local port 8080 is occupied. Make sure the port is "
"available and try again."
)
# exit early so we don't check non-existing airflow databases
return False
# we can't use StandaloneCommand().is_ready() here as the
# Airflow SequentialExecutor apparently does not send a heartbeat
# while running a task which would result in this returning `False`
# even if Airflow is running.
airflow_running = webserver_port_open and command.job_running(
TriggererJob
)
return airflow_running
@property
def is_provisioned(self) -> bool:
"""Returns whether the airflow daemon is currently running."""
return self.is_running
def provision(self) -> None:
"""Ensures that Airflow is running."""
if self.is_running:
logger.info("Airflow is already running.")
self._log_webserver_credentials()
return
if not fileio.exists(self.dags_directory):
zenml.io.utils.create_dir_recursive_if_not_exists(
self.dags_directory
)
from airflow.cli.commands.standalone_command import StandaloneCommand
try:
command = StandaloneCommand()
# Run the daemon with a working directory inside the current
# zenml repo so the same repo will be used to run the DAGs
daemon.run_as_daemon(
command.run,
pid_file=self.pid_file,
log_file=self.log_file,
working_directory=str(Repository().root),
)
while not self.is_running:
# Wait until the daemon started all the relevant airflow
# processes
time.sleep(0.1)
self._log_webserver_credentials()
except Exception as e:
logger.error(e)
logger.error(
"An error occurred while starting the Airflow daemon. If you "
"want to start it manually, use the commands described in the "
"official Airflow quickstart guide for running Airflow locally."
)
self.deprovision()
def deprovision(self) -> None:
"""Stops the airflow daemon if necessary and tears down resources."""
if self.is_running:
daemon.stop_daemon(self.pid_file)
fileio.rmtree(self.airflow_home)
logger.info("Airflow spun down.")
def run_pipeline(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> Any:
"""Schedules a pipeline to be run on Airflow.
Returns:
An Airflow DAG object that corresponds to the ZenML pipeline.
"""
if runtime_configuration.schedule:
airflow_config = {
"schedule_interval": datetime.timedelta(
seconds=runtime_configuration.schedule.interval_second
),
"start_date": runtime_configuration.schedule.start_time,
"end_date": runtime_configuration.schedule.end_time,
}
else:
airflow_config = {
"schedule_interval": "@once",
# Scheduled in the past to make sure it runs immediately
"start_date": datetime.datetime.now() - datetime.timedelta(7),
}
runner = AirflowDagRunner(AirflowPipelineConfig(airflow_config))
return runner.run(
pipeline=pipeline,
stack=stack,
runtime_configuration=runtime_configuration,
)
dags_directory: str
property
readonly
Returns path to the airflow dags directory.
is_provisioned: bool
property
readonly
Returns whether the airflow daemon is currently running.
is_running: bool
property
readonly
Returns whether the airflow daemon is currently running.
log_file: str
property
readonly
Returns path to the airflow log file.
password_file: str
property
readonly
Returns path to the webserver password file.
pid_file: str
property
readonly
Returns path to the daemon PID file.
__init__(self, **values)
special
Sets environment variables to configure airflow.
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
def __init__(self, **values: Any):
"""Sets environment variables to configure airflow."""
super().__init__(**values)
self._set_env()
deprovision(self)
Stops the airflow daemon if necessary and tears down resources.
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
def deprovision(self) -> None:
"""Stops the airflow daemon if necessary and tears down resources."""
if self.is_running:
daemon.stop_daemon(self.pid_file)
fileio.rmtree(self.airflow_home)
logger.info("Airflow spun down.")
prepare_pipeline_deployment(self, pipeline, stack, runtime_configuration)
Checks whether airflow is running and copies the DAG file to the airflow DAGs directory.
Exceptions:
| Type | Description |
|---|---|
RuntimeError |
If airflow is not running or no DAG filepath runtime option is provided. |
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
def prepare_pipeline_deployment(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> None:
"""Checks whether airflow is running and copies the DAG file to the
airflow DAGs directory.
Raises:
RuntimeError: If airflow is not running or no DAG filepath runtime
option is provided.
"""
if not self.is_running:
raise RuntimeError(
"Airflow orchestrator is currently not running. Run `zenml "
"stack up` to provision resources for the active stack."
)
try:
dag_filepath = runtime_configuration[DAG_FILEPATH_OPTION_KEY]
except KeyError:
raise RuntimeError(
f"No DAG filepath found in runtime configuration. Make sure "
f"to add the filepath to your airflow DAG file as a runtime "
f"option (key: '{DAG_FILEPATH_OPTION_KEY}')."
)
self._copy_to_dag_directory_if_necessary(dag_filepath=dag_filepath)
provision(self)
Ensures that Airflow is running.
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
def provision(self) -> None:
"""Ensures that Airflow is running."""
if self.is_running:
logger.info("Airflow is already running.")
self._log_webserver_credentials()
return
if not fileio.exists(self.dags_directory):
zenml.io.utils.create_dir_recursive_if_not_exists(
self.dags_directory
)
from airflow.cli.commands.standalone_command import StandaloneCommand
try:
command = StandaloneCommand()
# Run the daemon with a working directory inside the current
# zenml repo so the same repo will be used to run the DAGs
daemon.run_as_daemon(
command.run,
pid_file=self.pid_file,
log_file=self.log_file,
working_directory=str(Repository().root),
)
while not self.is_running:
# Wait until the daemon started all the relevant airflow
# processes
time.sleep(0.1)
self._log_webserver_credentials()
except Exception as e:
logger.error(e)
logger.error(
"An error occurred while starting the Airflow daemon. If you "
"want to start it manually, use the commands described in the "
"official Airflow quickstart guide for running Airflow locally."
)
self.deprovision()
run_pipeline(self, pipeline, stack, runtime_configuration)
Schedules a pipeline to be run on Airflow.
Returns:
| Type | Description |
|---|---|
Any |
An Airflow DAG object that corresponds to the ZenML pipeline. |
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
def run_pipeline(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> Any:
"""Schedules a pipeline to be run on Airflow.
Returns:
An Airflow DAG object that corresponds to the ZenML pipeline.
"""
if runtime_configuration.schedule:
airflow_config = {
"schedule_interval": datetime.timedelta(
seconds=runtime_configuration.schedule.interval_second
),
"start_date": runtime_configuration.schedule.start_time,
"end_date": runtime_configuration.schedule.end_time,
}
else:
airflow_config = {
"schedule_interval": "@once",
# Scheduled in the past to make sure it runs immediately
"start_date": datetime.datetime.now() - datetime.timedelta(7),
}
runner = AirflowDagRunner(AirflowPipelineConfig(airflow_config))
return runner.run(
pipeline=pipeline,
stack=stack,
runtime_configuration=runtime_configuration,
)
runtime_options(self)
Runtime options for the airflow orchestrator.
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
def runtime_options(self) -> Dict[str, Any]:
"""Runtime options for the airflow orchestrator."""
return {DAG_FILEPATH_OPTION_KEY: None}
set_airflow_home(values)
classmethod
Sets airflow home according to orchestrator UUID.
Source code in zenml/integrations/airflow/orchestrators/airflow_orchestrator.py
@root_validator
def set_airflow_home(cls, values: Dict[str, Any]) -> Dict[str, Any]:
"""Sets airflow home according to orchestrator UUID."""
if "uuid" not in values:
raise ValueError("`uuid` needs to exist for AirflowOrchestrator.")
values["airflow_home"] = os.path.join(
zenml.io.utils.get_global_config_directory(),
AIRFLOW_ROOT_DIR,
str(values["uuid"]),
)
return values
aws
special
The AWS integration provides a way for our users to manage their secrets through AWS.
AWSIntegration (Integration)
Definition of AWS integration for ZenML.
Source code in zenml/integrations/aws/__init__.py
class AWSIntegration(Integration):
"""Definition of AWS integration for ZenML."""
NAME = AWS
REQUIREMENTS = ["boto3==1.21.21"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.aws import secret_schemas # noqa
from zenml.integrations.aws import secrets_managers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/aws/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.aws import secret_schemas # noqa
from zenml.integrations.aws import secrets_managers # noqa
secret_schemas
special
Secret Schema
...
secrets_managers
special
Secret Manager
...
aws_secrets_manager
AWSSecretsManager (BaseSecretsManager)
pydantic-model
Class to interact with the AWS secrets manager.
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
class AWSSecretsManager(BaseSecretsManager):
"""Class to interact with the AWS secrets manager."""
region_name: str = DEFAULT_AWS_REGION
# Class configuration
FLAVOR: ClassVar[str] = "aws"
CLIENT: ClassVar[Any] = None
@classmethod
def _ensure_client_connected(cls, region_name: str) -> None:
if cls.CLIENT is None:
# Create a Secrets Manager client
session = boto3.session.Session()
cls.CLIENT = session.client(
service_name="secretsmanager", region_name=region_name
)
def register_secret(self, secret: BaseSecretSchema) -> None:
"""Registers a new secret.
Args:
secret: the secret to register"""
self._ensure_client_connected(self.region_name)
secret_value = jsonify_secret_contents(secret)
kwargs = {"Name": secret.name, "SecretString": secret_value}
self.CLIENT.create_secret(**kwargs)
def get_secret(self, secret_name: str) -> BaseSecretSchema:
"""Gets a secret.
Args:
secret_name: the name of the secret to get
Returns:
The secret.
Raises:
RuntimeError: if the secret does not exist"""
self._ensure_client_connected(self.region_name)
get_secret_value_response = self.CLIENT.get_secret_value(
SecretId=secret_name
)
if "SecretString" not in get_secret_value_response:
raise RuntimeError(f"No secrets found within the {secret_name}")
secret_contents: Dict[str, str] = json.loads(
get_secret_value_response["SecretString"]
)
zenml_schema_name = secret_contents.pop(ZENML_SCHEMA_NAME)
secret_contents["name"] = secret_name
secret_schema = SecretSchemaClassRegistry.get_class(
secret_schema=zenml_schema_name
)
return secret_schema(**secret_contents)
def get_all_secret_keys(self) -> List[str]:
"""Get all secret keys.
Returns:
A list of all secret keys."""
self._ensure_client_connected(self.region_name)
# TODO [ENG-720]: Deal with pagination in the aws secret manager when
# listing all secrets
# TODO [ENG-721]: take out this magic maxresults number
response = self.CLIENT.list_secrets(MaxResults=100)
return [secret["Name"] for secret in response["SecretList"]]
def update_secret(self, secret: BaseSecretSchema) -> None:
"""Update an existing secret.
Args:
secret: the secret to update"""
self._ensure_client_connected(self.region_name)
secret_value = jsonify_secret_contents(secret)
kwargs = {"SecretId": secret.name, "SecretString": secret_value}
self.CLIENT.put_secret_value(**kwargs)
def delete_secret(self, secret_name: str) -> None:
"""Delete an existing secret.
Args:
secret_name: the name of the secret to delete"""
self._ensure_client_connected(self.region_name)
self.CLIENT.delete_secret(
SecretId=secret_name, ForceDeleteWithoutRecovery=False
)
def delete_all_secrets(self, force: bool = False) -> None:
"""Delete all existing secrets.
Args:
force: whether to force delete all secrets"""
self._ensure_client_connected(self.region_name)
for secret_name in self.get_all_secret_keys():
self.CLIENT.delete_secret(
SecretId=secret_name, ForceDeleteWithoutRecovery=force
)
delete_all_secrets(self, force=False)
Delete all existing secrets.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
force |
bool |
whether to force delete all secrets |
False |
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
def delete_all_secrets(self, force: bool = False) -> None:
"""Delete all existing secrets.
Args:
force: whether to force delete all secrets"""
self._ensure_client_connected(self.region_name)
for secret_name in self.get_all_secret_keys():
self.CLIENT.delete_secret(
SecretId=secret_name, ForceDeleteWithoutRecovery=force
)
delete_secret(self, secret_name)
Delete an existing secret.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
secret_name |
str |
the name of the secret to delete |
required |
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
def delete_secret(self, secret_name: str) -> None:
"""Delete an existing secret.
Args:
secret_name: the name of the secret to delete"""
self._ensure_client_connected(self.region_name)
self.CLIENT.delete_secret(
SecretId=secret_name, ForceDeleteWithoutRecovery=False
)
get_all_secret_keys(self)
Get all secret keys.
Returns:
| Type | Description |
|---|---|
List[str] |
A list of all secret keys. |
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
def get_all_secret_keys(self) -> List[str]:
"""Get all secret keys.
Returns:
A list of all secret keys."""
self._ensure_client_connected(self.region_name)
# TODO [ENG-720]: Deal with pagination in the aws secret manager when
# listing all secrets
# TODO [ENG-721]: take out this magic maxresults number
response = self.CLIENT.list_secrets(MaxResults=100)
return [secret["Name"] for secret in response["SecretList"]]
get_secret(self, secret_name)
Gets a secret.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
secret_name |
str |
the name of the secret to get |
required |
Returns:
| Type | Description |
|---|---|
BaseSecretSchema |
The secret. |
Exceptions:
| Type | Description |
|---|---|
RuntimeError |
if the secret does not exist |
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
def get_secret(self, secret_name: str) -> BaseSecretSchema:
"""Gets a secret.
Args:
secret_name: the name of the secret to get
Returns:
The secret.
Raises:
RuntimeError: if the secret does not exist"""
self._ensure_client_connected(self.region_name)
get_secret_value_response = self.CLIENT.get_secret_value(
SecretId=secret_name
)
if "SecretString" not in get_secret_value_response:
raise RuntimeError(f"No secrets found within the {secret_name}")
secret_contents: Dict[str, str] = json.loads(
get_secret_value_response["SecretString"]
)
zenml_schema_name = secret_contents.pop(ZENML_SCHEMA_NAME)
secret_contents["name"] = secret_name
secret_schema = SecretSchemaClassRegistry.get_class(
secret_schema=zenml_schema_name
)
return secret_schema(**secret_contents)
register_secret(self, secret)
Registers a new secret.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
secret |
BaseSecretSchema |
the secret to register |
required |
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
def register_secret(self, secret: BaseSecretSchema) -> None:
"""Registers a new secret.
Args:
secret: the secret to register"""
self._ensure_client_connected(self.region_name)
secret_value = jsonify_secret_contents(secret)
kwargs = {"Name": secret.name, "SecretString": secret_value}
self.CLIENT.create_secret(**kwargs)
update_secret(self, secret)
Update an existing secret.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
secret |
BaseSecretSchema |
the secret to update |
required |
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
def update_secret(self, secret: BaseSecretSchema) -> None:
"""Update an existing secret.
Args:
secret: the secret to update"""
self._ensure_client_connected(self.region_name)
secret_value = jsonify_secret_contents(secret)
kwargs = {"SecretId": secret.name, "SecretString": secret_value}
self.CLIENT.put_secret_value(**kwargs)
jsonify_secret_contents(secret)
Adds the secret type to the secret contents to persist the schema type in the secrets backend, so that the correct SecretSchema can be retrieved when the secret is queried from the backend.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
secret |
BaseSecretSchema |
should be a subclass of the BaseSecretSchema class |
required |
Returns:
| Type | Description |
|---|---|
str |
jsonified dictionary containing all key-value pairs and the ZenML schema type |
Source code in zenml/integrations/aws/secrets_managers/aws_secrets_manager.py
def jsonify_secret_contents(secret: BaseSecretSchema) -> str:
"""Adds the secret type to the secret contents to persist the schema
type in the secrets backend, so that the correct SecretSchema can be
retrieved when the secret is queried from the backend.
Args:
secret: should be a subclass of the BaseSecretSchema class
Returns:
jsonified dictionary containing all key-value pairs and the ZenML schema
type
"""
secret_contents = secret.content
secret_contents[ZENML_SCHEMA_NAME] = secret.TYPE
return json.dumps(secret_contents)
azure
special
The Azure integration submodule provides a way to run ZenML pipelines in a cloud
environment. Specifically, it allows the use of cloud artifact stores,
and an io module to handle file operations on Azure Blob Storage.
AzureIntegration (Integration)
Definition of Azure integration for ZenML.
Source code in zenml/integrations/azure/__init__.py
class AzureIntegration(Integration):
"""Definition of Azure integration for ZenML."""
NAME = AZURE
REQUIREMENTS = ["adlfs==2021.10.0"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.azure import artifact_stores # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/azure/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.azure import artifact_stores # noqa
artifact_stores
special
azure_artifact_store
AzureArtifactStore (BaseArtifactStore)
pydantic-model
Artifact Store for Microsoft Azure based artifacts.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
class AzureArtifactStore(BaseArtifactStore):
"""Artifact Store for Microsoft Azure based artifacts."""
_filesystem: Optional[adlfs.AzureBlobFileSystem] = None
# Class Configuration
FLAVOR: ClassVar[str] = "azure"
SUPPORTED_SCHEMES: ClassVar[Set[str]] = {"abfs://", "az://"}
@property
def filesystem(self) -> adlfs.AzureBlobFileSystem:
"""The adlfs filesystem to access this artifact store."""
if not self._filesystem:
self._filesystem = adlfs.AzureBlobFileSystem(
anon=False,
use_listings_cache=False,
)
return self._filesystem
@classmethod
def _split_path(cls, path: PathType) -> Tuple[str, str]:
"""Splits a path into the filesystem prefix and remainder.
Example:
```python
prefix, remainder = ZenAzure._split_path("az://my_container/test.txt")
print(prefix, remainder) # "az://" "my_container/test.txt"
```
"""
path = convert_to_str(path)
prefix = ""
for potential_prefix in cls.SUPPORTED_SCHEMES:
if path.startswith(potential_prefix):
prefix = potential_prefix
path = path[len(potential_prefix) :]
break
return prefix, path
def open(self, path: PathType, mode: str = "r") -> Any:
"""Open a file at the given path.
Args:
path: Path of the file to open.
mode: Mode in which to open the file. Currently, only
'rb' and 'wb' to read and write binary files are supported.
"""
return self.filesystem.open(path=path, mode=mode)
def copyfile(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Copy a file.
Args:
src: The path to copy from.
dst: The path to copy to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to copy to destination '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to copy anyway."
)
# TODO [ENG-151]: Check if it works with overwrite=True or if we need to
# manually remove it first
self.filesystem.copy(path1=src, path2=dst)
def exists(self, path: PathType) -> bool:
"""Check whether a path exists."""
return self.filesystem.exists(path=path) # type: ignore[no-any-return]
def glob(self, pattern: PathType) -> List[PathType]:
"""Return all paths that match the given glob pattern.
The glob pattern may include:
- '*' to match any number of characters
- '?' to match a single character
- '[...]' to match one of the characters inside the brackets
- '**' as the full name of a path component to match to search
in subdirectories of any depth (e.g. '/some_dir/**/some_file)
Args:
pattern: The glob pattern to match, see details above.
Returns:
A list of paths that match the given glob pattern.
"""
prefix, _ = self._split_path(pattern)
return [
f"{prefix}{path}" for path in self.filesystem.glob(path=pattern)
]
def isdir(self, path: PathType) -> bool:
"""Check whether a path is a directory."""
return self.filesystem.isdir(path=path) # type: ignore[no-any-return]
def listdir(self, path: PathType) -> List[PathType]:
"""Return a list of files in a directory."""
_, path = self._split_path(path)
def _extract_basename(file_dict: Dict[str, Any]) -> str:
"""Extracts the basename from a file info dict returned by the Azure
filesystem."""
file_path = cast(str, file_dict["name"])
base_name = file_path[len(path) :]
return base_name.lstrip("/")
return [
_extract_basename(dict_)
for dict_ in self.filesystem.listdir(path=path)
]
def makedirs(self, path: PathType) -> None:
"""Create a directory at the given path. If needed also
create missing parent directories."""
self.filesystem.makedirs(path=path, exist_ok=True)
def mkdir(self, path: PathType) -> None:
"""Create a directory at the given path."""
self.filesystem.makedir(path=path)
def remove(self, path: PathType) -> None:
"""Remove the file at the given path."""
self.filesystem.rm_file(path=path)
def rename(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Rename source file to destination file.
Args:
src: The path of the file to rename.
dst: The path to rename the source file to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to rename file to '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to rename anyway."
)
# TODO [ENG-152]: Check if it works with overwrite=True or if we need
# to manually remove it first
self.filesystem.rename(path1=src, path2=dst)
def rmtree(self, path: PathType) -> None:
"""Remove the given directory."""
self.filesystem.delete(path=path, recursive=True)
def stat(self, path: PathType) -> Dict[str, Any]:
"""Return stat info for the given path."""
return self.filesystem.stat(path=path) # type: ignore[no-any-return]
def walk(
self,
top: PathType,
topdown: bool = True,
onerror: Optional[Callable[..., None]] = None,
) -> Iterable[Tuple[PathType, List[PathType], List[PathType]]]:
"""Return an iterator that walks the contents of the given directory.
Args:
top: Path of directory to walk.
topdown: Unused argument to conform to interface.
onerror: Unused argument to conform to interface.
Returns:
An Iterable of Tuples, each of which contain the path of the current
directory path, a list of directories inside the current directory
and a list of files inside the current directory.
"""
# TODO [ENG-153]: Additional params
prefix, _ = self._split_path(top)
for (
directory,
subdirectories,
files,
) in self.filesystem.walk(path=top):
yield f"{prefix}{directory}", subdirectories, files
filesystem: AzureBlobFileSystem
property
readonly
The adlfs filesystem to access this artifact store.
copyfile(self, src, dst, overwrite=False)
Copy a file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
src |
Union[bytes, str] |
The path to copy from. |
required |
dst |
Union[bytes, str] |
The path to copy to. |
required |
overwrite |
bool |
If a file already exists at the destination, this
method will overwrite it if overwrite= |
False |
Exceptions:
| Type | Description |
|---|---|
FileNotFoundError |
If the source file does not exist. |
FileExistsError |
If a file already exists at the destination
and overwrite is not set to |
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def copyfile(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Copy a file.
Args:
src: The path to copy from.
dst: The path to copy to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to copy to destination '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to copy anyway."
)
# TODO [ENG-151]: Check if it works with overwrite=True or if we need to
# manually remove it first
self.filesystem.copy(path1=src, path2=dst)
exists(self, path)
Check whether a path exists.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def exists(self, path: PathType) -> bool:
"""Check whether a path exists."""
return self.filesystem.exists(path=path) # type: ignore[no-any-return]
glob(self, pattern)
Return all paths that match the given glob pattern. The glob pattern may include: - '' to match any number of characters - '?' to match a single character - '[...]' to match one of the characters inside the brackets - '' as the full name of a path component to match to search in subdirectories of any depth (e.g. '/some_dir/*/some_file)
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pattern |
Union[bytes, str] |
The glob pattern to match, see details above. |
required |
Returns:
| Type | Description |
|---|---|
List[Union[bytes, str]] |
A list of paths that match the given glob pattern. |
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def glob(self, pattern: PathType) -> List[PathType]:
"""Return all paths that match the given glob pattern.
The glob pattern may include:
- '*' to match any number of characters
- '?' to match a single character
- '[...]' to match one of the characters inside the brackets
- '**' as the full name of a path component to match to search
in subdirectories of any depth (e.g. '/some_dir/**/some_file)
Args:
pattern: The glob pattern to match, see details above.
Returns:
A list of paths that match the given glob pattern.
"""
prefix, _ = self._split_path(pattern)
return [
f"{prefix}{path}" for path in self.filesystem.glob(path=pattern)
]
isdir(self, path)
Check whether a path is a directory.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def isdir(self, path: PathType) -> bool:
"""Check whether a path is a directory."""
return self.filesystem.isdir(path=path) # type: ignore[no-any-return]
listdir(self, path)
Return a list of files in a directory.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def listdir(self, path: PathType) -> List[PathType]:
"""Return a list of files in a directory."""
_, path = self._split_path(path)
def _extract_basename(file_dict: Dict[str, Any]) -> str:
"""Extracts the basename from a file info dict returned by the Azure
filesystem."""
file_path = cast(str, file_dict["name"])
base_name = file_path[len(path) :]
return base_name.lstrip("/")
return [
_extract_basename(dict_)
for dict_ in self.filesystem.listdir(path=path)
]
makedirs(self, path)
Create a directory at the given path. If needed also create missing parent directories.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def makedirs(self, path: PathType) -> None:
"""Create a directory at the given path. If needed also
create missing parent directories."""
self.filesystem.makedirs(path=path, exist_ok=True)
mkdir(self, path)
Create a directory at the given path.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def mkdir(self, path: PathType) -> None:
"""Create a directory at the given path."""
self.filesystem.makedir(path=path)
open(self, path, mode='r')
Open a file at the given path.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
path |
Union[bytes, str] |
Path of the file to open. |
required |
mode |
str |
Mode in which to open the file. Currently, only 'rb' and 'wb' to read and write binary files are supported. |
'r' |
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def open(self, path: PathType, mode: str = "r") -> Any:
"""Open a file at the given path.
Args:
path: Path of the file to open.
mode: Mode in which to open the file. Currently, only
'rb' and 'wb' to read and write binary files are supported.
"""
return self.filesystem.open(path=path, mode=mode)
remove(self, path)
Remove the file at the given path.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def remove(self, path: PathType) -> None:
"""Remove the file at the given path."""
self.filesystem.rm_file(path=path)
rename(self, src, dst, overwrite=False)
Rename source file to destination file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
src |
Union[bytes, str] |
The path of the file to rename. |
required |
dst |
Union[bytes, str] |
The path to rename the source file to. |
required |
overwrite |
bool |
If a file already exists at the destination, this
method will overwrite it if overwrite= |
False |
Exceptions:
| Type | Description |
|---|---|
FileNotFoundError |
If the source file does not exist. |
FileExistsError |
If a file already exists at the destination
and overwrite is not set to |
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def rename(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Rename source file to destination file.
Args:
src: The path of the file to rename.
dst: The path to rename the source file to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to rename file to '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to rename anyway."
)
# TODO [ENG-152]: Check if it works with overwrite=True or if we need
# to manually remove it first
self.filesystem.rename(path1=src, path2=dst)
rmtree(self, path)
Remove the given directory.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def rmtree(self, path: PathType) -> None:
"""Remove the given directory."""
self.filesystem.delete(path=path, recursive=True)
stat(self, path)
Return stat info for the given path.
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def stat(self, path: PathType) -> Dict[str, Any]:
"""Return stat info for the given path."""
return self.filesystem.stat(path=path) # type: ignore[no-any-return]
walk(self, top, topdown=True, onerror=None)
Return an iterator that walks the contents of the given directory.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
top |
Union[bytes, str] |
Path of directory to walk. |
required |
topdown |
bool |
Unused argument to conform to interface. |
True |
onerror |
Optional[Callable[..., NoneType]] |
Unused argument to conform to interface. |
None |
Returns:
| Type | Description |
|---|---|
Iterable[Tuple[Union[bytes, str], List[Union[bytes, str]], List[Union[bytes, str]]]] |
An Iterable of Tuples, each of which contain the path of the current directory path, a list of directories inside the current directory and a list of files inside the current directory. |
Source code in zenml/integrations/azure/artifact_stores/azure_artifact_store.py
def walk(
self,
top: PathType,
topdown: bool = True,
onerror: Optional[Callable[..., None]] = None,
) -> Iterable[Tuple[PathType, List[PathType], List[PathType]]]:
"""Return an iterator that walks the contents of the given directory.
Args:
top: Path of directory to walk.
topdown: Unused argument to conform to interface.
onerror: Unused argument to conform to interface.
Returns:
An Iterable of Tuples, each of which contain the path of the current
directory path, a list of directories inside the current directory
and a list of files inside the current directory.
"""
# TODO [ENG-153]: Additional params
prefix, _ = self._split_path(top)
for (
directory,
subdirectories,
files,
) in self.filesystem.walk(path=top):
yield f"{prefix}{directory}", subdirectories, files
azureml
special
The AzureML integration submodule provides a way to run ZenML steps in AzureML.
AzureMLIntegration (Integration)
Definition of AzureML integration for ZenML.
Source code in zenml/integrations/azureml/__init__.py
class AzureMLIntegration(Integration):
"""Definition of AzureML integration for ZenML."""
NAME = AZUREML
REQUIREMENTS = ["azureml-core==1.39.0.post1"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.azureml import step_operators # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/azureml/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.azureml import step_operators # noqa
step_operators
special
azureml_step_operator
AzureMLStepOperator (BaseStepOperator)
pydantic-model
Step operator to run a step on AzureML.
This class defines code that can set up an AzureML environment and run the ZenML entrypoint command in it.
Attributes:
| Name | Type | Description |
|---|---|---|
subscription_id |
str |
The Azure account's subscription ID |
resource_group |
str |
The resource group to which the AzureML workspace is deployed. |
workspace_name |
str |
The name of the AzureML Workspace. |
compute_target_name |
str |
The name of the configured ComputeTarget. An instance of it has to be created on the portal if it doesn't exist already. |
environment_name |
Optional[str] |
[Optional] The name of the environment if there already exists one. |
docker_base_image |
Optional[str] |
[Optional] The custom docker base image that the environment should use. |
tenant_id |
Optional[str] |
The Azure Tenant ID. |
service_principal_id |
Optional[str] |
The ID for the service principal that is created to allow apps to access secure resources. |
service_principal_password |
Optional[str] |
Password for the service principal. |
Source code in zenml/integrations/azureml/step_operators/azureml_step_operator.py
class AzureMLStepOperator(BaseStepOperator):
"""Step operator to run a step on AzureML.
This class defines code that can set up an AzureML environment and run the
ZenML entrypoint command in it.
Attributes:
subscription_id: The Azure account's subscription ID
resource_group: The resource group to which the AzureML workspace
is deployed.
workspace_name: The name of the AzureML Workspace.
compute_target_name: The name of the configured ComputeTarget.
An instance of it has to be created on the portal if it doesn't
exist already.
environment_name: [Optional] The name of the environment if there
already exists one.
docker_base_image: [Optional] The custom docker base image that the
environment should use.
tenant_id: The Azure Tenant ID.
service_principal_id: The ID for the service principal that is created
to allow apps to access secure resources.
service_principal_password: Password for the service principal.
"""
subscription_id: str
resource_group: str
workspace_name: str
compute_target_name: str
# Environment
environment_name: Optional[str] = None
docker_base_image: Optional[str] = None
# Service principal authentication
# https://docs.microsoft.com/en-us/azure/machine-learning/how-to-setup-authentication#configure-a-service-principal
tenant_id: Optional[str] = None
service_principal_id: Optional[str] = None
service_principal_password: Optional[str] = None
# Class Configuration
FLAVOR: ClassVar[str] = "azureml"
def _get_authentication(self) -> Optional[AbstractAuthentication]:
if (
self.tenant_id
and self.service_principal_id
and self.service_principal_password
):
return ServicePrincipalAuthentication(
tenant_id=self.tenant_id,
service_principal_id=self.service_principal_id,
service_principal_password=self.service_principal_password,
)
return None
def _prepare_environment(
self, workspace: Workspace, requirements: List[str], run_name: str
) -> Environment:
"""Prepares the environment in which Azure will run all jobs.
Args:
workspace: The AzureML Workspace that has configuration
for a storage account, container registry among other
things.
requirements: The list of requirements to be installed
in the environment.
run_name: The name of the pipeline run that can be used
for naming environments and runs.
"""
if self.environment_name:
environment = Environment.get(
workspace=workspace, name=self.environment_name
)
if not environment.python.conda_dependencies:
environment.python.conda_dependencies = (
CondaDependencies.create(
python_version=ZenMLEnvironment.python_version()
)
)
for requirement in requirements:
environment.python.conda_dependencies.add_pip_package(
requirement
)
else:
environment = Environment(name=f"zenml-{run_name}")
environment.python.conda_dependencies = CondaDependencies.create(
pip_packages=requirements,
python_version=ZenMLEnvironment.python_version(),
)
if self.docker_base_image:
# replace the default azure base image
environment.docker.base_image = self.docker_base_image
environment_variables = {
"ENV_ZENML_PREVENT_PIPELINE_EXECUTION": "True",
}
# set credentials to access azure storage
for key in [
"AZURE_STORAGE_ACCOUNT_KEY",
"AZURE_STORAGE_ACCOUNT_NAME",
"AZURE_STORAGE_CONNECTION_STRING",
"AZURE_STORAGE_SAS_TOKEN",
]:
value = os.getenv(key)
if value:
environment_variables[key] = value
environment_variables[
ENV_ZENML_CONFIG_PATH
] = f"./{CONTAINER_ZENML_CONFIG_DIR}"
environment.environment_variables = environment_variables
return environment
def launch(
self,
pipeline_name: str,
run_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> None:
"""Launches a step on AzureML.
Args:
pipeline_name: Name of the pipeline which the step to be executed
is part of.
run_name: Name of the pipeline run which the step to be executed
is part of.
entrypoint_command: Command that executes the step.
requirements: List of pip requirements that must be installed
inside the step operator environment.
"""
workspace = Workspace.get(
subscription_id=self.subscription_id,
resource_group=self.resource_group,
name=self.workspace_name,
auth=self._get_authentication(),
)
source_directory = get_source_root_path()
config_path = os.path.join(source_directory, CONTAINER_ZENML_CONFIG_DIR)
try:
# Save a copy of the current global configuration with the
# active profile contents into the build context, to have
# the configured stacks accessible from within the Azure ML
# environment.
GlobalConfiguration().copy_config_with_active_profile(
config_path,
load_config_path=f"./{CONTAINER_ZENML_CONFIG_DIR}",
)
environment = self._prepare_environment(
workspace=workspace,
requirements=requirements,
run_name=run_name,
)
compute_target = ComputeTarget(
workspace=workspace, name=self.compute_target_name
)
run_config = ScriptRunConfig(
source_directory=source_directory,
environment=environment,
compute_target=compute_target,
command=entrypoint_command,
)
experiment = Experiment(workspace=workspace, name=pipeline_name)
run = experiment.submit(config=run_config)
finally:
# Clean up the temporary build files
fileio.rmtree(config_path)
run.display_name = run_name
run.wait_for_completion(show_output=True)
launch(self, pipeline_name, run_name, requirements, entrypoint_command)
Launches a step on AzureML.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline_name |
str |
Name of the pipeline which the step to be executed is part of. |
required |
run_name |
str |
Name of the pipeline run which the step to be executed is part of. |
required |
entrypoint_command |
List[str] |
Command that executes the step. |
required |
requirements |
List[str] |
List of pip requirements that must be installed inside the step operator environment. |
required |
Source code in zenml/integrations/azureml/step_operators/azureml_step_operator.py
def launch(
self,
pipeline_name: str,
run_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> None:
"""Launches a step on AzureML.
Args:
pipeline_name: Name of the pipeline which the step to be executed
is part of.
run_name: Name of the pipeline run which the step to be executed
is part of.
entrypoint_command: Command that executes the step.
requirements: List of pip requirements that must be installed
inside the step operator environment.
"""
workspace = Workspace.get(
subscription_id=self.subscription_id,
resource_group=self.resource_group,
name=self.workspace_name,
auth=self._get_authentication(),
)
source_directory = get_source_root_path()
config_path = os.path.join(source_directory, CONTAINER_ZENML_CONFIG_DIR)
try:
# Save a copy of the current global configuration with the
# active profile contents into the build context, to have
# the configured stacks accessible from within the Azure ML
# environment.
GlobalConfiguration().copy_config_with_active_profile(
config_path,
load_config_path=f"./{CONTAINER_ZENML_CONFIG_DIR}",
)
environment = self._prepare_environment(
workspace=workspace,
requirements=requirements,
run_name=run_name,
)
compute_target = ComputeTarget(
workspace=workspace, name=self.compute_target_name
)
run_config = ScriptRunConfig(
source_directory=source_directory,
environment=environment,
compute_target=compute_target,
command=entrypoint_command,
)
experiment = Experiment(workspace=workspace, name=pipeline_name)
run = experiment.submit(config=run_config)
finally:
# Clean up the temporary build files
fileio.rmtree(config_path)
run.display_name = run_name
run.wait_for_completion(show_output=True)
dash
special
DashIntegration (Integration)
Definition of Dash integration for ZenML.
Source code in zenml/integrations/dash/__init__.py
class DashIntegration(Integration):
"""Definition of Dash integration for ZenML."""
NAME = DASH
REQUIREMENTS = [
"dash>=2.0.0",
"dash-cytoscape>=0.3.0",
"dash-bootstrap-components>=1.0.1",
]
visualizers
special
pipeline_run_lineage_visualizer
PipelineRunLineageVisualizer (BasePipelineRunVisualizer)
Implementation of a lineage diagram via the dash and dash-cyctoscape library.
Source code in zenml/integrations/dash/visualizers/pipeline_run_lineage_visualizer.py
class PipelineRunLineageVisualizer(BasePipelineRunVisualizer):
"""Implementation of a lineage diagram via the [dash](
https://plotly.com/dash/) and [dash-cyctoscape](
https://dash.plotly.com/cytoscape) library."""
ARTIFACT_PREFIX = "artifact_"
STEP_PREFIX = "step_"
STATUS_CLASS_MAPPING = {
ExecutionStatus.CACHED: "green",
ExecutionStatus.FAILED: "red",
ExecutionStatus.RUNNING: "yellow",
ExecutionStatus.COMPLETED: "blue",
}
def visualize(
self, object: PipelineRunView, *args: Any, **kwargs: Any
) -> dash.Dash:
"""Method to visualize pipeline runs via the Dash library. The layout
puts every layer of the dag in a column.
"""
app = dash.Dash(
__name__,
external_stylesheets=[
dbc.themes.BOOTSTRAP,
dbc.icons.BOOTSTRAP,
],
)
nodes, edges, first_step_id = [], [], None
first_step_id = None
for step in object.steps:
step_output_artifacts = list(step.outputs.values())
execution_id = (
step_output_artifacts[0].producer_step.id
if step_output_artifacts
else step.id
)
step_id = self.STEP_PREFIX + str(step.id)
if first_step_id is None:
first_step_id = step_id
nodes.append(
{
"data": {
"id": step_id,
"execution_id": execution_id,
"label": f"{execution_id} / {step.entrypoint_name}",
"entrypoint_name": step.entrypoint_name, # redundant for consistency
"name": step.name, # redundant for consistency
"type": "step",
"parameters": step.parameters,
"inputs": {k: v.uri for k, v in step.inputs.items()},
"outputs": {k: v.uri for k, v in step.outputs.items()},
},
"classes": self.STATUS_CLASS_MAPPING[step.status],
}
)
for artifact_name, artifact in step.outputs.items():
nodes.append(
{
"data": {
"id": self.ARTIFACT_PREFIX + str(artifact.id),
"execution_id": artifact.id,
"label": f"{artifact.id} / {artifact_name} ("
f"{artifact.data_type})",
"type": "artifact",
"name": artifact_name,
"is_cached": artifact.is_cached,
"artifact_type": artifact.type,
"artifact_data_type": artifact.data_type,
"parent_step_id": artifact.parent_step_id,
"producer_step_id": artifact.producer_step.id,
"uri": artifact.uri,
},
"classes": f"rectangle "
f"{self.STATUS_CLASS_MAPPING[step.status]}",
}
)
edges.append(
{
"data": {
"source": self.STEP_PREFIX + str(step.id),
"target": self.ARTIFACT_PREFIX + str(artifact.id),
},
"classes": f"edge-arrow "
f"{self.STATUS_CLASS_MAPPING[step.status]}"
+ (" dashed" if artifact.is_cached else " solid"),
}
)
for artifact_name, artifact in step.inputs.items():
edges.append(
{
"data": {
"source": self.ARTIFACT_PREFIX + str(artifact.id),
"target": self.STEP_PREFIX + str(step.id),
},
"classes": "edge-arrow "
+ (
f"{self.STATUS_CLASS_MAPPING[ExecutionStatus.CACHED]} dashed"
if artifact.is_cached
else f"{self.STATUS_CLASS_MAPPING[step.status]} solid"
),
}
)
app.layout = dbc.Row(
[
dbc.Container(f"Run: {object.name}", class_name="h1"),
dbc.Row(
[
dbc.Col(
[
dbc.Row(
[
html.Span(
[
html.Span(
[
html.I(
className="bi bi-circle-fill me-1"
),
"Step",
],
className="me-2",
),
html.Span(
[
html.I(
className="bi bi-square-fill me-1"
),
"Artifact",
],
className="me-4",
),
dbc.Badge(
"Completed",
color=COLOR_BLUE,
className="me-1",
),
dbc.Badge(
"Cached",
color=COLOR_GREEN,
className="me-1",
),
dbc.Badge(
"Running",
color=COLOR_YELLOW,
className="me-1",
),
dbc.Badge(
"Failed",
color=COLOR_RED,
className="me-1",
),
]
),
]
),
dbc.Row(
[
cyto.Cytoscape(
id="cytoscape",
layout={
"name": "breadthfirst",
"roots": f'[id = "{first_step_id}"]',
},
elements=edges + nodes,
stylesheet=STYLESHEET,
style={
"width": "100%",
"height": "800px",
},
zoom=1,
)
]
),
dbc.Row(
[
dbc.Button(
"Reset",
id="bt-reset",
color="primary",
className="me-1",
)
]
),
]
),
dbc.Col(
[
dcc.Markdown(id="markdown-selected-node-data"),
]
),
]
),
],
className="p-5",
)
@app.callback( # type: ignore[misc]
Output("markdown-selected-node-data", "children"),
Input("cytoscape", "selectedNodeData"),
)
def display_data(data_list: List[Dict[str, Any]]) -> str:
"""Callback for the text area below the graph"""
if data_list is None:
return "Click on a node in the diagram."
text = ""
for data in data_list:
text += f'## {data["execution_id"]} / {data["name"]}' + "\n\n"
if data["type"] == "artifact":
for item in [
"artifact_data_type",
"is_cached",
"producer_step_id",
"parent_step_id",
"uri",
]:
text += f"**{item}**: {data[item]}" + "\n\n"
elif data["type"] == "step":
text += "### Inputs:" + "\n\n"
for k, v in data["inputs"].items():
text += f"**{k}**: {v}" + "\n\n"
text += "### Outputs:" + "\n\n"
for k, v in data["outputs"].items():
text += f"**{k}**: {v}" + "\n\n"
text += "### Params:"
for k, v in data["parameters"].items():
text += f"**{k}**: {v}" + "\n\n"
return text
@app.callback( # type: ignore[misc]
[Output("cytoscape", "zoom"), Output("cytoscape", "elements")],
[Input("bt-reset", "n_clicks")],
)
def reset_layout(
n_clicks: int,
) -> List[Union[int, List[Dict[str, Collection[str]]]]]:
"""Resets the layout"""
logger.debug(n_clicks, "clicked in reset button.")
return [1, edges + nodes]
app.run_server()
return app
visualize(self, object, *args, **kwargs)
Method to visualize pipeline runs via the Dash library. The layout puts every layer of the dag in a column.
Source code in zenml/integrations/dash/visualizers/pipeline_run_lineage_visualizer.py
def visualize(
self, object: PipelineRunView, *args: Any, **kwargs: Any
) -> dash.Dash:
"""Method to visualize pipeline runs via the Dash library. The layout
puts every layer of the dag in a column.
"""
app = dash.Dash(
__name__,
external_stylesheets=[
dbc.themes.BOOTSTRAP,
dbc.icons.BOOTSTRAP,
],
)
nodes, edges, first_step_id = [], [], None
first_step_id = None
for step in object.steps:
step_output_artifacts = list(step.outputs.values())
execution_id = (
step_output_artifacts[0].producer_step.id
if step_output_artifacts
else step.id
)
step_id = self.STEP_PREFIX + str(step.id)
if first_step_id is None:
first_step_id = step_id
nodes.append(
{
"data": {
"id": step_id,
"execution_id": execution_id,
"label": f"{execution_id} / {step.entrypoint_name}",
"entrypoint_name": step.entrypoint_name, # redundant for consistency
"name": step.name, # redundant for consistency
"type": "step",
"parameters": step.parameters,
"inputs": {k: v.uri for k, v in step.inputs.items()},
"outputs": {k: v.uri for k, v in step.outputs.items()},
},
"classes": self.STATUS_CLASS_MAPPING[step.status],
}
)
for artifact_name, artifact in step.outputs.items():
nodes.append(
{
"data": {
"id": self.ARTIFACT_PREFIX + str(artifact.id),
"execution_id": artifact.id,
"label": f"{artifact.id} / {artifact_name} ("
f"{artifact.data_type})",
"type": "artifact",
"name": artifact_name,
"is_cached": artifact.is_cached,
"artifact_type": artifact.type,
"artifact_data_type": artifact.data_type,
"parent_step_id": artifact.parent_step_id,
"producer_step_id": artifact.producer_step.id,
"uri": artifact.uri,
},
"classes": f"rectangle "
f"{self.STATUS_CLASS_MAPPING[step.status]}",
}
)
edges.append(
{
"data": {
"source": self.STEP_PREFIX + str(step.id),
"target": self.ARTIFACT_PREFIX + str(artifact.id),
},
"classes": f"edge-arrow "
f"{self.STATUS_CLASS_MAPPING[step.status]}"
+ (" dashed" if artifact.is_cached else " solid"),
}
)
for artifact_name, artifact in step.inputs.items():
edges.append(
{
"data": {
"source": self.ARTIFACT_PREFIX + str(artifact.id),
"target": self.STEP_PREFIX + str(step.id),
},
"classes": "edge-arrow "
+ (
f"{self.STATUS_CLASS_MAPPING[ExecutionStatus.CACHED]} dashed"
if artifact.is_cached
else f"{self.STATUS_CLASS_MAPPING[step.status]} solid"
),
}
)
app.layout = dbc.Row(
[
dbc.Container(f"Run: {object.name}", class_name="h1"),
dbc.Row(
[
dbc.Col(
[
dbc.Row(
[
html.Span(
[
html.Span(
[
html.I(
className="bi bi-circle-fill me-1"
),
"Step",
],
className="me-2",
),
html.Span(
[
html.I(
className="bi bi-square-fill me-1"
),
"Artifact",
],
className="me-4",
),
dbc.Badge(
"Completed",
color=COLOR_BLUE,
className="me-1",
),
dbc.Badge(
"Cached",
color=COLOR_GREEN,
className="me-1",
),
dbc.Badge(
"Running",
color=COLOR_YELLOW,
className="me-1",
),
dbc.Badge(
"Failed",
color=COLOR_RED,
className="me-1",
),
]
),
]
),
dbc.Row(
[
cyto.Cytoscape(
id="cytoscape",
layout={
"name": "breadthfirst",
"roots": f'[id = "{first_step_id}"]',
},
elements=edges + nodes,
stylesheet=STYLESHEET,
style={
"width": "100%",
"height": "800px",
},
zoom=1,
)
]
),
dbc.Row(
[
dbc.Button(
"Reset",
id="bt-reset",
color="primary",
className="me-1",
)
]
),
]
),
dbc.Col(
[
dcc.Markdown(id="markdown-selected-node-data"),
]
),
]
),
],
className="p-5",
)
@app.callback( # type: ignore[misc]
Output("markdown-selected-node-data", "children"),
Input("cytoscape", "selectedNodeData"),
)
def display_data(data_list: List[Dict[str, Any]]) -> str:
"""Callback for the text area below the graph"""
if data_list is None:
return "Click on a node in the diagram."
text = ""
for data in data_list:
text += f'## {data["execution_id"]} / {data["name"]}' + "\n\n"
if data["type"] == "artifact":
for item in [
"artifact_data_type",
"is_cached",
"producer_step_id",
"parent_step_id",
"uri",
]:
text += f"**{item}**: {data[item]}" + "\n\n"
elif data["type"] == "step":
text += "### Inputs:" + "\n\n"
for k, v in data["inputs"].items():
text += f"**{k}**: {v}" + "\n\n"
text += "### Outputs:" + "\n\n"
for k, v in data["outputs"].items():
text += f"**{k}**: {v}" + "\n\n"
text += "### Params:"
for k, v in data["parameters"].items():
text += f"**{k}**: {v}" + "\n\n"
return text
@app.callback( # type: ignore[misc]
[Output("cytoscape", "zoom"), Output("cytoscape", "elements")],
[Input("bt-reset", "n_clicks")],
)
def reset_layout(
n_clicks: int,
) -> List[Union[int, List[Dict[str, Collection[str]]]]]:
"""Resets the layout"""
logger.debug(n_clicks, "clicked in reset button.")
return [1, edges + nodes]
app.run_server()
return app
evidently
special
The Evidently integration provides a way to monitor your models in production. It includes a way to detect data drift and different kinds of model performance issues.
The results of Evidently calculations can either be exported as an interactive dashboard (visualized as an html file or in your Jupyter notebook), or as a JSON file.
EvidentlyIntegration (Integration)
Definition of Evidently integration for ZenML.
Source code in zenml/integrations/evidently/__init__.py
class EvidentlyIntegration(Integration):
"""Definition of [Evidently](https://github.com/evidentlyai/evidently) integration
for ZenML."""
NAME = EVIDENTLY
REQUIREMENTS = ["evidently==v0.1.41.dev0"]
steps
special
evidently_profile
EvidentlyProfileConfig (BaseDriftDetectionConfig)
pydantic-model
Config class for Evidently profile steps.
column_mapping: properties of the dataframe's columns used !!! profile_section "a string that identifies the profile section to be used." The following are valid options supported by Evidently: - "datadrift" - "categoricaltargetdrift" - "numericaltargetdrift" - "classificationmodelperformance" - "regressionmodelperformance" - "probabilisticmodelperformance"
Source code in zenml/integrations/evidently/steps/evidently_profile.py
class EvidentlyProfileConfig(BaseDriftDetectionConfig):
"""Config class for Evidently profile steps.
column_mapping: properties of the dataframe's columns used
profile_section: a string that identifies the profile section to be used.
The following are valid options supported by Evidently:
- "datadrift"
- "categoricaltargetdrift"
- "numericaltargetdrift"
- "classificationmodelperformance"
- "regressionmodelperformance"
- "probabilisticmodelperformance"
"""
def get_profile_sections_and_tabs(
self,
) -> Tuple[List[ProfileSection], List[Tab]]:
try:
return (
[
profile_mapper[profile]()
for profile in self.profile_sections
],
[
dashboard_mapper[profile]()
for profile in self.profile_sections
],
)
except KeyError:
nl = "\n"
raise ValueError(
f"Invalid profile section: {self.profile_sections} \n\n"
f"Valid and supported options are: {nl}- "
f'{f"{nl}- ".join(list(profile_mapper.keys()))}'
)
column_mapping: Optional[ColumnMapping]
profile_sections: Sequence[str]
EvidentlyProfileStep (BaseDriftDetectionStep)
Simple step implementation which implements Evidently's functionality for creating a profile.
Source code in zenml/integrations/evidently/steps/evidently_profile.py
class EvidentlyProfileStep(BaseDriftDetectionStep):
"""Simple step implementation which implements Evidently's functionality for
creating a profile."""
OUTPUT_SPEC = {
"profile": DataAnalysisArtifact,
"dashboard": DataAnalysisArtifact,
}
def entrypoint( # type: ignore[override]
self,
reference_dataset: pd.DataFrame,
comparison_dataset: pd.DataFrame,
config: EvidentlyProfileConfig,
) -> Output( # type:ignore[valid-type]
profile=dict, dashboard=str
):
"""Main entrypoint for the Evidently categorical target drift detection
step.
Args:
reference_dataset: a Pandas dataframe
comparison_dataset: a Pandas dataframe of new data you wish to
compare against the reference data
config: the configuration for the step
Returns:
profile: dictionary report extracted from an Evidently Profile
generated for the data drift
dashboard: HTML report extracted from an Evidently Dashboard
generated for the data drift
"""
sections, tabs = config.get_profile_sections_and_tabs()
data_drift_dashboard = Dashboard(tabs=tabs)
data_drift_dashboard.calculate(
reference_dataset,
comparison_dataset,
column_mapping=config.column_mapping or None,
)
data_drift_profile = Profile(sections=sections)
data_drift_profile.calculate(
reference_dataset,
comparison_dataset,
column_mapping=config.column_mapping or None,
)
return [data_drift_profile.object(), data_drift_dashboard.html()]
CONFIG_CLASS (BaseDriftDetectionConfig)
pydantic-model
Config class for Evidently profile steps.
column_mapping: properties of the dataframe's columns used !!! profile_section "a string that identifies the profile section to be used." The following are valid options supported by Evidently: - "datadrift" - "categoricaltargetdrift" - "numericaltargetdrift" - "classificationmodelperformance" - "regressionmodelperformance" - "probabilisticmodelperformance"
Source code in zenml/integrations/evidently/steps/evidently_profile.py
class EvidentlyProfileConfig(BaseDriftDetectionConfig):
"""Config class for Evidently profile steps.
column_mapping: properties of the dataframe's columns used
profile_section: a string that identifies the profile section to be used.
The following are valid options supported by Evidently:
- "datadrift"
- "categoricaltargetdrift"
- "numericaltargetdrift"
- "classificationmodelperformance"
- "regressionmodelperformance"
- "probabilisticmodelperformance"
"""
def get_profile_sections_and_tabs(
self,
) -> Tuple[List[ProfileSection], List[Tab]]:
try:
return (
[
profile_mapper[profile]()
for profile in self.profile_sections
],
[
dashboard_mapper[profile]()
for profile in self.profile_sections
],
)
except KeyError:
nl = "\n"
raise ValueError(
f"Invalid profile section: {self.profile_sections} \n\n"
f"Valid and supported options are: {nl}- "
f'{f"{nl}- ".join(list(profile_mapper.keys()))}'
)
column_mapping: Optional[ColumnMapping]
profile_sections: Sequence[str]
entrypoint(self, reference_dataset, comparison_dataset, config)
Main entrypoint for the Evidently categorical target drift detection step.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
reference_dataset |
DataFrame |
a Pandas dataframe |
required |
comparison_dataset |
DataFrame |
a Pandas dataframe of new data you wish to compare against the reference data |
required |
config |
EvidentlyProfileConfig |
the configuration for the step |
required |
Returns:
| Type | Description |
|---|---|
profile |
dictionary report extracted from an Evidently Profile generated for the data drift dashboard: HTML report extracted from an Evidently Dashboard generated for the data drift |
Source code in zenml/integrations/evidently/steps/evidently_profile.py
def entrypoint( # type: ignore[override]
self,
reference_dataset: pd.DataFrame,
comparison_dataset: pd.DataFrame,
config: EvidentlyProfileConfig,
) -> Output( # type:ignore[valid-type]
profile=dict, dashboard=str
):
"""Main entrypoint for the Evidently categorical target drift detection
step.
Args:
reference_dataset: a Pandas dataframe
comparison_dataset: a Pandas dataframe of new data you wish to
compare against the reference data
config: the configuration for the step
Returns:
profile: dictionary report extracted from an Evidently Profile
generated for the data drift
dashboard: HTML report extracted from an Evidently Dashboard
generated for the data drift
"""
sections, tabs = config.get_profile_sections_and_tabs()
data_drift_dashboard = Dashboard(tabs=tabs)
data_drift_dashboard.calculate(
reference_dataset,
comparison_dataset,
column_mapping=config.column_mapping or None,
)
data_drift_profile = Profile(sections=sections)
data_drift_profile.calculate(
reference_dataset,
comparison_dataset,
column_mapping=config.column_mapping or None,
)
return [data_drift_profile.object(), data_drift_dashboard.html()]
visualizers
special
evidently_visualizer
EvidentlyVisualizer (BaseStepVisualizer)
The implementation of an Evidently Visualizer.
Source code in zenml/integrations/evidently/visualizers/evidently_visualizer.py
class EvidentlyVisualizer(BaseStepVisualizer):
"""The implementation of an Evidently Visualizer."""
@abstractmethod
def visualize(self, object: StepView, *args: Any, **kwargs: Any) -> None:
"""Method to visualize components
Args:
object: StepView fetched from run.get_step().
"""
for artifact_view in object.outputs.values():
# filter out anything but data analysis artifacts
if (
artifact_view.type == DataAnalysisArtifact.__name__
and artifact_view.data_type == "builtins.str"
):
artifact = artifact_view.read()
self.generate_facet(artifact)
def generate_facet(self, html_: str) -> None:
"""Generate a Facet Overview
Args:
html_: HTML represented as a string.
"""
if Environment.in_notebook():
from IPython.core.display import HTML, display
display(HTML(html_))
else:
logger.warning(
"The magic functions are only usable in a Jupyter notebook."
)
with tempfile.NamedTemporaryFile(
mode="w", delete=False, suffix=".html", encoding="utf-8"
) as f:
f.write(html_)
url = f"file:///{f.name}"
logger.info("Opening %s in a new browser.." % f.name)
webbrowser.open(url, new=2)
generate_facet(self, html_)
Generate a Facet Overview
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
html_ |
str |
HTML represented as a string. |
required |
Source code in zenml/integrations/evidently/visualizers/evidently_visualizer.py
def generate_facet(self, html_: str) -> None:
"""Generate a Facet Overview
Args:
html_: HTML represented as a string.
"""
if Environment.in_notebook():
from IPython.core.display import HTML, display
display(HTML(html_))
else:
logger.warning(
"The magic functions are only usable in a Jupyter notebook."
)
with tempfile.NamedTemporaryFile(
mode="w", delete=False, suffix=".html", encoding="utf-8"
) as f:
f.write(html_)
url = f"file:///{f.name}"
logger.info("Opening %s in a new browser.." % f.name)
webbrowser.open(url, new=2)
visualize(self, object, *args, **kwargs)
Method to visualize components
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
object |
StepView |
StepView fetched from run.get_step(). |
required |
Source code in zenml/integrations/evidently/visualizers/evidently_visualizer.py
@abstractmethod
def visualize(self, object: StepView, *args: Any, **kwargs: Any) -> None:
"""Method to visualize components
Args:
object: StepView fetched from run.get_step().
"""
for artifact_view in object.outputs.values():
# filter out anything but data analysis artifacts
if (
artifact_view.type == DataAnalysisArtifact.__name__
and artifact_view.data_type == "builtins.str"
):
artifact = artifact_view.read()
self.generate_facet(artifact)
facets
special
The Facets integration provides a simple way to visualize post-execution objects
like PipelineView, PipelineRunView and StepView. These objects can be
extended using the BaseVisualization class. This integration requires
facets-overview be installed in your Python environment.
FacetsIntegration (Integration)
Definition of Facet integration for ZenML.
Source code in zenml/integrations/facets/__init__.py
class FacetsIntegration(Integration):
"""Definition of [Facet](https://pair-code.github.io/facets/) integration
for ZenML."""
NAME = FACETS
REQUIREMENTS = ["facets-overview>=1.0.0", "IPython"]
visualizers
special
facet_statistics_visualizer
FacetStatisticsVisualizer (BaseStepVisualizer)
The base implementation of a ZenML Visualizer.
Source code in zenml/integrations/facets/visualizers/facet_statistics_visualizer.py
class FacetStatisticsVisualizer(BaseStepVisualizer):
"""The base implementation of a ZenML Visualizer."""
@abstractmethod
def visualize(
self, object: StepView, magic: bool = False, *args: Any, **kwargs: Any
) -> None:
"""Method to visualize components
Args:
object: StepView fetched from run.get_step().
magic: Whether to render in a Jupyter notebook or not.
"""
datasets = []
for output_name, artifact_view in object.outputs.items():
df = artifact_view.read()
if type(df) is not pd.DataFrame:
logger.warning(
"`%s` is not a pd.DataFrame. You can only visualize "
"statistics of steps that output pandas dataframes. "
"Skipping this output.." % output_name
)
else:
datasets.append({"name": output_name, "table": df})
h = self.generate_html(datasets)
self.generate_facet(h, magic)
def generate_html(self, datasets: List[Dict[Text, pd.DataFrame]]) -> str:
"""Generates html for facet.
Args:
datasets: List of dicts of dataframes to be visualized as stats.
Returns:
HTML template with proto string embedded.
"""
proto = GenericFeatureStatisticsGenerator().ProtoFromDataFrames(
datasets
)
protostr = base64.b64encode(proto.SerializeToString()).decode("utf-8")
template = os.path.join(
os.path.abspath(os.path.dirname(__file__)),
"stats.html",
)
html_template = zenml.io.utils.read_file_contents_as_string(template)
html_ = html_template.replace("protostr", protostr)
return html_
def generate_facet(self, html_: str, magic: bool = False) -> None:
"""Generate a Facet Overview
Args:
html_: HTML represented as a string.
magic: Whether to magically materialize facet in a notebook.
"""
if magic:
if not Environment.in_notebook():
raise EnvironmentError(
"The magic functions are only usable in a Jupyter notebook."
)
display(HTML(html_))
else:
with tempfile.NamedTemporaryFile(delete=False, suffix=".html") as f:
zenml.io.utils.write_file_contents_as_string(f.name, html_)
url = f"file:///{f.name}"
logger.info("Opening %s in a new browser.." % f.name)
webbrowser.open(url, new=2)
generate_facet(self, html_, magic=False)
Generate a Facet Overview
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
html_ |
str |
HTML represented as a string. |
required |
magic |
bool |
Whether to magically materialize facet in a notebook. |
False |
Source code in zenml/integrations/facets/visualizers/facet_statistics_visualizer.py
def generate_facet(self, html_: str, magic: bool = False) -> None:
"""Generate a Facet Overview
Args:
html_: HTML represented as a string.
magic: Whether to magically materialize facet in a notebook.
"""
if magic:
if not Environment.in_notebook():
raise EnvironmentError(
"The magic functions are only usable in a Jupyter notebook."
)
display(HTML(html_))
else:
with tempfile.NamedTemporaryFile(delete=False, suffix=".html") as f:
zenml.io.utils.write_file_contents_as_string(f.name, html_)
url = f"file:///{f.name}"
logger.info("Opening %s in a new browser.." % f.name)
webbrowser.open(url, new=2)
generate_html(self, datasets)
Generates html for facet.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
datasets |
List[Dict[str, pandas.core.frame.DataFrame]] |
List of dicts of dataframes to be visualized as stats. |
required |
Returns:
| Type | Description |
|---|---|
str |
HTML template with proto string embedded. |
Source code in zenml/integrations/facets/visualizers/facet_statistics_visualizer.py
def generate_html(self, datasets: List[Dict[Text, pd.DataFrame]]) -> str:
"""Generates html for facet.
Args:
datasets: List of dicts of dataframes to be visualized as stats.
Returns:
HTML template with proto string embedded.
"""
proto = GenericFeatureStatisticsGenerator().ProtoFromDataFrames(
datasets
)
protostr = base64.b64encode(proto.SerializeToString()).decode("utf-8")
template = os.path.join(
os.path.abspath(os.path.dirname(__file__)),
"stats.html",
)
html_template = zenml.io.utils.read_file_contents_as_string(template)
html_ = html_template.replace("protostr", protostr)
return html_
visualize(self, object, magic=False, *args, **kwargs)
Method to visualize components
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
object |
StepView |
StepView fetched from run.get_step(). |
required |
magic |
bool |
Whether to render in a Jupyter notebook or not. |
False |
Source code in zenml/integrations/facets/visualizers/facet_statistics_visualizer.py
@abstractmethod
def visualize(
self, object: StepView, magic: bool = False, *args: Any, **kwargs: Any
) -> None:
"""Method to visualize components
Args:
object: StepView fetched from run.get_step().
magic: Whether to render in a Jupyter notebook or not.
"""
datasets = []
for output_name, artifact_view in object.outputs.items():
df = artifact_view.read()
if type(df) is not pd.DataFrame:
logger.warning(
"`%s` is not a pd.DataFrame. You can only visualize "
"statistics of steps that output pandas dataframes. "
"Skipping this output.." % output_name
)
else:
datasets.append({"name": output_name, "table": df})
h = self.generate_html(datasets)
self.generate_facet(h, magic)
feast
special
The Feast integration offers a way to connect to a Feast Feature Store. ZenML implements a dedicated stack component that you can access as part of your ZenML steps in the usual ways.
FeastIntegration (Integration)
Definition of Feast integration for ZenML.
Source code in zenml/integrations/feast/__init__.py
class FeastIntegration(Integration):
"""Definition of Feast integration for ZenML."""
NAME = FEAST
REQUIREMENTS = ["feast[redis]>=0.19.4", "redis-server"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.feast import feature_stores # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/feast/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.feast import feature_stores # noqa
feature_stores
special
Feast Feature Store
Feature stores allow data teams to serve data via an offline store and an online low-latency store where data is kept in sync between the two. It also offers a centralized registry where features (and feature schemas) are stored for use within a team or wider organization. Feature stores are a relatively recent addition to commonly-used machine learning stacks. Feast is a leading open-source feature store, first developed by Gojek in collaboration with Google.
feast_feature_store
FeastFeatureStore (BaseFeatureStore)
pydantic-model
Class to interact with the Feast feature store.
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
class FeastFeatureStore(BaseFeatureStore):
"""Class to interact with the Feast feature store."""
FLAVOR: ClassVar[str] = "feast"
online_host: str = "localhost"
online_port: int = 6379
feast_repo: str
def _validate_connection(self) -> None:
"""Validates the connection to the feature store.
Raises:
RuntimeError: If the online component (Redis) is not available.
"""
client = redis.Redis(host=self.online_host, port=self.online_port)
try:
client.ping()
except redis.exceptions.ConnectionError as e:
raise redis.exceptions.ConnectionError(
"Could not connect to feature store's online component. "
"Please make sure that Redis is running."
) from e
def get_historical_features(
self,
entity_df: Union[pd.DataFrame, str],
features: List[str],
full_feature_names: bool = False,
) -> pd.DataFrame:
"""Returns the historical features for training or batch scoring.
Args:
entity_df: The entity dataframe or entity name.
features: The features to retrieve.
full_feature_names: Whether to return the full feature names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The historical features as a Pandas DataFrame.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return fs.get_historical_features(
entity_df=entity_df,
features=features,
full_feature_names=full_feature_names,
).to_df()
def get_online_features(
self,
entity_rows: List[Dict[str, Any]],
features: List[str],
full_feature_names: bool = False,
) -> Dict[str, Any]:
"""Returns the latest online feature data.
Args:
entity_rows: The entity rows to retrieve.
features: The features to retrieve.
full_feature_names: Whether to return the full feature names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The latest online feature data as a dictionary.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return fs.get_online_features( # type: ignore[no-any-return]
entity_rows=entity_rows,
features=features,
full_feature_names=full_feature_names,
).to_dict()
def get_data_sources(self) -> List[str]:
"""Returns the data sources' names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The data sources' names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_data_sources()]
def get_entities(self) -> List[str]:
"""Returns the entity names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The entity names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_entities()]
def get_feature_services(self) -> List[str]:
"""Returns the feature service names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The feature service names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_feature_services()]
def get_feature_views(self) -> List[str]:
"""Returns the feature view names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The feature view names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_feature_views()]
def get_project(self) -> str:
"""Returns the project name.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The project name.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return str(fs.project)
def get_registry(self) -> Registry:
"""Returns the feature store registry.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The registry.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return fs.registry
def get_feast_version(self) -> str:
"""Returns the version of Feast used.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The version of Feast currently being used.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return str(fs.version())
get_data_sources(self)
Returns the data sources' names.
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
List[str] |
The data sources' names. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_data_sources(self) -> List[str]:
"""Returns the data sources' names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The data sources' names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_data_sources()]
get_entities(self)
Returns the entity names.
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
List[str] |
The entity names. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_entities(self) -> List[str]:
"""Returns the entity names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The entity names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_entities()]
get_feast_version(self)
Returns the version of Feast used.
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
str |
The version of Feast currently being used. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_feast_version(self) -> str:
"""Returns the version of Feast used.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The version of Feast currently being used.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return str(fs.version())
get_feature_services(self)
Returns the feature service names.
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
List[str] |
The feature service names. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_feature_services(self) -> List[str]:
"""Returns the feature service names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The feature service names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_feature_services()]
get_feature_views(self)
Returns the feature view names.
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
List[str] |
The feature view names. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_feature_views(self) -> List[str]:
"""Returns the feature view names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The feature view names.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return [ds.name for ds in fs.list_feature_views()]
get_historical_features(self, entity_df, features, full_feature_names=False)
Returns the historical features for training or batch scoring.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
entity_df |
Union[pandas.core.frame.DataFrame, str] |
The entity dataframe or entity name. |
required |
features |
List[str] |
The features to retrieve. |
required |
full_feature_names |
bool |
Whether to return the full feature names. |
False |
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
DataFrame |
The historical features as a Pandas DataFrame. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_historical_features(
self,
entity_df: Union[pd.DataFrame, str],
features: List[str],
full_feature_names: bool = False,
) -> pd.DataFrame:
"""Returns the historical features for training or batch scoring.
Args:
entity_df: The entity dataframe or entity name.
features: The features to retrieve.
full_feature_names: Whether to return the full feature names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The historical features as a Pandas DataFrame.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return fs.get_historical_features(
entity_df=entity_df,
features=features,
full_feature_names=full_feature_names,
).to_df()
get_online_features(self, entity_rows, features, full_feature_names=False)
Returns the latest online feature data.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
entity_rows |
List[Dict[str, Any]] |
The entity rows to retrieve. |
required |
features |
List[str] |
The features to retrieve. |
required |
full_feature_names |
bool |
Whether to return the full feature names. |
False |
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
Dict[str, Any] |
The latest online feature data as a dictionary. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_online_features(
self,
entity_rows: List[Dict[str, Any]],
features: List[str],
full_feature_names: bool = False,
) -> Dict[str, Any]:
"""Returns the latest online feature data.
Args:
entity_rows: The entity rows to retrieve.
features: The features to retrieve.
full_feature_names: Whether to return the full feature names.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The latest online feature data as a dictionary.
"""
self._validate_connection()
fs = FeatureStore(repo_path=self.feast_repo)
return fs.get_online_features( # type: ignore[no-any-return]
entity_rows=entity_rows,
features=features,
full_feature_names=full_feature_names,
).to_dict()
get_project(self)
Returns the project name.
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
str |
The project name. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_project(self) -> str:
"""Returns the project name.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The project name.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return str(fs.project)
get_registry(self)
Returns the feature store registry.
Exceptions:
| Type | Description |
|---|---|
ConnectionError |
If the online component (Redis) is not available. |
Returns:
| Type | Description |
|---|---|
Registry |
The registry. |
Source code in zenml/integrations/feast/feature_stores/feast_feature_store.py
def get_registry(self) -> Registry:
"""Returns the feature store registry.
Raise:
ConnectionError: If the online component (Redis) is not available.
Returns:
The registry.
"""
fs = FeatureStore(repo_path=self.feast_repo)
return fs.registry
gcp
special
The GCP integration submodule provides a way to run ZenML pipelines in a cloud
environment. Specifically, it allows the use of cloud artifact stores, metadata
stores, and an io module to handle file operations on Google Cloud Storage (GCS).
GcpIntegration (Integration)
Definition of Google Cloud Platform integration for ZenML.
Source code in zenml/integrations/gcp/__init__.py
class GcpIntegration(Integration):
"""Definition of Google Cloud Platform integration for ZenML."""
NAME = GCP
REQUIREMENTS = ["gcsfs"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.gcp import artifact_stores # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/gcp/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.gcp import artifact_stores # noqa
artifact_stores
special
gcp_artifact_store
GCPArtifactStore (BaseArtifactStore)
pydantic-model
Artifact Store for Google Cloud Storage based artifacts.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
class GCPArtifactStore(BaseArtifactStore):
"""Artifact Store for Google Cloud Storage based artifacts."""
_filesystem: Optional[gcsfs.GCSFileSystem] = None
# Class Configuration
FLAVOR: ClassVar[str] = "gcp"
SUPPORTED_SCHEMES: ClassVar[Set[str]] = {"gs://"}
@property
def filesystem(self) -> gcsfs.GCSFileSystem:
"""The gcsfs filesystem to access this artifact store."""
if not self._filesystem:
self._filesystem = gcsfs.GCSFileSystem()
return self._filesystem
def open(self, path: PathType, mode: str = "r") -> Any:
"""Open a file at the given path.
Args:
path: Path of the file to open.
mode: Mode in which to open the file. Currently, only
'rb' and 'wb' to read and write binary files are supported.
"""
return self.filesystem.open(path=path, mode=mode)
def copyfile(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Copy a file.
Args:
src: The path to copy from.
dst: The path to copy to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to copy to destination '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to copy anyway."
)
# TODO [ENG-151]: Check if it works with overwrite=True or if we need to
# manually remove it first
self.filesystem.copy(path1=src, path2=dst)
def exists(self, path: PathType) -> bool:
"""Check whether a path exists."""
return self.filesystem.exists(path=path) # type: ignore[no-any-return]
def glob(self, pattern: PathType) -> List[PathType]:
"""Return all paths that match the given glob pattern.
The glob pattern may include:
- '*' to match any number of characters
- '?' to match a single character
- '[...]' to match one of the characters inside the brackets
- '**' as the full name of a path component to match to search
in subdirectories of any depth (e.g. '/some_dir/**/some_file)
Args:
pattern: The glob pattern to match, see details above.
Returns:
A list of paths that match the given glob pattern.
"""
return self.filesystem.glob(path=pattern) # type: ignore[no-any-return]
def isdir(self, path: PathType) -> bool:
"""Check whether a path is a directory."""
return self.filesystem.isdir(path=path) # type: ignore[no-any-return]
def listdir(self, path: PathType) -> List[PathType]:
"""Return a list of files in a directory."""
return self.filesystem.listdir(path=path) # type: ignore[no-any-return]
def makedirs(self, path: PathType) -> None:
"""Create a directory at the given path. If needed also
create missing parent directories."""
self.filesystem.makedirs(path=path, exist_ok=True)
def mkdir(self, path: PathType) -> None:
"""Create a directory at the given path."""
self.filesystem.makedir(path=path)
def remove(self, path: PathType) -> None:
"""Remove the file at the given path."""
self.filesystem.rm_file(path=path)
def rename(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Rename source file to destination file.
Args:
src: The path of the file to rename.
dst: The path to rename the source file to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to rename file to '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to rename anyway."
)
# TODO [ENG-152]: Check if it works with overwrite=True or if we need
# to manually remove it first
self.filesystem.rename(path1=src, path2=dst)
def rmtree(self, path: PathType) -> None:
"""Remove the given directory."""
self.filesystem.delete(path=path, recursive=True)
def stat(self, path: PathType) -> Dict[str, Any]:
"""Return stat info for the given path."""
return self.filesystem.stat(path=path) # type: ignore[no-any-return]
def walk(
self,
top: PathType,
topdown: bool = True,
onerror: Optional[Callable[..., None]] = None,
) -> Iterable[Tuple[PathType, List[PathType], List[PathType]]]:
"""Return an iterator that walks the contents of the given directory.
Args:
top: Path of directory to walk.
topdown: Unused argument to conform to interface.
onerror: Unused argument to conform to interface.
Returns:
An Iterable of Tuples, each of which contain the path of the current
directory path, a list of directories inside the current directory
and a list of files inside the current directory.
"""
# TODO [ENG-153]: Additional params
return self.filesystem.walk(path=top) # type: ignore[no-any-return]
filesystem: GCSFileSystem
property
readonly
The gcsfs filesystem to access this artifact store.
copyfile(self, src, dst, overwrite=False)
Copy a file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
src |
Union[bytes, str] |
The path to copy from. |
required |
dst |
Union[bytes, str] |
The path to copy to. |
required |
overwrite |
bool |
If a file already exists at the destination, this
method will overwrite it if overwrite= |
False |
Exceptions:
| Type | Description |
|---|---|
FileNotFoundError |
If the source file does not exist. |
FileExistsError |
If a file already exists at the destination
and overwrite is not set to |
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def copyfile(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Copy a file.
Args:
src: The path to copy from.
dst: The path to copy to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to copy to destination '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to copy anyway."
)
# TODO [ENG-151]: Check if it works with overwrite=True or if we need to
# manually remove it first
self.filesystem.copy(path1=src, path2=dst)
exists(self, path)
Check whether a path exists.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def exists(self, path: PathType) -> bool:
"""Check whether a path exists."""
return self.filesystem.exists(path=path) # type: ignore[no-any-return]
glob(self, pattern)
Return all paths that match the given glob pattern. The glob pattern may include: - '' to match any number of characters - '?' to match a single character - '[...]' to match one of the characters inside the brackets - '' as the full name of a path component to match to search in subdirectories of any depth (e.g. '/some_dir/*/some_file)
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pattern |
Union[bytes, str] |
The glob pattern to match, see details above. |
required |
Returns:
| Type | Description |
|---|---|
List[Union[bytes, str]] |
A list of paths that match the given glob pattern. |
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def glob(self, pattern: PathType) -> List[PathType]:
"""Return all paths that match the given glob pattern.
The glob pattern may include:
- '*' to match any number of characters
- '?' to match a single character
- '[...]' to match one of the characters inside the brackets
- '**' as the full name of a path component to match to search
in subdirectories of any depth (e.g. '/some_dir/**/some_file)
Args:
pattern: The glob pattern to match, see details above.
Returns:
A list of paths that match the given glob pattern.
"""
return self.filesystem.glob(path=pattern) # type: ignore[no-any-return]
isdir(self, path)
Check whether a path is a directory.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def isdir(self, path: PathType) -> bool:
"""Check whether a path is a directory."""
return self.filesystem.isdir(path=path) # type: ignore[no-any-return]
listdir(self, path)
Return a list of files in a directory.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def listdir(self, path: PathType) -> List[PathType]:
"""Return a list of files in a directory."""
return self.filesystem.listdir(path=path) # type: ignore[no-any-return]
makedirs(self, path)
Create a directory at the given path. If needed also create missing parent directories.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def makedirs(self, path: PathType) -> None:
"""Create a directory at the given path. If needed also
create missing parent directories."""
self.filesystem.makedirs(path=path, exist_ok=True)
mkdir(self, path)
Create a directory at the given path.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def mkdir(self, path: PathType) -> None:
"""Create a directory at the given path."""
self.filesystem.makedir(path=path)
open(self, path, mode='r')
Open a file at the given path.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
path |
Union[bytes, str] |
Path of the file to open. |
required |
mode |
str |
Mode in which to open the file. Currently, only 'rb' and 'wb' to read and write binary files are supported. |
'r' |
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def open(self, path: PathType, mode: str = "r") -> Any:
"""Open a file at the given path.
Args:
path: Path of the file to open.
mode: Mode in which to open the file. Currently, only
'rb' and 'wb' to read and write binary files are supported.
"""
return self.filesystem.open(path=path, mode=mode)
remove(self, path)
Remove the file at the given path.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def remove(self, path: PathType) -> None:
"""Remove the file at the given path."""
self.filesystem.rm_file(path=path)
rename(self, src, dst, overwrite=False)
Rename source file to destination file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
src |
Union[bytes, str] |
The path of the file to rename. |
required |
dst |
Union[bytes, str] |
The path to rename the source file to. |
required |
overwrite |
bool |
If a file already exists at the destination, this
method will overwrite it if overwrite= |
False |
Exceptions:
| Type | Description |
|---|---|
FileNotFoundError |
If the source file does not exist. |
FileExistsError |
If a file already exists at the destination
and overwrite is not set to |
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def rename(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Rename source file to destination file.
Args:
src: The path of the file to rename.
dst: The path to rename the source file to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to rename file to '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to rename anyway."
)
# TODO [ENG-152]: Check if it works with overwrite=True or if we need
# to manually remove it first
self.filesystem.rename(path1=src, path2=dst)
rmtree(self, path)
Remove the given directory.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def rmtree(self, path: PathType) -> None:
"""Remove the given directory."""
self.filesystem.delete(path=path, recursive=True)
stat(self, path)
Return stat info for the given path.
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def stat(self, path: PathType) -> Dict[str, Any]:
"""Return stat info for the given path."""
return self.filesystem.stat(path=path) # type: ignore[no-any-return]
walk(self, top, topdown=True, onerror=None)
Return an iterator that walks the contents of the given directory.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
top |
Union[bytes, str] |
Path of directory to walk. |
required |
topdown |
bool |
Unused argument to conform to interface. |
True |
onerror |
Optional[Callable[..., NoneType]] |
Unused argument to conform to interface. |
None |
Returns:
| Type | Description |
|---|---|
Iterable[Tuple[Union[bytes, str], List[Union[bytes, str]], List[Union[bytes, str]]]] |
An Iterable of Tuples, each of which contain the path of the current directory path, a list of directories inside the current directory and a list of files inside the current directory. |
Source code in zenml/integrations/gcp/artifact_stores/gcp_artifact_store.py
def walk(
self,
top: PathType,
topdown: bool = True,
onerror: Optional[Callable[..., None]] = None,
) -> Iterable[Tuple[PathType, List[PathType], List[PathType]]]:
"""Return an iterator that walks the contents of the given directory.
Args:
top: Path of directory to walk.
topdown: Unused argument to conform to interface.
onerror: Unused argument to conform to interface.
Returns:
An Iterable of Tuples, each of which contain the path of the current
directory path, a list of directories inside the current directory
and a list of files inside the current directory.
"""
# TODO [ENG-153]: Additional params
return self.filesystem.walk(path=top) # type: ignore[no-any-return]
graphviz
special
GraphvizIntegration (Integration)
Definition of Graphviz integration for ZenML.
Source code in zenml/integrations/graphviz/__init__.py
class GraphvizIntegration(Integration):
"""Definition of Graphviz integration for ZenML."""
NAME = GRAPHVIZ
REQUIREMENTS = ["graphviz>=0.17"]
SYSTEM_REQUIREMENTS = {"graphviz": "dot"}
visualizers
special
pipeline_run_dag_visualizer
PipelineRunDagVisualizer (BasePipelineRunVisualizer)
Visualize the lineage of runs in a pipeline.
Source code in zenml/integrations/graphviz/visualizers/pipeline_run_dag_visualizer.py
class PipelineRunDagVisualizer(BasePipelineRunVisualizer):
"""Visualize the lineage of runs in a pipeline."""
ARTIFACT_DEFAULT_COLOR = "blue"
ARTIFACT_CACHED_COLOR = "green"
ARTIFACT_SHAPE = "box"
ARTIFACT_PREFIX = "artifact_"
STEP_COLOR = "#431D93"
STEP_SHAPE = "ellipse"
STEP_PREFIX = "step_"
FONT = "Roboto"
@abstractmethod
def visualize(
self, object: PipelineRunView, *args: Any, **kwargs: Any
) -> graphviz.Digraph:
"""Creates a pipeline lineage diagram using graphviz."""
logger.warning(
"This integration is not completed yet. Results might be unexpected."
)
dot = graphviz.Digraph(comment=object.name)
# link the steps together
for step in object.steps:
# add each step as a node
dot.node(
self.STEP_PREFIX + str(step.id),
step.entrypoint_name,
shape=self.STEP_SHAPE,
)
# for each parent of a step, add an edge
for artifact_name, artifact in step.outputs.items():
dot.node(
self.ARTIFACT_PREFIX + str(artifact.id),
f"{artifact_name} \n" f"({artifact._data_type})",
shape=self.ARTIFACT_SHAPE,
)
dot.edge(
self.STEP_PREFIX + str(step.id),
self.ARTIFACT_PREFIX + str(artifact.id),
)
for artifact_name, artifact in step.inputs.items():
dot.edge(
self.ARTIFACT_PREFIX + str(artifact.id),
self.STEP_PREFIX + str(step.id),
)
with tempfile.NamedTemporaryFile(delete=False, suffix=".html") as f:
dot.render(filename=f.name, format="png", view=True, cleanup=True)
return dot
visualize(self, object, *args, **kwargs)
Creates a pipeline lineage diagram using graphviz.
Source code in zenml/integrations/graphviz/visualizers/pipeline_run_dag_visualizer.py
@abstractmethod
def visualize(
self, object: PipelineRunView, *args: Any, **kwargs: Any
) -> graphviz.Digraph:
"""Creates a pipeline lineage diagram using graphviz."""
logger.warning(
"This integration is not completed yet. Results might be unexpected."
)
dot = graphviz.Digraph(comment=object.name)
# link the steps together
for step in object.steps:
# add each step as a node
dot.node(
self.STEP_PREFIX + str(step.id),
step.entrypoint_name,
shape=self.STEP_SHAPE,
)
# for each parent of a step, add an edge
for artifact_name, artifact in step.outputs.items():
dot.node(
self.ARTIFACT_PREFIX + str(artifact.id),
f"{artifact_name} \n" f"({artifact._data_type})",
shape=self.ARTIFACT_SHAPE,
)
dot.edge(
self.STEP_PREFIX + str(step.id),
self.ARTIFACT_PREFIX + str(artifact.id),
)
for artifact_name, artifact in step.inputs.items():
dot.edge(
self.ARTIFACT_PREFIX + str(artifact.id),
self.STEP_PREFIX + str(step.id),
)
with tempfile.NamedTemporaryFile(delete=False, suffix=".html") as f:
dot.render(filename=f.name, format="png", view=True, cleanup=True)
return dot
huggingface
special
HuggingfaceIntegration (Integration)
Definition of Huggingface integration for ZenML.
Source code in zenml/integrations/huggingface/__init__.py
class HuggingfaceIntegration(Integration):
"""Definition of Huggingface integration for ZenML."""
NAME = HUGGINGFACE
REQUIREMENTS = ["transformers", "datasets"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.huggingface import materializers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/huggingface/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.huggingface import materializers # noqa
materializers
special
huggingface_datasets_materializer
HFDatasetMaterializer (BaseMaterializer)
Materializer to read data to and from huggingface datasets.
Source code in zenml/integrations/huggingface/materializers/huggingface_datasets_materializer.py
class HFDatasetMaterializer(BaseMaterializer):
"""Materializer to read data to and from huggingface datasets."""
ASSOCIATED_TYPES = (Dataset, DatasetDict)
ASSOCIATED_ARTIFACT_TYPES = (DataArtifact,)
def handle_input(self, data_type: Type[Any]) -> Dataset:
"""Reads Dataset"""
super().handle_input(data_type)
return load_from_disk(
os.path.join(self.artifact.uri, DEFAULT_DATASET_DIR)
)
def handle_return(self, ds: Type[Any]) -> None:
"""Writes a Dataset to the specified dir.
Args:
Dataset: The Dataset to write.
"""
super().handle_return(ds)
temp_dir = TemporaryDirectory()
ds.save_to_disk(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name, os.path.join(self.artifact.uri, DEFAULT_DATASET_DIR)
)
handle_input(self, data_type)
Reads Dataset
Source code in zenml/integrations/huggingface/materializers/huggingface_datasets_materializer.py
def handle_input(self, data_type: Type[Any]) -> Dataset:
"""Reads Dataset"""
super().handle_input(data_type)
return load_from_disk(
os.path.join(self.artifact.uri, DEFAULT_DATASET_DIR)
)
handle_return(self, ds)
Writes a Dataset to the specified dir.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
Dataset |
The Dataset to write. |
required |
Source code in zenml/integrations/huggingface/materializers/huggingface_datasets_materializer.py
def handle_return(self, ds: Type[Any]) -> None:
"""Writes a Dataset to the specified dir.
Args:
Dataset: The Dataset to write.
"""
super().handle_return(ds)
temp_dir = TemporaryDirectory()
ds.save_to_disk(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name, os.path.join(self.artifact.uri, DEFAULT_DATASET_DIR)
)
huggingface_pt_model_materializer
HFPTModelMaterializer (BaseMaterializer)
Materializer to read torch model to and from huggingface pretrained model.
Source code in zenml/integrations/huggingface/materializers/huggingface_pt_model_materializer.py
class HFPTModelMaterializer(BaseMaterializer):
"""Materializer to read torch model to and from huggingface pretrained model."""
ASSOCIATED_TYPES = (PreTrainedModel,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> PreTrainedModel:
"""Reads HFModel"""
super().handle_input(data_type)
config = AutoConfig.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_PT_MODEL_DIR)
)
architecture = config.architectures[0]
model_cls = getattr(
importlib.import_module("transformers"), architecture
)
return model_cls.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_PT_MODEL_DIR)
)
def handle_return(self, model: Type[Any]) -> None:
"""Writes a Model to the specified dir.
Args:
PreTrainedModel: The Torch Model to write.
"""
super().handle_return(model)
temp_dir = TemporaryDirectory()
model.save_pretrained(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name,
os.path.join(self.artifact.uri, DEFAULT_PT_MODEL_DIR),
)
handle_input(self, data_type)
Reads HFModel
Source code in zenml/integrations/huggingface/materializers/huggingface_pt_model_materializer.py
def handle_input(self, data_type: Type[Any]) -> PreTrainedModel:
"""Reads HFModel"""
super().handle_input(data_type)
config = AutoConfig.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_PT_MODEL_DIR)
)
architecture = config.architectures[0]
model_cls = getattr(
importlib.import_module("transformers"), architecture
)
return model_cls.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_PT_MODEL_DIR)
)
handle_return(self, model)
Writes a Model to the specified dir.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
PreTrainedModel |
The Torch Model to write. |
required |
Source code in zenml/integrations/huggingface/materializers/huggingface_pt_model_materializer.py
def handle_return(self, model: Type[Any]) -> None:
"""Writes a Model to the specified dir.
Args:
PreTrainedModel: The Torch Model to write.
"""
super().handle_return(model)
temp_dir = TemporaryDirectory()
model.save_pretrained(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name,
os.path.join(self.artifact.uri, DEFAULT_PT_MODEL_DIR),
)
huggingface_tf_model_materializer
HFTFModelMaterializer (BaseMaterializer)
Materializer to read tensorflow model to and from huggingface pretrained model.
Source code in zenml/integrations/huggingface/materializers/huggingface_tf_model_materializer.py
class HFTFModelMaterializer(BaseMaterializer):
"""Materializer to read tensorflow model to and from huggingface pretrained model."""
ASSOCIATED_TYPES = (TFPreTrainedModel,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> TFPreTrainedModel:
"""Reads HFModel"""
super().handle_input(data_type)
config = AutoConfig.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_TF_MODEL_DIR)
)
architecture = "TF" + config.architectures[0]
model_cls = getattr(
importlib.import_module("transformers"), architecture
)
return model_cls.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_TF_MODEL_DIR)
)
def handle_return(self, model: Type[Any]) -> None:
"""Writes a Model to the specified dir.
Args:
TFPreTrainedModel: The TF Model to write.
"""
super().handle_return(model)
temp_dir = TemporaryDirectory()
model.save_pretrained(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name,
os.path.join(self.artifact.uri, DEFAULT_TF_MODEL_DIR),
)
handle_input(self, data_type)
Reads HFModel
Source code in zenml/integrations/huggingface/materializers/huggingface_tf_model_materializer.py
def handle_input(self, data_type: Type[Any]) -> TFPreTrainedModel:
"""Reads HFModel"""
super().handle_input(data_type)
config = AutoConfig.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_TF_MODEL_DIR)
)
architecture = "TF" + config.architectures[0]
model_cls = getattr(
importlib.import_module("transformers"), architecture
)
return model_cls.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_TF_MODEL_DIR)
)
handle_return(self, model)
Writes a Model to the specified dir.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
TFPreTrainedModel |
The TF Model to write. |
required |
Source code in zenml/integrations/huggingface/materializers/huggingface_tf_model_materializer.py
def handle_return(self, model: Type[Any]) -> None:
"""Writes a Model to the specified dir.
Args:
TFPreTrainedModel: The TF Model to write.
"""
super().handle_return(model)
temp_dir = TemporaryDirectory()
model.save_pretrained(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name,
os.path.join(self.artifact.uri, DEFAULT_TF_MODEL_DIR),
)
huggingface_tokenizer_materializer
HFTokenizerMaterializer (BaseMaterializer)
Materializer to read tokenizer to and from huggingface tokenizer.
Source code in zenml/integrations/huggingface/materializers/huggingface_tokenizer_materializer.py
class HFTokenizerMaterializer(BaseMaterializer):
"""Materializer to read tokenizer to and from huggingface tokenizer."""
ASSOCIATED_TYPES = (PreTrainedTokenizerBase,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> PreTrainedTokenizerBase:
"""Reads Tokenizer"""
super().handle_input(data_type)
return AutoTokenizer.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_TOKENIZER_DIR)
)
def handle_return(self, tokenizer: Type[Any]) -> None:
"""Writes a Tokenizer to the specified dir.
Args:
PreTrainedTokenizerBase: The HFTokenizer to write.
"""
super().handle_return(tokenizer)
temp_dir = TemporaryDirectory()
tokenizer.save_pretrained(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name,
os.path.join(self.artifact.uri, DEFAULT_TOKENIZER_DIR),
)
handle_input(self, data_type)
Reads Tokenizer
Source code in zenml/integrations/huggingface/materializers/huggingface_tokenizer_materializer.py
def handle_input(self, data_type: Type[Any]) -> PreTrainedTokenizerBase:
"""Reads Tokenizer"""
super().handle_input(data_type)
return AutoTokenizer.from_pretrained(
os.path.join(self.artifact.uri, DEFAULT_TOKENIZER_DIR)
)
handle_return(self, tokenizer)
Writes a Tokenizer to the specified dir.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
PreTrainedTokenizerBase |
The HFTokenizer to write. |
required |
Source code in zenml/integrations/huggingface/materializers/huggingface_tokenizer_materializer.py
def handle_return(self, tokenizer: Type[Any]) -> None:
"""Writes a Tokenizer to the specified dir.
Args:
PreTrainedTokenizerBase: The HFTokenizer to write.
"""
super().handle_return(tokenizer)
temp_dir = TemporaryDirectory()
tokenizer.save_pretrained(temp_dir.name)
fileio_utils.copy_dir(
temp_dir.name,
os.path.join(self.artifact.uri, DEFAULT_TOKENIZER_DIR),
)
integration
Integration
Base class for integration in ZenML
Source code in zenml/integrations/integration.py
class Integration(metaclass=IntegrationMeta):
"""Base class for integration in ZenML"""
NAME = "base_integration"
REQUIREMENTS: List[str] = []
SYSTEM_REQUIREMENTS: Dict[str, str] = {}
@classmethod
def check_installation(cls) -> bool:
"""Method to check whether the required packages are installed"""
try:
for requirement, command in cls.SYSTEM_REQUIREMENTS.items():
result = shutil.which(command)
if result is None:
logger.debug(
"Unable to find the required packages for %s on your "
"system. Please install the packages on your system "
"and try again.",
requirement,
)
return False
for r in cls.REQUIREMENTS:
pkg_resources.get_distribution(r)
logger.debug(
f"Integration {cls.NAME} is installed correctly with "
f"requirements {cls.REQUIREMENTS}."
)
return True
except pkg_resources.DistributionNotFound as e:
logger.debug(
f"Unable to find required package '{e.req}' for "
f"integration {cls.NAME}."
)
return False
except pkg_resources.VersionConflict as e:
logger.debug(
f"VersionConflict error when loading installation {cls.NAME}: "
f"{str(e)}"
)
return False
@staticmethod
def activate() -> None:
"""Abstract method to activate the integration"""
activate()
staticmethod
Abstract method to activate the integration
Source code in zenml/integrations/integration.py
@staticmethod
def activate() -> None:
"""Abstract method to activate the integration"""
check_installation()
classmethod
Method to check whether the required packages are installed
Source code in zenml/integrations/integration.py
@classmethod
def check_installation(cls) -> bool:
"""Method to check whether the required packages are installed"""
try:
for requirement, command in cls.SYSTEM_REQUIREMENTS.items():
result = shutil.which(command)
if result is None:
logger.debug(
"Unable to find the required packages for %s on your "
"system. Please install the packages on your system "
"and try again.",
requirement,
)
return False
for r in cls.REQUIREMENTS:
pkg_resources.get_distribution(r)
logger.debug(
f"Integration {cls.NAME} is installed correctly with "
f"requirements {cls.REQUIREMENTS}."
)
return True
except pkg_resources.DistributionNotFound as e:
logger.debug(
f"Unable to find required package '{e.req}' for "
f"integration {cls.NAME}."
)
return False
except pkg_resources.VersionConflict as e:
logger.debug(
f"VersionConflict error when loading installation {cls.NAME}: "
f"{str(e)}"
)
return False
IntegrationMeta (type)
Metaclass responsible for registering different Integration subclasses
Source code in zenml/integrations/integration.py
class IntegrationMeta(type):
"""Metaclass responsible for registering different Integration
subclasses"""
def __new__(
mcs, name: str, bases: Tuple[Type[Any], ...], dct: Dict[str, Any]
) -> "IntegrationMeta":
"""Hook into creation of an Integration class."""
cls = cast(Type["Integration"], super().__new__(mcs, name, bases, dct))
if name != "Integration":
integration_registry.register_integration(cls.NAME, cls)
return cls
__new__(mcs, name, bases, dct)
special
staticmethod
Hook into creation of an Integration class.
Source code in zenml/integrations/integration.py
def __new__(
mcs, name: str, bases: Tuple[Type[Any], ...], dct: Dict[str, Any]
) -> "IntegrationMeta":
"""Hook into creation of an Integration class."""
cls = cast(Type["Integration"], super().__new__(mcs, name, bases, dct))
if name != "Integration":
integration_registry.register_integration(cls.NAME, cls)
return cls
kubeflow
special
The Kubeflow integration sub-module powers an alternative to the local orchestrator. You can enable it by registering the Kubeflow orchestrator with the CLI tool.
KubeflowIntegration (Integration)
Definition of Kubeflow Integration for ZenML.
Source code in zenml/integrations/kubeflow/__init__.py
class KubeflowIntegration(Integration):
"""Definition of Kubeflow Integration for ZenML."""
NAME = KUBEFLOW
REQUIREMENTS = ["kfp==1.8.9"]
@classmethod
def activate(cls) -> None:
"""Activates all classes required for the airflow integration."""
from zenml.integrations.kubeflow import metadata_stores # noqa
from zenml.integrations.kubeflow import orchestrators # noqa
activate()
classmethod
Activates all classes required for the airflow integration.
Source code in zenml/integrations/kubeflow/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates all classes required for the airflow integration."""
from zenml.integrations.kubeflow import metadata_stores # noqa
from zenml.integrations.kubeflow import orchestrators # noqa
container_entrypoint
Main entrypoint for containers with Kubeflow TFX component executors.
main()
Runs a single step defined by the command line arguments.
Source code in zenml/integrations/kubeflow/container_entrypoint.py
def main() -> None:
"""Runs a single step defined by the command line arguments."""
# Log to the container's stdout so Kubeflow Pipelines UI can display logs to
# the user.
logging.basicConfig(stream=sys.stdout, level=logging.INFO)
logging.getLogger().setLevel(logging.INFO)
args = _parse_command_line_arguments()
tfx_pipeline = pipeline_pb2.Pipeline()
json_format.Parse(args.tfx_ir, tfx_pipeline)
run_name = _get_run_name()
_resolve_runtime_parameters(tfx_pipeline, run_name, args.runtime_parameter)
node_id = args.node_id
pipeline_node = _get_pipeline_node(tfx_pipeline, node_id)
deployment_config = runner_utils.extract_local_deployment_config(
tfx_pipeline
)
executor_spec = runner_utils.extract_executor_spec(
deployment_config, node_id
)
custom_driver_spec = runner_utils.extract_custom_driver_spec(
deployment_config, node_id
)
# make sure all integrations are activated so all materializers etc. are
# available
integration_registry.activate_integrations()
repo = Repository()
metadata_store = repo.active_stack.metadata_store
metadata_connection = metadata.Metadata(
metadata_store.get_tfx_metadata_config()
)
# import the user main module to register all the materializers
importlib.import_module(args.main_module)
zenml.constants.USER_MAIN_MODULE = args.main_module
step_module = importlib.import_module(args.step_module)
step_class = getattr(step_module, args.step_function_name)
step_instance = cast(BaseStep, step_class())
if hasattr(executor_spec, "class_path"):
executor_module_parts = getattr(executor_spec, "class_path").split(".")
executor_class_target_module_name = ".".join(executor_module_parts[:-1])
_create_executor_class(
step=step_instance,
executor_class_target_module_name=executor_class_target_module_name,
input_artifact_type_mapping=json.loads(args.input_artifact_types),
)
else:
raise RuntimeError(
f"No class path found inside executor spec: {executor_spec}."
)
custom_executor_operators = {
executable_spec_pb2.PythonClassExecutableSpec: step_instance.executor_operator
}
component_launcher = launcher.Launcher(
pipeline_node=pipeline_node,
mlmd_connection=metadata_connection,
pipeline_info=tfx_pipeline.pipeline_info,
pipeline_runtime_spec=tfx_pipeline.runtime_spec,
executor_spec=executor_spec,
custom_driver_spec=custom_driver_spec,
custom_executor_operators=custom_executor_operators,
)
repo.active_stack.prepare_step_run()
execution_info = execute_step(component_launcher)
repo.active_stack.prepare_step_run()
if execution_info:
_dump_ui_metadata(pipeline_node, execution_info, args.metadata_ui_path)
metadata_stores
special
kubeflow_metadata_store
KubeflowMetadataStore (BaseMetadataStore)
pydantic-model
Kubeflow GRPC backend for ZenML metadata store.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
class KubeflowMetadataStore(BaseMetadataStore):
"""Kubeflow GRPC backend for ZenML metadata store."""
upgrade_migration_enabled: bool = False
host: str = "127.0.0.1"
port: int = DEFAULT_KFP_METADATA_GRPC_PORT
# Class Configuration
FLAVOR: ClassVar[str] = KUBEFLOW
@property
def validator(self) -> Optional[StackValidator]:
"""Validates that the stack contains a KFP orchestrator."""
def _ensure_kfp_orchestrator(stack: Stack) -> Tuple[bool, str]:
return (
stack.orchestrator.FLAVOR == KUBEFLOW,
"The Kubeflow metadata store can only be used with a Kubeflow "
"orchestrator.",
)
return StackValidator(
custom_validation_function=_ensure_kfp_orchestrator
)
def get_tfx_metadata_config(
self,
) -> Union[
metadata_store_pb2.ConnectionConfig,
metadata_store_pb2.MetadataStoreClientConfig,
]:
"""Return tfx metadata config for the kubeflow metadata store."""
connection_config = metadata_store_pb2.MetadataStoreClientConfig()
if inside_kfp_pod():
connection_config.host = os.environ["METADATA_GRPC_SERVICE_HOST"]
connection_config.port = int(
os.environ["METADATA_GRPC_SERVICE_PORT"]
)
else:
if not self.is_running:
raise RuntimeError(
"The KFP metadata daemon is not running. Please run the "
"following command to start it first:\n\n"
" 'zenml metadata-store up'\n"
)
connection_config.host = self.host
connection_config.port = self.port
return connection_config
@property
def kfp_orchestrator(self) -> KubeflowOrchestrator:
"""Returns the Kubeflow orchestrator in the active stack."""
repo = Repository(skip_repository_check=True) # type: ignore[call-arg]
return cast(KubeflowOrchestrator, repo.active_stack.orchestrator)
@property
def kubernetes_context(self) -> str:
"""Returns the kubernetes context to the cluster where the Kubeflow
Pipelines services are running."""
kubernetes_context = self.kfp_orchestrator.kubernetes_context
# will never happen, but mypy doesn't know that
assert kubernetes_context is not None
return kubernetes_context
@property
def root_directory(self) -> str:
"""Returns path to the root directory for all files concerning
this KFP metadata store.
Note: the root directory for the KFP metadata store is relative to the
root directory of the KFP orchestrator, because it is a sub-component
of it.
"""
return os.path.join(
self.kfp_orchestrator.root_directory,
"metadata-store",
str(self.uuid),
)
@property
def _pid_file_path(self) -> str:
"""Returns path to the daemon PID file."""
return os.path.join(self.root_directory, "kubeflow_daemon.pid")
@property
def _log_file(self) -> str:
"""Path of the daemon log file."""
return os.path.join(self.root_directory, "kubeflow_daemon.log")
@property
def is_provisioned(self) -> bool:
"""If the component provisioned resources to run locally."""
return fileio.exists(self.root_directory)
@property
def is_running(self) -> bool:
"""If the component is running locally."""
if sys.platform != "win32":
from zenml.utils.daemon import check_if_daemon_is_running
if not check_if_daemon_is_running(self._pid_file_path):
return False
else:
# Daemon functionality is not supported on Windows, so the PID
# file won't exist. This if clause exists just for mypy to not
# complain about missing functions
pass
return True
def provision(self) -> None:
"""Provisions resources to run the component locally."""
logger.info("Provisioning local Kubeflow Pipelines deployment...")
fileio.makedirs(self.root_directory)
def deprovision(self) -> None:
"""Deprovisions all local resources of the component."""
if fileio.exists(self._log_file):
fileio.remove(self._log_file)
logger.info("Local kubeflow pipelines deployment deprovisioned.")
def resume(self) -> None:
"""Resumes the local k3d cluster."""
if self.is_running:
logger.info("Local kubeflow pipelines deployment already running.")
return
self.start_kfp_metadata_daemon()
self.wait_until_metadata_store_ready()
def suspend(self) -> None:
"""Suspends the local k3d cluster."""
if not self.is_running:
logger.info("Local kubeflow pipelines deployment not running.")
return
self.stop_kfp_metadata_daemon()
def start_kfp_metadata_daemon(self) -> None:
"""Starts a daemon process that forwards ports so the Kubeflow Pipelines
Metadata MySQL database is accessible on the localhost."""
command = [
"kubectl",
"--context",
self.kubernetes_context,
"--namespace",
"kubeflow",
"port-forward",
"svc/metadata-grpc-service",
f"{self.port}:8080",
]
if sys.platform == "win32":
logger.warning(
"Daemon functionality not supported on Windows. "
"In order to access the Kubeflow Pipelines Metadata locally, "
"please run '%s' in a separate command line shell.",
self.port,
" ".join(command),
)
elif not networking_utils.port_available(self.port):
raise ProvisioningError(
f"Unable to port-forward Kubeflow Pipelines Metadata to local "
f"port {self.port} because the port is occupied. In order to "
f"access the Kubeflow Pipelines Metadata locally, please "
f"change the metadata store configuration to use an available "
f"port or stop the other process currently using the port."
)
else:
from zenml.utils import daemon
def _daemon_function() -> None:
"""Forwards the port of the Kubeflow Pipelines Metadata pod ."""
subprocess.check_call(command)
daemon.run_as_daemon(
_daemon_function,
pid_file=self._pid_file_path,
log_file=self._log_file,
)
logger.info(
"Started Kubeflow Pipelines Metadata daemon (check the daemon"
"logs at %s in case you're not able to access the pipeline"
"metadata).",
self._log_file,
)
def stop_kfp_metadata_daemon(self) -> None:
"""Stops the KFP Metadata daemon process if it is running."""
if fileio.exists(self._pid_file_path):
if sys.platform == "win32":
# Daemon functionality is not supported on Windows, so the PID
# file won't exist. This if clause exists just for mypy to not
# complain about missing functions
pass
else:
from zenml.utils import daemon
daemon.stop_daemon(self._pid_file_path)
fileio.remove(self._pid_file_path)
def wait_until_metadata_store_ready(
self, timeout: int = DEFAULT_KFP_METADATA_DAEMON_TIMEOUT
) -> None:
"""Waits until the metadata store connection is ready, an irrecoverable
error occurs or the timeout expires."""
logger.info(
"Waiting for the Kubeflow metadata store to be ready (this might "
"take a few minutes)."
)
while True:
try:
# it doesn't matter what we call here as long as it exercises
# the MLMD connection
self.get_pipelines()
break
except Exception as e:
logger.info(
"The Kubeflow metadata store is not ready yet. Waiting for "
"10 seconds..."
)
if timeout <= 0:
raise RuntimeError(
f"An unexpected error was encountered while waiting for the "
f"Kubeflow metadata store to be functional: {str(e)}"
) from e
timeout -= 10
time.sleep(10)
logger.info("The Kubeflow metadata store is functional.")
is_provisioned: bool
property
readonly
If the component provisioned resources to run locally.
is_running: bool
property
readonly
If the component is running locally.
kfp_orchestrator: KubeflowOrchestrator
property
readonly
Returns the Kubeflow orchestrator in the active stack.
kubernetes_context: str
property
readonly
Returns the kubernetes context to the cluster where the Kubeflow Pipelines services are running.
root_directory: str
property
readonly
Returns path to the root directory for all files concerning this KFP metadata store.
Note: the root directory for the KFP metadata store is relative to the root directory of the KFP orchestrator, because it is a sub-component of it.
validator: Optional[zenml.stack.stack_validator.StackValidator]
property
readonly
Validates that the stack contains a KFP orchestrator.
deprovision(self)
Deprovisions all local resources of the component.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def deprovision(self) -> None:
"""Deprovisions all local resources of the component."""
if fileio.exists(self._log_file):
fileio.remove(self._log_file)
logger.info("Local kubeflow pipelines deployment deprovisioned.")
get_tfx_metadata_config(self)
Return tfx metadata config for the kubeflow metadata store.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def get_tfx_metadata_config(
self,
) -> Union[
metadata_store_pb2.ConnectionConfig,
metadata_store_pb2.MetadataStoreClientConfig,
]:
"""Return tfx metadata config for the kubeflow metadata store."""
connection_config = metadata_store_pb2.MetadataStoreClientConfig()
if inside_kfp_pod():
connection_config.host = os.environ["METADATA_GRPC_SERVICE_HOST"]
connection_config.port = int(
os.environ["METADATA_GRPC_SERVICE_PORT"]
)
else:
if not self.is_running:
raise RuntimeError(
"The KFP metadata daemon is not running. Please run the "
"following command to start it first:\n\n"
" 'zenml metadata-store up'\n"
)
connection_config.host = self.host
connection_config.port = self.port
return connection_config
provision(self)
Provisions resources to run the component locally.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def provision(self) -> None:
"""Provisions resources to run the component locally."""
logger.info("Provisioning local Kubeflow Pipelines deployment...")
fileio.makedirs(self.root_directory)
resume(self)
Resumes the local k3d cluster.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def resume(self) -> None:
"""Resumes the local k3d cluster."""
if self.is_running:
logger.info("Local kubeflow pipelines deployment already running.")
return
self.start_kfp_metadata_daemon()
self.wait_until_metadata_store_ready()
start_kfp_metadata_daemon(self)
Starts a daemon process that forwards ports so the Kubeflow Pipelines Metadata MySQL database is accessible on the localhost.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def start_kfp_metadata_daemon(self) -> None:
"""Starts a daemon process that forwards ports so the Kubeflow Pipelines
Metadata MySQL database is accessible on the localhost."""
command = [
"kubectl",
"--context",
self.kubernetes_context,
"--namespace",
"kubeflow",
"port-forward",
"svc/metadata-grpc-service",
f"{self.port}:8080",
]
if sys.platform == "win32":
logger.warning(
"Daemon functionality not supported on Windows. "
"In order to access the Kubeflow Pipelines Metadata locally, "
"please run '%s' in a separate command line shell.",
self.port,
" ".join(command),
)
elif not networking_utils.port_available(self.port):
raise ProvisioningError(
f"Unable to port-forward Kubeflow Pipelines Metadata to local "
f"port {self.port} because the port is occupied. In order to "
f"access the Kubeflow Pipelines Metadata locally, please "
f"change the metadata store configuration to use an available "
f"port or stop the other process currently using the port."
)
else:
from zenml.utils import daemon
def _daemon_function() -> None:
"""Forwards the port of the Kubeflow Pipelines Metadata pod ."""
subprocess.check_call(command)
daemon.run_as_daemon(
_daemon_function,
pid_file=self._pid_file_path,
log_file=self._log_file,
)
logger.info(
"Started Kubeflow Pipelines Metadata daemon (check the daemon"
"logs at %s in case you're not able to access the pipeline"
"metadata).",
self._log_file,
)
stop_kfp_metadata_daemon(self)
Stops the KFP Metadata daemon process if it is running.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def stop_kfp_metadata_daemon(self) -> None:
"""Stops the KFP Metadata daemon process if it is running."""
if fileio.exists(self._pid_file_path):
if sys.platform == "win32":
# Daemon functionality is not supported on Windows, so the PID
# file won't exist. This if clause exists just for mypy to not
# complain about missing functions
pass
else:
from zenml.utils import daemon
daemon.stop_daemon(self._pid_file_path)
fileio.remove(self._pid_file_path)
suspend(self)
Suspends the local k3d cluster.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def suspend(self) -> None:
"""Suspends the local k3d cluster."""
if not self.is_running:
logger.info("Local kubeflow pipelines deployment not running.")
return
self.stop_kfp_metadata_daemon()
wait_until_metadata_store_ready(self, timeout=60)
Waits until the metadata store connection is ready, an irrecoverable error occurs or the timeout expires.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def wait_until_metadata_store_ready(
self, timeout: int = DEFAULT_KFP_METADATA_DAEMON_TIMEOUT
) -> None:
"""Waits until the metadata store connection is ready, an irrecoverable
error occurs or the timeout expires."""
logger.info(
"Waiting for the Kubeflow metadata store to be ready (this might "
"take a few minutes)."
)
while True:
try:
# it doesn't matter what we call here as long as it exercises
# the MLMD connection
self.get_pipelines()
break
except Exception as e:
logger.info(
"The Kubeflow metadata store is not ready yet. Waiting for "
"10 seconds..."
)
if timeout <= 0:
raise RuntimeError(
f"An unexpected error was encountered while waiting for the "
f"Kubeflow metadata store to be functional: {str(e)}"
) from e
timeout -= 10
time.sleep(10)
logger.info("The Kubeflow metadata store is functional.")
inside_kfp_pod()
Returns if the current python process is running inside a KFP Pod.
Source code in zenml/integrations/kubeflow/metadata_stores/kubeflow_metadata_store.py
def inside_kfp_pod() -> bool:
"""Returns if the current python process is running inside a KFP Pod."""
if "KFP_POD_NAME" not in os.environ:
return False
try:
k8s_config.load_incluster_config()
return True
except k8s_config.ConfigException:
return False
orchestrators
special
kubeflow_component
Kubeflow Pipelines based implementation of TFX components. These components are lightweight wrappers around the KFP DSL's ContainerOp, and ensure that the container gets called with the right set of input arguments. It also ensures that each component exports named output attributes that are consistent with those provided by the native TFX components, thus ensuring that both types of pipeline definitions are compatible. Note: This requires Kubeflow Pipelines SDK to be installed.
KubeflowComponent
Base component for all Kubeflow pipelines TFX components. Returns a wrapper around a KFP DSL ContainerOp class, and adds named output attributes that match the output names for the corresponding native TFX components.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_component.py
class KubeflowComponent:
"""Base component for all Kubeflow pipelines TFX components.
Returns a wrapper around a KFP DSL ContainerOp class, and adds named output
attributes that match the output names for the corresponding native TFX
components.
"""
def __init__(
self,
component: tfx_base_component.BaseComponent,
depends_on: Set[dsl.ContainerOp],
image: str,
tfx_ir: pipeline_pb2.Pipeline,
pod_labels_to_attach: Dict[str, str],
main_module: str,
step_module: str,
step_function_name: str,
runtime_parameters: List[data_types.RuntimeParameter],
):
"""Creates a new Kubeflow-based component.
This class essentially wraps a dsl.ContainerOp construct in Kubeflow
Pipelines.
Args:
component: The logical TFX component to wrap.
depends_on: The set of upstream KFP ContainerOp components that this
component will depend on.
image: The container image to use for this component.
tfx_ir: The TFX intermedia representation of the pipeline.
pod_labels_to_attach: Dict of pod labels to attach to the GKE pod.
runtime_parameters: Runtime parameters of the pipeline.
"""
# Path to a metadata file that will be displayed in the KFP UI
# This metadata file needs to be in a mounted emptyDir to avoid
# sporadic failures with the (not mature) PNS executor
# See these links for more information about limitations of PNS +
# security context:
# https://www.kubeflow.org/docs/components/pipelines/installation/localcluster-deployment/#deploying-kubeflow-pipelines
# https://argoproj.github.io/argo-workflows/empty-dir/
# KFP will switch to the Emissary executor (soon), when this emptyDir
# mount will not be necessary anymore, but for now it's still in alpha
# status (https://www.kubeflow.org/docs/components/pipelines/installation/choose-executor/#emissary-executor)
metadata_ui_path = "/outputs/mlpipeline-ui-metadata.json"
volumes: Dict[str, k8s_client.V1Volume] = {
"/outputs": k8s_client.V1Volume(
name="outputs", empty_dir=k8s_client.V1EmptyDirVolumeSource()
),
}
utils.replace_placeholder(component)
input_artifact_type_mapping = _get_input_artifact_type_mapping(
component
)
arguments = [
"--node_id",
component.id,
"--tfx_ir",
json_format.MessageToJson(tfx_ir),
"--metadata_ui_path",
metadata_ui_path,
"--main_module",
main_module,
"--step_module",
step_module,
"--step_function_name",
step_function_name,
"--input_artifact_types",
json.dumps(input_artifact_type_mapping),
]
for param in runtime_parameters:
arguments.append("--runtime_parameter")
arguments.append(_encode_runtime_parameter(param))
stack = Repository().active_stack
global_cfg_dir = get_global_config_directory()
# go through all stack components and identify those that advertise
# a local path where they persist information that they need to be
# available when running pipelines. For those that do, mount them
# into the Kubeflow container.
has_local_repos = False
for stack_comp in stack.components.values():
local_path = stack_comp.local_path
if not local_path:
continue
# double-check this convention, just in case it wasn't respected
# as documented in `StackComponent.local_path`
if not local_path.startswith(global_cfg_dir):
raise ValueError(
f"Local path {local_path} for component {stack_comp.name} "
f"is not in the global config directory ({global_cfg_dir})."
)
has_local_repos = True
host_path = k8s_client.V1HostPathVolumeSource(
path=local_path, type="Directory"
)
volume_name = f"{stack_comp.TYPE.value}-{stack_comp.name}"
volumes[local_path] = k8s_client.V1Volume(
name=re.sub(r"[^0-9a-zA-Z-]+", "-", volume_name)
.strip("-")
.lower(),
host_path=host_path,
)
logger.debug(
"Adding host path volume for %s %s (path: %s) "
"in kubeflow pipelines container.",
stack_comp.TYPE.value,
stack_comp.name,
local_path,
)
self.container_op = dsl.ContainerOp(
name=component.id,
command=CONTAINER_ENTRYPOINT_COMMAND,
image=image,
arguments=arguments,
output_artifact_paths={
"mlpipeline-ui-metadata": metadata_ui_path,
},
pvolumes=volumes,
)
if has_local_repos:
if sys.platform == "win32":
# File permissions are not checked on Windows. This if clause
# prevents mypy from complaining about unused 'type: ignore'
# statements
pass
else:
# Run KFP containers in the context of the local UID/GID
# to ensure that the artifact and metadata stores can be shared
# with the local pipeline runs.
self.container_op.container.security_context = (
k8s_client.V1SecurityContext(
run_as_user=os.getuid(),
run_as_group=os.getgid(),
)
)
logger.debug(
"Setting security context UID and GID to local user/group "
"in kubeflow pipelines container."
)
for op in depends_on:
self.container_op.after(op)
self.container_op.container.add_env_variable(
k8s_client.V1EnvVar(
name=ENV_ZENML_PREVENT_PIPELINE_EXECUTION, value="True"
)
)
# Add environment variables for Azure Blob Storage to pod in case they
# are set locally
# TODO [ENG-699]: remove this as soon as we implement credential handling
for key in [
"AZURE_STORAGE_ACCOUNT_KEY",
"AZURE_STORAGE_ACCOUNT_NAME",
"AZURE_STORAGE_CONNECTION_STRING",
"AZURE_STORAGE_SAS_TOKEN",
]:
value = os.getenv(key)
if value:
self.container_op.container.add_env_variable(
k8s_client.V1EnvVar(name=key, value=value)
)
for k, v in pod_labels_to_attach.items():
self.container_op.add_pod_label(k, v)
__init__(self, component, depends_on, image, tfx_ir, pod_labels_to_attach, main_module, step_module, step_function_name, runtime_parameters)
special
Creates a new Kubeflow-based component. This class essentially wraps a dsl.ContainerOp construct in Kubeflow Pipelines.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
component |
BaseComponent |
The logical TFX component to wrap. |
required |
depends_on |
Set[kfp.dsl._container_op.ContainerOp] |
The set of upstream KFP ContainerOp components that this component will depend on. |
required |
image |
str |
The container image to use for this component. |
required |
tfx_ir |
Pipeline |
The TFX intermedia representation of the pipeline. |
required |
pod_labels_to_attach |
Dict[str, str] |
Dict of pod labels to attach to the GKE pod. |
required |
runtime_parameters |
List[tfx.orchestration.data_types.RuntimeParameter] |
Runtime parameters of the pipeline. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_component.py
def __init__(
self,
component: tfx_base_component.BaseComponent,
depends_on: Set[dsl.ContainerOp],
image: str,
tfx_ir: pipeline_pb2.Pipeline,
pod_labels_to_attach: Dict[str, str],
main_module: str,
step_module: str,
step_function_name: str,
runtime_parameters: List[data_types.RuntimeParameter],
):
"""Creates a new Kubeflow-based component.
This class essentially wraps a dsl.ContainerOp construct in Kubeflow
Pipelines.
Args:
component: The logical TFX component to wrap.
depends_on: The set of upstream KFP ContainerOp components that this
component will depend on.
image: The container image to use for this component.
tfx_ir: The TFX intermedia representation of the pipeline.
pod_labels_to_attach: Dict of pod labels to attach to the GKE pod.
runtime_parameters: Runtime parameters of the pipeline.
"""
# Path to a metadata file that will be displayed in the KFP UI
# This metadata file needs to be in a mounted emptyDir to avoid
# sporadic failures with the (not mature) PNS executor
# See these links for more information about limitations of PNS +
# security context:
# https://www.kubeflow.org/docs/components/pipelines/installation/localcluster-deployment/#deploying-kubeflow-pipelines
# https://argoproj.github.io/argo-workflows/empty-dir/
# KFP will switch to the Emissary executor (soon), when this emptyDir
# mount will not be necessary anymore, but for now it's still in alpha
# status (https://www.kubeflow.org/docs/components/pipelines/installation/choose-executor/#emissary-executor)
metadata_ui_path = "/outputs/mlpipeline-ui-metadata.json"
volumes: Dict[str, k8s_client.V1Volume] = {
"/outputs": k8s_client.V1Volume(
name="outputs", empty_dir=k8s_client.V1EmptyDirVolumeSource()
),
}
utils.replace_placeholder(component)
input_artifact_type_mapping = _get_input_artifact_type_mapping(
component
)
arguments = [
"--node_id",
component.id,
"--tfx_ir",
json_format.MessageToJson(tfx_ir),
"--metadata_ui_path",
metadata_ui_path,
"--main_module",
main_module,
"--step_module",
step_module,
"--step_function_name",
step_function_name,
"--input_artifact_types",
json.dumps(input_artifact_type_mapping),
]
for param in runtime_parameters:
arguments.append("--runtime_parameter")
arguments.append(_encode_runtime_parameter(param))
stack = Repository().active_stack
global_cfg_dir = get_global_config_directory()
# go through all stack components and identify those that advertise
# a local path where they persist information that they need to be
# available when running pipelines. For those that do, mount them
# into the Kubeflow container.
has_local_repos = False
for stack_comp in stack.components.values():
local_path = stack_comp.local_path
if not local_path:
continue
# double-check this convention, just in case it wasn't respected
# as documented in `StackComponent.local_path`
if not local_path.startswith(global_cfg_dir):
raise ValueError(
f"Local path {local_path} for component {stack_comp.name} "
f"is not in the global config directory ({global_cfg_dir})."
)
has_local_repos = True
host_path = k8s_client.V1HostPathVolumeSource(
path=local_path, type="Directory"
)
volume_name = f"{stack_comp.TYPE.value}-{stack_comp.name}"
volumes[local_path] = k8s_client.V1Volume(
name=re.sub(r"[^0-9a-zA-Z-]+", "-", volume_name)
.strip("-")
.lower(),
host_path=host_path,
)
logger.debug(
"Adding host path volume for %s %s (path: %s) "
"in kubeflow pipelines container.",
stack_comp.TYPE.value,
stack_comp.name,
local_path,
)
self.container_op = dsl.ContainerOp(
name=component.id,
command=CONTAINER_ENTRYPOINT_COMMAND,
image=image,
arguments=arguments,
output_artifact_paths={
"mlpipeline-ui-metadata": metadata_ui_path,
},
pvolumes=volumes,
)
if has_local_repos:
if sys.platform == "win32":
# File permissions are not checked on Windows. This if clause
# prevents mypy from complaining about unused 'type: ignore'
# statements
pass
else:
# Run KFP containers in the context of the local UID/GID
# to ensure that the artifact and metadata stores can be shared
# with the local pipeline runs.
self.container_op.container.security_context = (
k8s_client.V1SecurityContext(
run_as_user=os.getuid(),
run_as_group=os.getgid(),
)
)
logger.debug(
"Setting security context UID and GID to local user/group "
"in kubeflow pipelines container."
)
for op in depends_on:
self.container_op.after(op)
self.container_op.container.add_env_variable(
k8s_client.V1EnvVar(
name=ENV_ZENML_PREVENT_PIPELINE_EXECUTION, value="True"
)
)
# Add environment variables for Azure Blob Storage to pod in case they
# are set locally
# TODO [ENG-699]: remove this as soon as we implement credential handling
for key in [
"AZURE_STORAGE_ACCOUNT_KEY",
"AZURE_STORAGE_ACCOUNT_NAME",
"AZURE_STORAGE_CONNECTION_STRING",
"AZURE_STORAGE_SAS_TOKEN",
]:
value = os.getenv(key)
if value:
self.container_op.container.add_env_variable(
k8s_client.V1EnvVar(name=key, value=value)
)
for k, v in pod_labels_to_attach.items():
self.container_op.add_pod_label(k, v)
kubeflow_dag_runner
The below code is copied from the TFX source repo with minor changes. All credits go to the TFX team for the core implementation
KubeflowDagRunner
Kubeflow Pipelines runner. Constructs a pipeline definition YAML file based on the TFX logical pipeline.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_dag_runner.py
class KubeflowDagRunner:
"""Kubeflow Pipelines runner. Constructs a pipeline definition YAML file
based on the TFX logical pipeline.
"""
def __init__(
self,
config: KubeflowDagRunnerConfig,
output_path: str,
pod_labels_to_attach: Optional[Dict[str, str]] = None,
):
"""Initializes KubeflowDagRunner for compiling a Kubeflow Pipeline.
Args:
config: A KubeflowDagRunnerConfig object to specify runtime
configuration when running the pipeline under Kubeflow.
output_path: Path where the pipeline definition file will be stored.
pod_labels_to_attach: Optional set of pod labels to attach to GKE pod
spun up for this pipeline. Default to the 3 labels:
1. add-pod-env: true,
2. pipeline SDK type,
3. pipeline unique ID,
where 2 and 3 are instrumentation of usage tracking.
"""
self._config = config or pipeline_config.PipelineConfig()
self._kubeflow_config = config
self._output_path = output_path
self._compiler = compiler.Compiler()
self._tfx_compiler = tfx_compiler.Compiler()
self._params: List[dsl.PipelineParam] = []
self._params_by_component_id: Dict[
str, List[data_types.RuntimeParameter]
] = collections.defaultdict(list)
self._deduped_parameter_names: Set[str] = set()
self._pod_labels_to_attach = (
pod_labels_to_attach or get_default_pod_labels()
)
@property
def config(self) -> pipeline_config.PipelineConfig:
"""The config property"""
return self._config
def _parse_parameter_from_component(
self, component: tfx_base_component.BaseComponent
) -> None:
"""Extract embedded RuntimeParameter placeholders from a component.
Extract embedded RuntimeParameter placeholders from a component, then
append the corresponding dsl.PipelineParam to KubeflowDagRunner.
Args:
component: a TFX component.
"""
deduped_parameter_names_for_component = set()
for parameter in component.exec_properties.values():
if not isinstance(parameter, data_types.RuntimeParameter):
continue
# Ignore pipeline root because it will be added later.
if parameter.name == TFX_PIPELINE_ROOT_PARAMETER.name:
continue
if parameter.name in deduped_parameter_names_for_component:
continue
deduped_parameter_names_for_component.add(parameter.name)
self._params_by_component_id[component.id].append(parameter)
if parameter.name not in self._deduped_parameter_names:
self._deduped_parameter_names.add(parameter.name)
dsl_parameter = dsl.PipelineParam(
name=parameter.name, value=str(parameter.default)
)
self._params.append(dsl_parameter)
def _parse_parameter_from_pipeline(self, pipeline: TfxPipeline) -> None:
"""Extract all the RuntimeParameter placeholders from the pipeline."""
for component in pipeline.components:
self._parse_parameter_from_component(component)
def _construct_pipeline_graph(
self,
pipeline: "BasePipeline",
tfx_pipeline: TfxPipeline,
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> None:
"""Constructs a Kubeflow Pipeline graph.
Args:
pipeline: ZenML pipeline instance.
tfx_pipeline: The logical TFX pipeline to base the construction on.
stack: The ZenML stack that the pipeline is running on
runtime_configuration: The runtime configuration
"""
component_to_kfp_op: Dict[base_node.BaseNode, dsl.ContainerOp] = {}
tfx_ir: Pb2Pipeline = self._generate_tfx_ir(tfx_pipeline)
for node in tfx_ir.nodes:
pipeline_node: PipelineNode = node.pipeline_node
# Add the stack as context to each pipeline node:
context_utils.add_context_to_node(
pipeline_node,
type_=MetadataContextTypes.STACK.value,
name=str(hash(json.dumps(stack.dict(), sort_keys=True))),
properties=stack.dict(),
)
# Add all pydantic objects from runtime_configuration to the
# context
context_utils.add_runtime_configuration_to_node(
pipeline_node, runtime_configuration
)
# Add pipeline requirements as a context
requirements = " ".join(sorted(pipeline.requirements))
context_utils.add_context_to_node(
pipeline_node,
type_=MetadataContextTypes.PIPELINE_REQUIREMENTS.value,
name=str(hash(requirements)),
properties={"pipeline_requirements": requirements},
)
# Assumption: There is a partial ordering of components in the list,
# i.e. if component A depends on component B and C, then A appears
# after B and C in the list.
for component in tfx_pipeline.components:
# Keep track of the set of upstream dsl.ContainerOps for this
# component.
depends_on = set()
for upstream_component in component.upstream_nodes:
depends_on.add(component_to_kfp_op[upstream_component])
# remove the extra pipeline node information
tfx_node_ir = self._dehydrate_tfx_ir(tfx_ir, component.id)
main_module = get_module_source_from_module(sys.modules["__main__"])
step_module = component.component_type.split(".")[:-1]
if step_module[0] == "__main__":
step_module = main_module
else:
step_module = ".".join(step_module)
kfp_component = KubeflowComponent(
main_module=main_module,
step_module=step_module,
step_function_name=component.id,
component=component,
depends_on=depends_on,
image=self._kubeflow_config.image,
pod_labels_to_attach=self._pod_labels_to_attach,
tfx_ir=tfx_node_ir,
runtime_parameters=self._params_by_component_id[component.id],
)
for operator in self._kubeflow_config.pipeline_operator_funcs:
kfp_component.container_op.apply(operator)
component_to_kfp_op[component] = kfp_component.container_op
def _del_unused_field(
self, node_id: str, message_dict: MutableMapping[str, Any]
) -> None:
"""Remove fields that are not used by the pipeline."""
for item in list(message_dict.keys()):
if item != node_id:
del message_dict[item]
def _dehydrate_tfx_ir(
self, original_pipeline: Pb2Pipeline, node_id: str
) -> Pb2Pipeline:
"""Dehydrate the TFX IR to remove unused fields."""
pipeline = copy.deepcopy(original_pipeline)
for node in pipeline.nodes:
if (
node.WhichOneof("node") == "pipeline_node"
and node.pipeline_node.node_info.id == node_id
):
del pipeline.nodes[:]
pipeline.nodes.extend([node])
break
deployment_config = IntermediateDeploymentConfig()
pipeline.deployment_config.Unpack(deployment_config)
self._del_unused_field(node_id, deployment_config.executor_specs)
self._del_unused_field(node_id, deployment_config.custom_driver_specs)
self._del_unused_field(
node_id, deployment_config.node_level_platform_configs
)
pipeline.deployment_config.Pack(deployment_config)
return pipeline
def _generate_tfx_ir(self, pipeline: TfxPipeline) -> Pb2Pipeline:
"""Generate the TFX IR from the logical TFX pipeline."""
result = self._tfx_compiler.compile(pipeline)
return result
def run(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> None:
"""Compiles and outputs a Kubeflow Pipeline YAML definition file.
Args:
pipeline: The logical TFX pipeline to use when building the Kubeflow
pipeline.
stack: The ZenML stack that the pipeline is running on.
runtime_configuration: The runtime configuration.
"""
tfx_pipeline = create_tfx_pipeline(pipeline, stack=stack)
pipeline_root = tfx_pipeline.pipeline_info.pipeline_root
if not isinstance(pipeline_root, str):
raise TypeError(
"TFX Pipeline root may not be a Placeholder, "
"but must be a specific string."
)
for component in tfx_pipeline.components:
# TODO(b/187122662): Pass through pip dependencies as a first-class
# component flag.
if isinstance(component, tfx_base_component.BaseComponent):
component._resolve_pip_dependencies(pipeline_root)
def _construct_pipeline() -> None:
"""Creates Kubeflow ContainerOps for each TFX component
encountered in the pipeline definition."""
self._construct_pipeline_graph(
pipeline, tfx_pipeline, stack, runtime_configuration
)
# Need to run this first to get self._params populated. Then KFP
# compiler can correctly match default value with PipelineParam.
self._parse_parameter_from_pipeline(tfx_pipeline)
# Create workflow spec and write out to package.
self._compiler._create_and_write_workflow(
# pylint: disable=protected-access
pipeline_func=_construct_pipeline,
pipeline_name=tfx_pipeline.pipeline_info.pipeline_name,
params_list=self._params,
package_path=self._output_path,
)
logger.info(
"Finished writing kubeflow pipeline definition file '%s'.",
self._output_path,
)
config: PipelineConfig
property
readonly
The config property
__init__(self, config, output_path, pod_labels_to_attach=None)
special
Initializes KubeflowDagRunner for compiling a Kubeflow Pipeline.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
config |
KubeflowDagRunnerConfig |
A KubeflowDagRunnerConfig object to specify runtime configuration when running the pipeline under Kubeflow. |
required |
output_path |
str |
Path where the pipeline definition file will be stored. |
required |
pod_labels_to_attach |
Optional[Dict[str, str]] |
Optional set of pod labels to attach to GKE pod spun up for this pipeline. Default to the 3 labels: 1. add-pod-env: true, 2. pipeline SDK type, 3. pipeline unique ID, where 2 and 3 are instrumentation of usage tracking. |
None |
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_dag_runner.py
def __init__(
self,
config: KubeflowDagRunnerConfig,
output_path: str,
pod_labels_to_attach: Optional[Dict[str, str]] = None,
):
"""Initializes KubeflowDagRunner for compiling a Kubeflow Pipeline.
Args:
config: A KubeflowDagRunnerConfig object to specify runtime
configuration when running the pipeline under Kubeflow.
output_path: Path where the pipeline definition file will be stored.
pod_labels_to_attach: Optional set of pod labels to attach to GKE pod
spun up for this pipeline. Default to the 3 labels:
1. add-pod-env: true,
2. pipeline SDK type,
3. pipeline unique ID,
where 2 and 3 are instrumentation of usage tracking.
"""
self._config = config or pipeline_config.PipelineConfig()
self._kubeflow_config = config
self._output_path = output_path
self._compiler = compiler.Compiler()
self._tfx_compiler = tfx_compiler.Compiler()
self._params: List[dsl.PipelineParam] = []
self._params_by_component_id: Dict[
str, List[data_types.RuntimeParameter]
] = collections.defaultdict(list)
self._deduped_parameter_names: Set[str] = set()
self._pod_labels_to_attach = (
pod_labels_to_attach or get_default_pod_labels()
)
run(self, pipeline, stack, runtime_configuration)
Compiles and outputs a Kubeflow Pipeline YAML definition file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline |
BasePipeline |
The logical TFX pipeline to use when building the Kubeflow pipeline. |
required |
stack |
Stack |
The ZenML stack that the pipeline is running on. |
required |
runtime_configuration |
RuntimeConfiguration |
The runtime configuration. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_dag_runner.py
def run(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> None:
"""Compiles and outputs a Kubeflow Pipeline YAML definition file.
Args:
pipeline: The logical TFX pipeline to use when building the Kubeflow
pipeline.
stack: The ZenML stack that the pipeline is running on.
runtime_configuration: The runtime configuration.
"""
tfx_pipeline = create_tfx_pipeline(pipeline, stack=stack)
pipeline_root = tfx_pipeline.pipeline_info.pipeline_root
if not isinstance(pipeline_root, str):
raise TypeError(
"TFX Pipeline root may not be a Placeholder, "
"but must be a specific string."
)
for component in tfx_pipeline.components:
# TODO(b/187122662): Pass through pip dependencies as a first-class
# component flag.
if isinstance(component, tfx_base_component.BaseComponent):
component._resolve_pip_dependencies(pipeline_root)
def _construct_pipeline() -> None:
"""Creates Kubeflow ContainerOps for each TFX component
encountered in the pipeline definition."""
self._construct_pipeline_graph(
pipeline, tfx_pipeline, stack, runtime_configuration
)
# Need to run this first to get self._params populated. Then KFP
# compiler can correctly match default value with PipelineParam.
self._parse_parameter_from_pipeline(tfx_pipeline)
# Create workflow spec and write out to package.
self._compiler._create_and_write_workflow(
# pylint: disable=protected-access
pipeline_func=_construct_pipeline,
pipeline_name=tfx_pipeline.pipeline_info.pipeline_name,
params_list=self._params,
package_path=self._output_path,
)
logger.info(
"Finished writing kubeflow pipeline definition file '%s'.",
self._output_path,
)
KubeflowDagRunnerConfig (PipelineConfig)
Runtime configuration parameters specific to execution on Kubeflow.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_dag_runner.py
class KubeflowDagRunnerConfig(pipeline_config.PipelineConfig):
"""Runtime configuration parameters specific to execution on Kubeflow."""
def __init__(
self,
image: str,
pipeline_operator_funcs: Optional[List[OpFunc]] = None,
supported_launcher_classes: Optional[
List[Type[base_component_launcher.BaseComponentLauncher]]
] = None,
**kwargs: Any
):
"""Creates a KubeflowDagRunnerConfig object.
The user can use pipeline_operator_funcs to apply modifications to
ContainerOps used in the pipeline. For example, to ensure the pipeline
steps mount a GCP secret, and a Persistent Volume, one can create config
object like so:
from kfp import gcp, onprem
mount_secret_op = gcp.use_secret('my-secret-name)
mount_volume_op = onprem.mount_pvc(
"my-persistent-volume-claim",
"my-volume-name",
"/mnt/volume-mount-path")
config = KubeflowDagRunnerConfig(
pipeline_operator_funcs=[mount_secret_op, mount_volume_op]
)
Args:
image: The docker image to use in the pipeline.
pipeline_operator_funcs: A list of ContainerOp modifying functions
that will be applied to every container step in the pipeline.
supported_launcher_classes: A list of component launcher classes that
are supported by the current pipeline. List sequence determines the
order in which launchers are chosen for each component being run.
**kwargs: keyword args for PipelineConfig.
"""
supported_launcher_classes = supported_launcher_classes or [
in_process_component_launcher.InProcessComponentLauncher,
kubernetes_component_launcher.KubernetesComponentLauncher,
]
super().__init__(
supported_launcher_classes=supported_launcher_classes, **kwargs
)
self.pipeline_operator_funcs = (
pipeline_operator_funcs or get_default_pipeline_operator_funcs()
)
self.image = image
__init__(self, image, pipeline_operator_funcs=None, supported_launcher_classes=None, **kwargs)
special
Creates a KubeflowDagRunnerConfig object. The user can use pipeline_operator_funcs to apply modifications to ContainerOps used in the pipeline. For example, to ensure the pipeline steps mount a GCP secret, and a Persistent Volume, one can create config object like so: from kfp import gcp, onprem mount_secret_op = gcp.use_secret('my-secret-name) mount_volume_op = onprem.mount_pvc( "my-persistent-volume-claim", "my-volume-name", "/mnt/volume-mount-path") config = KubeflowDagRunnerConfig( pipeline_operator_funcs=[mount_secret_op, mount_volume_op] )
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
image |
str |
The docker image to use in the pipeline. |
required |
pipeline_operator_funcs |
Optional[List[Callable[[kfp.dsl._container_op.ContainerOp], Union[kfp.dsl._container_op.ContainerOp, NoneType]]]] |
A list of ContainerOp modifying functions that will be applied to every container step in the pipeline. |
None |
supported_launcher_classes |
Optional[List[Type[tfx.orchestration.launcher.base_component_launcher.BaseComponentLauncher]]] |
A list of component launcher classes that are supported by the current pipeline. List sequence determines the order in which launchers are chosen for each component being run. |
None |
**kwargs |
Any |
keyword args for PipelineConfig. |
{} |
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_dag_runner.py
def __init__(
self,
image: str,
pipeline_operator_funcs: Optional[List[OpFunc]] = None,
supported_launcher_classes: Optional[
List[Type[base_component_launcher.BaseComponentLauncher]]
] = None,
**kwargs: Any
):
"""Creates a KubeflowDagRunnerConfig object.
The user can use pipeline_operator_funcs to apply modifications to
ContainerOps used in the pipeline. For example, to ensure the pipeline
steps mount a GCP secret, and a Persistent Volume, one can create config
object like so:
from kfp import gcp, onprem
mount_secret_op = gcp.use_secret('my-secret-name)
mount_volume_op = onprem.mount_pvc(
"my-persistent-volume-claim",
"my-volume-name",
"/mnt/volume-mount-path")
config = KubeflowDagRunnerConfig(
pipeline_operator_funcs=[mount_secret_op, mount_volume_op]
)
Args:
image: The docker image to use in the pipeline.
pipeline_operator_funcs: A list of ContainerOp modifying functions
that will be applied to every container step in the pipeline.
supported_launcher_classes: A list of component launcher classes that
are supported by the current pipeline. List sequence determines the
order in which launchers are chosen for each component being run.
**kwargs: keyword args for PipelineConfig.
"""
supported_launcher_classes = supported_launcher_classes or [
in_process_component_launcher.InProcessComponentLauncher,
kubernetes_component_launcher.KubernetesComponentLauncher,
]
super().__init__(
supported_launcher_classes=supported_launcher_classes, **kwargs
)
self.pipeline_operator_funcs = (
pipeline_operator_funcs or get_default_pipeline_operator_funcs()
)
self.image = image
get_default_pipeline_operator_funcs(use_gcp_sa=False)
Returns a default list of pipeline operator functions.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
use_gcp_sa |
bool |
If true, mount a GCP service account secret to each pod, with the name _KUBEFLOW_GCP_SECRET_NAME. |
False |
Returns:
| Type | Description |
|---|---|
List[Callable[[kfp.dsl._container_op.ContainerOp], Optional[kfp.dsl._container_op.ContainerOp]]] |
A list of functions with type OpFunc. |
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_dag_runner.py
def get_default_pipeline_operator_funcs(
use_gcp_sa: bool = False,
) -> List[OpFunc]:
"""Returns a default list of pipeline operator functions.
Args:
use_gcp_sa: If true, mount a GCP service account secret to each pod, with
the name _KUBEFLOW_GCP_SECRET_NAME.
Returns:
A list of functions with type OpFunc.
"""
# Enables authentication for GCP services if needed.
gcp_secret_op = gcp.use_gcp_secret(_KUBEFLOW_GCP_SECRET_NAME)
# Mounts configmap containing Metadata gRPC server configuration.
mount_config_map_op = _mount_config_map_op("metadata-grpc-configmap")
if use_gcp_sa:
return [gcp_secret_op, mount_config_map_op]
else:
return [mount_config_map_op]
get_default_pod_labels()
Returns the default pod label dict for Kubeflow.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_dag_runner.py
def get_default_pod_labels() -> Dict[str, str]:
"""Returns the default pod label dict for Kubeflow."""
# KFP default transformers add pod env:
# https://github.com/kubeflow/pipelines/blob/0.1.32/sdk/python/kfp/compiler/_default_transformers.py
result = {"add-pod-env": "true", telemetry_utils.LABEL_KFP_SDK_ENV: "tfx"}
return result
kubeflow_orchestrator
KubeflowOrchestrator (BaseOrchestrator)
pydantic-model
Orchestrator responsible for running pipelines using Kubeflow.
Attributes:
| Name | Type | Description |
|---|---|---|
custom_docker_base_image_name |
Optional[str] |
Name of a docker image that should be used as the base for the image that will be run on KFP pods. If no custom image is given, a basic image of the active ZenML version will be used. Note: This image needs to have ZenML installed, otherwise the pipeline execution will fail. For that reason, you might want to extend the ZenML docker images found here: https://hub.docker.com/r/zenmldocker/zenml/ |
kubeflow_pipelines_ui_port |
int |
A local port to which the KFP UI will be forwarded. |
kubernetes_context |
Optional[str] |
Optional name of a kubernetes context to run
pipelines in. If not set, the current active context will be used.
You can find the active context by running |
synchronous |
If |
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
class KubeflowOrchestrator(BaseOrchestrator):
"""Orchestrator responsible for running pipelines using Kubeflow.
Attributes:
custom_docker_base_image_name: Name of a docker image that should be
used as the base for the image that will be run on KFP pods. If no
custom image is given, a basic image of the active ZenML version
will be used. **Note**: This image needs to have ZenML installed,
otherwise the pipeline execution will fail. For that reason, you
might want to extend the ZenML docker images found here:
https://hub.docker.com/r/zenmldocker/zenml/
kubeflow_pipelines_ui_port: A local port to which the KFP UI will be
forwarded.
kubernetes_context: Optional name of a kubernetes context to run
pipelines in. If not set, the current active context will be used.
You can find the active context by running `kubectl config
current-context`.
synchronous: If `True`, running a pipeline using this orchestrator will
block until all steps finished running on KFP.
"""
custom_docker_base_image_name: Optional[str] = None
kubeflow_pipelines_ui_port: int = DEFAULT_KFP_UI_PORT
kubernetes_context: Optional[str] = None
synchronous = False
# Class Configuration
FLAVOR: ClassVar[str] = KUBEFLOW
@staticmethod
def _get_k3d_cluster_name(uuid: UUID) -> str:
"""Returns the k3d cluster name corresponding to the orchestrator
UUID."""
# k3d only allows cluster names with up to 32 characters; use the
# first 8 chars of the orchestrator UUID as identifier
return f"zenml-kubeflow-{str(uuid)[:8]}"
@staticmethod
def _get_k3d_kubernetes_context(uuid: UUID) -> str:
"""Returns the name of the kubernetes context associated with the k3d
cluster managed locally by ZenML corresponding to the orchestrator
UUID."""
return f"k3d-{KubeflowOrchestrator._get_k3d_cluster_name(uuid)}"
@root_validator
def set_default_kubernetes_context(
cls, values: Dict[str, Any]
) -> Dict[str, Any]:
"""Pydantic root_validator that sets the default `kubernetes_context`
value to the value that is used to create the locally managed k3d
cluster, if not explicitly set.
Args:
values: Values passed to the object constructor
Returns:
Values passed to the Pydantic constructor
"""
if values.get("kubernetes_context"):
return values
# not likely, due to Pydantic validation, but mypy complains
assert "uuid" in values
values["kubernetes_context"] = cls._get_k3d_kubernetes_context(
values["uuid"]
)
return values
@property
def validator(self) -> Optional[StackValidator]:
"""Validates that the stack contains a container registry and that
requirements are met for local components."""
def _validate_local_requirements(stack: Stack) -> Tuple[bool, str]:
container_registry = stack.container_registry
# should not happen, because the stack validation takes care of
# this, but just in case
assert container_registry is not None
if not self.is_local:
# if the orchestrator is not running in a local k3d cluster,
# we cannot have any other local components in our stack, because
# we cannot mount the local path into the container. This
# may result in problems when running the pipeline, because
# the local components will not be available inside the
# Kubeflow containers.
# go through all stack components and identify those that
# advertise a local path where they persist information that
# they need to be available when running pipelines.
for stack_comp in stack.components.values():
local_path = stack_comp.local_path
if not local_path:
continue
return False, (
f"The Kubeflow orchestrator is not running in a local "
f"k3d cluster. The '{stack_comp.name}' "
f"{stack_comp.TYPE.value} is a local stack component "
f"and will not be available in the Kubeflow pipeline "
f"step. Please ensure that you always use non-local "
f"stack components with a remote Kubeflow orchestrator, "
f"otherwise you may run into pipeline execution "
f"problems."
)
# if the orchestrator is remote, the container registry must
# also be remote.
if container_registry.is_local:
return False, (
f"The Kubeflow orchestrator is not running in a local "
f"k3d cluster but the {container_registry.name} "
f"container registry URI '{container_registry.uri}' "
f"points to a local container registry. Please ensure "
f"that you always use non-local stack components with "
f"a remote Kubeflow orchestrator, otherwise you will "
f"run into problems."
)
else:
# if the orchestrator is local, the container registry must
# also be local.
if not container_registry.is_local:
return False, (
f"The container registry URI '{container_registry.uri}' "
f"doesn't match the expected format 'localhost:$PORT'. "
f"The local Kubeflow orchestrator only works with a "
f"local container registry because it cannot "
f"authenticate to external container registries."
)
return True, ""
return StackValidator(
required_components={StackComponentType.CONTAINER_REGISTRY},
custom_validation_function=_validate_local_requirements,
)
def get_docker_image_name(self, pipeline_name: str) -> str:
"""Returns the full docker image name including registry and tag."""
base_image_name = f"zenml-kubeflow:{pipeline_name}"
container_registry = Repository().active_stack.container_registry
if container_registry:
registry_uri = container_registry.uri.rstrip("/")
return f"{registry_uri}/{base_image_name}"
else:
return base_image_name
@property
def is_local(self) -> bool:
"""Returns `True` if the KFP orchestrator is running locally (i.e. in
the local k3d cluster managed by ZenML).
"""
return self.kubernetes_context == self._get_k3d_kubernetes_context(
self.uuid
)
@property
def root_directory(self) -> str:
"""Returns path to the root directory for all files concerning
this orchestrator."""
return os.path.join(
zenml.io.utils.get_global_config_directory(),
"kubeflow",
str(self.uuid),
)
@property
def pipeline_directory(self) -> str:
"""Returns path to a directory in which the kubeflow pipeline files
are stored."""
return os.path.join(self.root_directory, "pipelines")
def prepare_pipeline_deployment(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> None:
"""Builds a docker image for the current environment and uploads it to
a container registry if configured.
"""
from zenml.utils.docker_utils import (
build_docker_image,
push_docker_image,
)
# if the orchestrator is not running in a local k3d cluster,
# we cannot mount the local path into the container. This
# may result in problems when running the pipeline, because
# the local components will not be available inside the
# Kubeflow containers.
if self.kubernetes_context:
# go through all stack components and identify those that advertise
# a local path where they persist information that they need to be
# available when running pipelines.
for stack_comp in stack.components.values():
local_path = stack_comp.local_path
if not local_path:
continue
logger.warning(
"The Kubeflow orchestrator is not running in a local k3d "
"cluster. The '%s' %s is a local stack component and will "
"not be available in the Kubeflow pipeline step. Please "
"ensure that you never combine non-local stack components "
"with a remote orchestrator, otherwise you may run into "
"pipeline execution problems.",
stack_comp.name,
stack_comp.TYPE.value,
)
image_name = self.get_docker_image_name(pipeline.name)
requirements = {*stack.requirements(), *pipeline.requirements}
logger.debug("Kubeflow docker container requirements: %s", requirements)
build_docker_image(
build_context_path=get_source_root_path(),
image_name=image_name,
dockerignore_path=pipeline.dockerignore_file,
requirements=requirements,
base_image=self.custom_docker_base_image_name,
environment_vars=self._get_environment_vars_from_secrets(
pipeline.secrets
),
)
if stack.container_registry:
push_docker_image(image_name)
def run_pipeline(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> Any:
"""Runs a pipeline on Kubeflow Pipelines."""
# First check whether its running in a notebok
from zenml.environment import Environment
if Environment.in_notebook():
raise RuntimeError(
"The Kubeflow orchestrator cannot run pipelines in a notebook "
"environment. The reason is that it is non-trivial to create "
"a Docker image of a notebook. Please consider refactoring "
"your notebook cells into separate scripts in a Python module "
"and run the code outside of a notebook when using this "
"orchestrator."
)
from zenml.utils.docker_utils import get_image_digest
image_name = self.get_docker_image_name(pipeline.name)
image_name = get_image_digest(image_name) or image_name
fileio.makedirs(self.pipeline_directory)
pipeline_file_path = os.path.join(
self.pipeline_directory, f"{pipeline.name}.yaml"
)
runner_config = KubeflowDagRunnerConfig(image=image_name)
runner = KubeflowDagRunner(
config=runner_config, output_path=pipeline_file_path
)
runner.run(
pipeline=pipeline,
stack=stack,
runtime_configuration=runtime_configuration,
)
self._upload_and_run_pipeline(
pipeline_name=pipeline.name,
pipeline_file_path=pipeline_file_path,
runtime_configuration=runtime_configuration,
enable_cache=pipeline.enable_cache,
)
def _upload_and_run_pipeline(
self,
pipeline_name: str,
pipeline_file_path: str,
runtime_configuration: "RuntimeConfiguration",
enable_cache: bool,
) -> None:
"""Tries to upload and run a KFP pipeline.
Args:
pipeline_name: Name of the pipeline.
pipeline_file_path: Path to the pipeline definition file.
runtime_configuration: Runtime configuration of the pipeline run.
enable_cache: Whether caching is enabled for this pipeline run.
"""
try:
logger.info(
"Running in kubernetes context '%s'.",
self.kubernetes_context,
)
# upload the pipeline to Kubeflow and start it
client = kfp.Client(kube_context=self.kubernetes_context)
if runtime_configuration.schedule:
try:
experiment = client.get_experiment(pipeline_name)
logger.info(
"A recurring run has already been created with this "
"pipeline. Creating new recurring run now.."
)
except (ValueError, ApiException):
experiment = client.create_experiment(pipeline_name)
logger.info(
"Creating a new recurring run for pipeline '%s'.. ",
pipeline_name,
)
logger.info(
"You can see all recurring runs under the '%s' experiment.'",
pipeline_name,
)
schedule = runtime_configuration.schedule
result = client.create_recurring_run(
experiment_id=experiment.id,
job_name=runtime_configuration.run_name,
pipeline_package_path=pipeline_file_path,
enable_caching=enable_cache,
start_time=schedule.utc_start_time,
end_time=schedule.utc_end_time,
interval_second=schedule.interval_second,
no_catchup=not schedule.catchup,
)
logger.info("Started recurring run with ID '%s'.", result.id)
else:
logger.info(
"No schedule detected. Creating a one-off pipeline run.."
)
result = client.create_run_from_pipeline_package(
pipeline_file_path,
arguments={},
run_name=runtime_configuration.run_name,
enable_caching=enable_cache,
)
logger.info(
"Started one-off pipeline run with ID '%s'.", result.run_id
)
if self.synchronous:
# TODO [ENG-698]: Allow configuration of the timeout as a
# runtime option
client.wait_for_run_completion(
run_id=result.run_id, timeout=1200
)
except urllib3.exceptions.HTTPError as error:
logger.warning(
"Failed to upload Kubeflow pipeline: %s. "
"Please make sure your kube config is configured and the "
"current context is set correctly.",
error,
)
@property
def _pid_file_path(self) -> str:
"""Returns path to the daemon PID file."""
return os.path.join(self.root_directory, "kubeflow_daemon.pid")
@property
def log_file(self) -> str:
"""Path of the daemon log file."""
return os.path.join(self.root_directory, "kubeflow_daemon.log")
@property
def _k3d_cluster_name(self) -> str:
"""Returns the K3D cluster name."""
return self._get_k3d_cluster_name(self.uuid)
def _get_k3d_registry_name(self, port: int) -> str:
"""Returns the K3D registry name."""
return f"k3d-zenml-kubeflow-registry.localhost:{port}"
@property
def _k3d_registry_config_path(self) -> str:
"""Returns the path to the K3D registry config yaml."""
return os.path.join(self.root_directory, "k3d_registry.yaml")
def _get_kfp_ui_daemon_port(self) -> int:
"""Port to use for the KFP UI daemon."""
port = self.kubeflow_pipelines_ui_port
if port == DEFAULT_KFP_UI_PORT and not networking_utils.port_available(
port
):
# if the user didn't specify a specific port and the default
# port is occupied, fallback to a random open port
port = networking_utils.find_available_port()
return port
def list_manual_setup_steps(
self, container_registry_name: str, container_registry_path: str
) -> None:
"""Logs manual steps needed to setup the Kubeflow local orchestrator."""
global_config_dir_path = zenml.io.utils.get_global_config_directory()
kubeflow_commands = [
f"> k3d cluster create CLUSTER_NAME --image {local_deployment_utils.K3S_IMAGE_NAME} --registry-create {container_registry_name} --registry-config {container_registry_path} --volume {global_config_dir_path}:{global_config_dir_path}\n",
f"> kubectl --context CLUSTER_NAME apply -k github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref={KFP_VERSION}&timeout=1m",
"> kubectl --context CLUSTER_NAME wait --timeout=60s --for condition=established crd/applications.app.k8s.io",
f"> kubectl --context CLUSTER_NAME apply -k github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref={KFP_VERSION}&timeout=1m",
f"> kubectl --namespace kubeflow port-forward svc/ml-pipeline-ui {self.kubeflow_pipelines_ui_port}:80",
]
logger.error("Unable to spin up local Kubeflow Pipelines deployment.")
logger.info(
"If you wish to spin up this Kubeflow local orchestrator manually, "
"please enter the following commands (substituting where appropriate):\n"
)
logger.info("\n".join(kubeflow_commands))
@property
def is_provisioned(self) -> bool:
"""Returns if a local k3d cluster for this orchestrator exists."""
if not local_deployment_utils.check_prerequisites():
# if any prerequisites are missing there is certainly no
# local deployment running
return False
return self.is_cluster_provisioned
@property
def is_running(self) -> bool:
"""Returns if the local k3d cluster and the UI daemon for this
orchestrator are both running."""
return (
self.is_provisioned
and self.is_cluster_running
and self.is_daemon_running
)
@property
def is_suspended(self) -> bool:
"""Returns if the local k3d cluster and the UI daemon for this
orchestrator are both stopped."""
return (
self.is_provisioned
and not self.is_cluster_running
and not self.is_daemon_running
)
@property
def is_cluster_provisioned(self) -> bool:
"""Returns if the local k3d cluster for this orchestrator is provisioned.
For remote (i.e. not managed by ZenML) Kubeflow Pipelines installations,
this always returns True.
"""
if not self.is_local:
return True
return local_deployment_utils.k3d_cluster_exists(
cluster_name=self._k3d_cluster_name
)
@property
def is_cluster_running(self) -> bool:
"""Returns if the local k3d cluster for this orchestrator is running.
For remote (i.e. not managed by ZenML) Kubeflow Pipelines installations,
this always returns True.
"""
if not self.is_local:
return True
return local_deployment_utils.k3d_cluster_running(
cluster_name=self._k3d_cluster_name
)
@property
def is_daemon_running(self) -> bool:
"""Returns if the local Kubeflow UI daemon for this orchestrator is
running."""
if sys.platform != "win32":
from zenml.utils.daemon import check_if_daemon_is_running
return check_if_daemon_is_running(self._pid_file_path)
else:
return True
def provision(self) -> None:
"""Provisions a local Kubeflow Pipelines deployment."""
if self.is_running:
logger.info(
"Found already existing local Kubeflow Pipelines deployment. "
"If there are any issues with the existing deployment, please "
"run 'zenml stack down --force' to delete it."
)
return
if not local_deployment_utils.check_prerequisites():
raise ProvisioningError(
"Unable to provision local Kubeflow Pipelines deployment: "
"Please install 'k3d' and 'kubectl' and try again."
)
container_registry = Repository().active_stack.container_registry
# should not happen, because the stack validation takes care of this,
# but just in case
assert container_registry is not None
fileio.makedirs(self.root_directory)
if not self.is_local:
# don't provision any resources if using a remote KFP installation
return
logger.info("Provisioning local Kubeflow Pipelines deployment...")
container_registry_port = int(container_registry.uri.split(":")[-1])
container_registry_name = self._get_k3d_registry_name(
port=container_registry_port
)
local_deployment_utils.write_local_registry_yaml(
yaml_path=self._k3d_registry_config_path,
registry_name=container_registry_name,
registry_uri=container_registry.uri,
)
try:
local_deployment_utils.create_k3d_cluster(
cluster_name=self._k3d_cluster_name,
registry_name=container_registry_name,
registry_config_path=self._k3d_registry_config_path,
)
kubernetes_context = self.kubernetes_context
# will never happen, but mypy doesn't know that
assert kubernetes_context is not None
local_deployment_utils.deploy_kubeflow_pipelines(
kubernetes_context=kubernetes_context
)
artifact_store = Repository().active_stack.artifact_store
if isinstance(artifact_store, LocalArtifactStore):
local_deployment_utils.add_hostpath_to_kubeflow_pipelines(
kubernetes_context=kubernetes_context,
local_path=artifact_store.path,
)
except Exception as e:
logger.error(e)
self.list_manual_setup_steps(
container_registry_name, self._k3d_registry_config_path
)
self.deprovision()
def deprovision(self) -> None:
"""Deprovisions a local Kubeflow Pipelines deployment."""
if self.is_daemon_running:
local_deployment_utils.stop_kfp_ui_daemon(
pid_file_path=self._pid_file_path
)
if self.is_local:
# don't deprovision any resources if using a remote KFP installation
local_deployment_utils.delete_k3d_cluster(
cluster_name=self._k3d_cluster_name
)
logger.info("Local kubeflow pipelines deployment deprovisioned.")
if fileio.exists(self.log_file):
fileio.remove(self.log_file)
def resume(self) -> None:
"""Resumes the local k3d cluster."""
if self.is_running:
logger.info("Local kubeflow pipelines deployment already running.")
return
if not self.is_provisioned:
raise ProvisioningError(
"Unable to resume local kubeflow pipelines deployment: No "
"resources provisioned for local deployment."
)
kubernetes_context = self.kubernetes_context
# will never happen, but mypy doesn't know that
assert kubernetes_context is not None
if self.is_local and not self.is_cluster_running:
# don't resume any resources if using a remote KFP installation
local_deployment_utils.start_k3d_cluster(
cluster_name=self._k3d_cluster_name
)
local_deployment_utils.wait_until_kubeflow_pipelines_ready(
kubernetes_context=kubernetes_context
)
if not self.is_daemon_running:
local_deployment_utils.start_kfp_ui_daemon(
pid_file_path=self._pid_file_path,
log_file_path=self.log_file,
port=self._get_kfp_ui_daemon_port(),
kubernetes_context=kubernetes_context,
)
def suspend(self) -> None:
"""Suspends the local k3d cluster."""
if not self.is_provisioned:
logger.info("Local kubeflow pipelines deployment not provisioned.")
return
if self.is_daemon_running:
local_deployment_utils.stop_kfp_ui_daemon(
pid_file_path=self._pid_file_path
)
if self.is_local and self.is_cluster_running:
# don't suspend any resources if using a remote KFP installation
local_deployment_utils.stop_k3d_cluster(
cluster_name=self._k3d_cluster_name
)
def _get_environment_vars_from_secrets(
self, secrets: List[str]
) -> Dict[str, str]:
"""Get key-value pairs from list of secrets provided by the user.
Args:
secrets: List of secrets provided by the user.
Returns:
A dictionary of key-value pairs.
Raises:
ProvisioningError: If the stack has no secrets manager."""
environment_vars: Dict[str, str] = {}
secret_manager = Repository().active_stack.secrets_manager
if secrets and secret_manager:
for secret in secrets:
secret_schema = secret_manager.get_secret(secret)
environment_vars.update(secret_schema.content)
elif secrets and not secret_manager:
raise ProvisioningError(
"Unable to provision local Kubeflow Pipelines deployment: "
f"You passed in the following secrets: { ', '.join(secrets) }, "
"however, no secrets manager is registered for the current stack."
)
else:
# No secrets provided by the user.
pass
return environment_vars
is_cluster_provisioned: bool
property
readonly
Returns if the local k3d cluster for this orchestrator is provisioned.
For remote (i.e. not managed by ZenML) Kubeflow Pipelines installations, this always returns True.
is_cluster_running: bool
property
readonly
Returns if the local k3d cluster for this orchestrator is running.
For remote (i.e. not managed by ZenML) Kubeflow Pipelines installations, this always returns True.
is_daemon_running: bool
property
readonly
Returns if the local Kubeflow UI daemon for this orchestrator is running.
is_local: bool
property
readonly
Returns True if the KFP orchestrator is running locally (i.e. in
the local k3d cluster managed by ZenML).
is_provisioned: bool
property
readonly
Returns if a local k3d cluster for this orchestrator exists.
is_running: bool
property
readonly
Returns if the local k3d cluster and the UI daemon for this orchestrator are both running.
is_suspended: bool
property
readonly
Returns if the local k3d cluster and the UI daemon for this orchestrator are both stopped.
log_file: str
property
readonly
Path of the daemon log file.
pipeline_directory: str
property
readonly
Returns path to a directory in which the kubeflow pipeline files are stored.
root_directory: str
property
readonly
Returns path to the root directory for all files concerning this orchestrator.
validator: Optional[zenml.stack.stack_validator.StackValidator]
property
readonly
Validates that the stack contains a container registry and that requirements are met for local components.
deprovision(self)
Deprovisions a local Kubeflow Pipelines deployment.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def deprovision(self) -> None:
"""Deprovisions a local Kubeflow Pipelines deployment."""
if self.is_daemon_running:
local_deployment_utils.stop_kfp_ui_daemon(
pid_file_path=self._pid_file_path
)
if self.is_local:
# don't deprovision any resources if using a remote KFP installation
local_deployment_utils.delete_k3d_cluster(
cluster_name=self._k3d_cluster_name
)
logger.info("Local kubeflow pipelines deployment deprovisioned.")
if fileio.exists(self.log_file):
fileio.remove(self.log_file)
get_docker_image_name(self, pipeline_name)
Returns the full docker image name including registry and tag.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def get_docker_image_name(self, pipeline_name: str) -> str:
"""Returns the full docker image name including registry and tag."""
base_image_name = f"zenml-kubeflow:{pipeline_name}"
container_registry = Repository().active_stack.container_registry
if container_registry:
registry_uri = container_registry.uri.rstrip("/")
return f"{registry_uri}/{base_image_name}"
else:
return base_image_name
list_manual_setup_steps(self, container_registry_name, container_registry_path)
Logs manual steps needed to setup the Kubeflow local orchestrator.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def list_manual_setup_steps(
self, container_registry_name: str, container_registry_path: str
) -> None:
"""Logs manual steps needed to setup the Kubeflow local orchestrator."""
global_config_dir_path = zenml.io.utils.get_global_config_directory()
kubeflow_commands = [
f"> k3d cluster create CLUSTER_NAME --image {local_deployment_utils.K3S_IMAGE_NAME} --registry-create {container_registry_name} --registry-config {container_registry_path} --volume {global_config_dir_path}:{global_config_dir_path}\n",
f"> kubectl --context CLUSTER_NAME apply -k github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref={KFP_VERSION}&timeout=1m",
"> kubectl --context CLUSTER_NAME wait --timeout=60s --for condition=established crd/applications.app.k8s.io",
f"> kubectl --context CLUSTER_NAME apply -k github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref={KFP_VERSION}&timeout=1m",
f"> kubectl --namespace kubeflow port-forward svc/ml-pipeline-ui {self.kubeflow_pipelines_ui_port}:80",
]
logger.error("Unable to spin up local Kubeflow Pipelines deployment.")
logger.info(
"If you wish to spin up this Kubeflow local orchestrator manually, "
"please enter the following commands (substituting where appropriate):\n"
)
logger.info("\n".join(kubeflow_commands))
prepare_pipeline_deployment(self, pipeline, stack, runtime_configuration)
Builds a docker image for the current environment and uploads it to a container registry if configured.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def prepare_pipeline_deployment(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> None:
"""Builds a docker image for the current environment and uploads it to
a container registry if configured.
"""
from zenml.utils.docker_utils import (
build_docker_image,
push_docker_image,
)
# if the orchestrator is not running in a local k3d cluster,
# we cannot mount the local path into the container. This
# may result in problems when running the pipeline, because
# the local components will not be available inside the
# Kubeflow containers.
if self.kubernetes_context:
# go through all stack components and identify those that advertise
# a local path where they persist information that they need to be
# available when running pipelines.
for stack_comp in stack.components.values():
local_path = stack_comp.local_path
if not local_path:
continue
logger.warning(
"The Kubeflow orchestrator is not running in a local k3d "
"cluster. The '%s' %s is a local stack component and will "
"not be available in the Kubeflow pipeline step. Please "
"ensure that you never combine non-local stack components "
"with a remote orchestrator, otherwise you may run into "
"pipeline execution problems.",
stack_comp.name,
stack_comp.TYPE.value,
)
image_name = self.get_docker_image_name(pipeline.name)
requirements = {*stack.requirements(), *pipeline.requirements}
logger.debug("Kubeflow docker container requirements: %s", requirements)
build_docker_image(
build_context_path=get_source_root_path(),
image_name=image_name,
dockerignore_path=pipeline.dockerignore_file,
requirements=requirements,
base_image=self.custom_docker_base_image_name,
environment_vars=self._get_environment_vars_from_secrets(
pipeline.secrets
),
)
if stack.container_registry:
push_docker_image(image_name)
provision(self)
Provisions a local Kubeflow Pipelines deployment.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def provision(self) -> None:
"""Provisions a local Kubeflow Pipelines deployment."""
if self.is_running:
logger.info(
"Found already existing local Kubeflow Pipelines deployment. "
"If there are any issues with the existing deployment, please "
"run 'zenml stack down --force' to delete it."
)
return
if not local_deployment_utils.check_prerequisites():
raise ProvisioningError(
"Unable to provision local Kubeflow Pipelines deployment: "
"Please install 'k3d' and 'kubectl' and try again."
)
container_registry = Repository().active_stack.container_registry
# should not happen, because the stack validation takes care of this,
# but just in case
assert container_registry is not None
fileio.makedirs(self.root_directory)
if not self.is_local:
# don't provision any resources if using a remote KFP installation
return
logger.info("Provisioning local Kubeflow Pipelines deployment...")
container_registry_port = int(container_registry.uri.split(":")[-1])
container_registry_name = self._get_k3d_registry_name(
port=container_registry_port
)
local_deployment_utils.write_local_registry_yaml(
yaml_path=self._k3d_registry_config_path,
registry_name=container_registry_name,
registry_uri=container_registry.uri,
)
try:
local_deployment_utils.create_k3d_cluster(
cluster_name=self._k3d_cluster_name,
registry_name=container_registry_name,
registry_config_path=self._k3d_registry_config_path,
)
kubernetes_context = self.kubernetes_context
# will never happen, but mypy doesn't know that
assert kubernetes_context is not None
local_deployment_utils.deploy_kubeflow_pipelines(
kubernetes_context=kubernetes_context
)
artifact_store = Repository().active_stack.artifact_store
if isinstance(artifact_store, LocalArtifactStore):
local_deployment_utils.add_hostpath_to_kubeflow_pipelines(
kubernetes_context=kubernetes_context,
local_path=artifact_store.path,
)
except Exception as e:
logger.error(e)
self.list_manual_setup_steps(
container_registry_name, self._k3d_registry_config_path
)
self.deprovision()
resume(self)
Resumes the local k3d cluster.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def resume(self) -> None:
"""Resumes the local k3d cluster."""
if self.is_running:
logger.info("Local kubeflow pipelines deployment already running.")
return
if not self.is_provisioned:
raise ProvisioningError(
"Unable to resume local kubeflow pipelines deployment: No "
"resources provisioned for local deployment."
)
kubernetes_context = self.kubernetes_context
# will never happen, but mypy doesn't know that
assert kubernetes_context is not None
if self.is_local and not self.is_cluster_running:
# don't resume any resources if using a remote KFP installation
local_deployment_utils.start_k3d_cluster(
cluster_name=self._k3d_cluster_name
)
local_deployment_utils.wait_until_kubeflow_pipelines_ready(
kubernetes_context=kubernetes_context
)
if not self.is_daemon_running:
local_deployment_utils.start_kfp_ui_daemon(
pid_file_path=self._pid_file_path,
log_file_path=self.log_file,
port=self._get_kfp_ui_daemon_port(),
kubernetes_context=kubernetes_context,
)
run_pipeline(self, pipeline, stack, runtime_configuration)
Runs a pipeline on Kubeflow Pipelines.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def run_pipeline(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> Any:
"""Runs a pipeline on Kubeflow Pipelines."""
# First check whether its running in a notebok
from zenml.environment import Environment
if Environment.in_notebook():
raise RuntimeError(
"The Kubeflow orchestrator cannot run pipelines in a notebook "
"environment. The reason is that it is non-trivial to create "
"a Docker image of a notebook. Please consider refactoring "
"your notebook cells into separate scripts in a Python module "
"and run the code outside of a notebook when using this "
"orchestrator."
)
from zenml.utils.docker_utils import get_image_digest
image_name = self.get_docker_image_name(pipeline.name)
image_name = get_image_digest(image_name) or image_name
fileio.makedirs(self.pipeline_directory)
pipeline_file_path = os.path.join(
self.pipeline_directory, f"{pipeline.name}.yaml"
)
runner_config = KubeflowDagRunnerConfig(image=image_name)
runner = KubeflowDagRunner(
config=runner_config, output_path=pipeline_file_path
)
runner.run(
pipeline=pipeline,
stack=stack,
runtime_configuration=runtime_configuration,
)
self._upload_and_run_pipeline(
pipeline_name=pipeline.name,
pipeline_file_path=pipeline_file_path,
runtime_configuration=runtime_configuration,
enable_cache=pipeline.enable_cache,
)
set_default_kubernetes_context(values)
classmethod
Pydantic root_validator that sets the default kubernetes_context
value to the value that is used to create the locally managed k3d
cluster, if not explicitly set.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
values |
Dict[str, Any] |
Values passed to the object constructor |
required |
Returns:
| Type | Description |
|---|---|
Dict[str, Any] |
Values passed to the Pydantic constructor |
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
@root_validator
def set_default_kubernetes_context(
cls, values: Dict[str, Any]
) -> Dict[str, Any]:
"""Pydantic root_validator that sets the default `kubernetes_context`
value to the value that is used to create the locally managed k3d
cluster, if not explicitly set.
Args:
values: Values passed to the object constructor
Returns:
Values passed to the Pydantic constructor
"""
if values.get("kubernetes_context"):
return values
# not likely, due to Pydantic validation, but mypy complains
assert "uuid" in values
values["kubernetes_context"] = cls._get_k3d_kubernetes_context(
values["uuid"]
)
return values
suspend(self)
Suspends the local k3d cluster.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_orchestrator.py
def suspend(self) -> None:
"""Suspends the local k3d cluster."""
if not self.is_provisioned:
logger.info("Local kubeflow pipelines deployment not provisioned.")
return
if self.is_daemon_running:
local_deployment_utils.stop_kfp_ui_daemon(
pid_file_path=self._pid_file_path
)
if self.is_local and self.is_cluster_running:
# don't suspend any resources if using a remote KFP installation
local_deployment_utils.stop_k3d_cluster(
cluster_name=self._k3d_cluster_name
)
kubeflow_utils
Common utility for Kubeflow-based orchestrator.
replace_placeholder(component)
Replaces the RuntimeParameter placeholders with kfp.dsl.PipelineParam.
Source code in zenml/integrations/kubeflow/orchestrators/kubeflow_utils.py
def replace_placeholder(component: base_node.BaseNode) -> None:
"""Replaces the RuntimeParameter placeholders with kfp.dsl.PipelineParam."""
keys = list(component.exec_properties.keys())
for key in keys:
exec_property = component.exec_properties[key]
if not isinstance(exec_property, data_types.RuntimeParameter):
continue
component.exec_properties[key] = str(
dsl.PipelineParam(name=exec_property.name)
)
local_deployment_utils
add_hostpath_to_kubeflow_pipelines(kubernetes_context, local_path)
Patches the Kubeflow Pipelines deployment to mount a local folder as a hostpath for visualization purposes.
This function reconfigures the Kubeflow pipelines deployment to use a shared local folder to support loading the Tensorboard viewer and other pipeline visualization results from a local artifact store, as described here:
https://github.com/kubeflow/pipelines/blob/master/docs/config/volume-support.md
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
kubernetes_context |
str |
The kubernetes context on which Kubeflow Pipelines should be patched. |
required |
local_path |
str |
The path to the local folder to mount as a hostpath. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def add_hostpath_to_kubeflow_pipelines(
kubernetes_context: str, local_path: str
) -> None:
"""Patches the Kubeflow Pipelines deployment to mount a local folder as
a hostpath for visualization purposes.
This function reconfigures the Kubeflow pipelines deployment to use a
shared local folder to support loading the Tensorboard viewer and other
pipeline visualization results from a local artifact store, as described
here:
https://github.com/kubeflow/pipelines/blob/master/docs/config/volume-support.md
Args:
kubernetes_context: The kubernetes context on which Kubeflow Pipelines
should be patched.
local_path: The path to the local folder to mount as a hostpath.
"""
logger.info("Patching Kubeflow Pipelines to mount a local folder.")
pod_template = {
"spec": {
"serviceAccountName": "kubeflow-pipelines-viewer",
"containers": [
{
"volumeMounts": [
{
"mountPath": local_path,
"name": "local-artifact-store",
}
]
}
],
"volumes": [
{
"hostPath": {
"path": local_path,
"type": "Directory",
},
"name": "local-artifact-store",
}
],
}
}
pod_template_json = json.dumps(pod_template, indent=2)
config_map_data = {"data": {"viewer-pod-template.json": pod_template_json}}
config_map_data_json = json.dumps(config_map_data, indent=2)
logger.debug(
"Adding host path volume for local path `%s` to kubeflow pipeline"
"viewer pod template configuration.",
local_path,
)
subprocess.check_call(
[
"kubectl",
"--context",
kubernetes_context,
"-n",
"kubeflow",
"patch",
"configmap/ml-pipeline-ui-configmap",
"--type",
"merge",
"-p",
config_map_data_json,
]
)
deployment_patch = {
"spec": {
"template": {
"spec": {
"containers": [
{
"name": "ml-pipeline-ui",
"volumeMounts": [
{
"mountPath": local_path,
"name": "local-artifact-store",
}
],
}
],
"volumes": [
{
"hostPath": {
"path": local_path,
"type": "Directory",
},
"name": "local-artifact-store",
}
],
}
}
}
}
deployment_patch_json = json.dumps(deployment_patch, indent=2)
logger.debug(
"Adding host path volume for local path `%s` to the kubeflow UI",
local_path,
)
subprocess.check_call(
[
"kubectl",
"--context",
kubernetes_context,
"-n",
"kubeflow",
"patch",
"deployment/ml-pipeline-ui",
"--type",
"strategic",
"-p",
deployment_patch_json,
]
)
wait_until_kubeflow_pipelines_ready(kubernetes_context=kubernetes_context)
logger.info("Finished patching Kubeflow Pipelines setup.")
check_prerequisites()
Checks whether all prerequisites for a local kubeflow pipelines deployment are installed.
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def check_prerequisites() -> bool:
"""Checks whether all prerequisites for a local kubeflow pipelines
deployment are installed."""
k3d_installed = shutil.which("k3d") is not None
kubectl_installed = shutil.which("kubectl") is not None
logger.debug(
"Local kubeflow deployment prerequisites: K3D - %s, Kubectl - %s",
k3d_installed,
kubectl_installed,
)
return k3d_installed and kubectl_installed
create_k3d_cluster(cluster_name, registry_name, registry_config_path)
Creates a K3D cluster.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
cluster_name |
str |
Name of the cluster to create. |
required |
registry_name |
str |
Name of the registry to create for this cluster. |
required |
registry_config_path |
str |
Path to the registry config file. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def create_k3d_cluster(
cluster_name: str, registry_name: str, registry_config_path: str
) -> None:
"""Creates a K3D cluster.
Args:
cluster_name: Name of the cluster to create.
registry_name: Name of the registry to create for this cluster.
registry_config_path: Path to the registry config file.
"""
logger.info("Creating local K3D cluster '%s'.", cluster_name)
global_config_dir_path = zenml.io.utils.get_global_config_directory()
subprocess.check_call(
[
"k3d",
"cluster",
"create",
cluster_name,
"--image",
K3S_IMAGE_NAME,
"--registry-create",
registry_name,
"--registry-config",
registry_config_path,
"--volume",
f"{global_config_dir_path}:{global_config_dir_path}",
]
)
logger.info("Finished K3D cluster creation.")
delete_k3d_cluster(cluster_name)
Deletes a K3D cluster with the given name.
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def delete_k3d_cluster(cluster_name: str) -> None:
"""Deletes a K3D cluster with the given name."""
subprocess.check_call(["k3d", "cluster", "delete", cluster_name])
logger.info("Deleted local k3d cluster '%s'.", cluster_name)
deploy_kubeflow_pipelines(kubernetes_context)
Deploys Kubeflow Pipelines.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
kubernetes_context |
str |
The kubernetes context on which Kubeflow Pipelines should be deployed. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def deploy_kubeflow_pipelines(kubernetes_context: str) -> None:
"""Deploys Kubeflow Pipelines.
Args:
kubernetes_context: The kubernetes context on which Kubeflow Pipelines
should be deployed.
"""
logger.info("Deploying Kubeflow Pipelines.")
subprocess.check_call(
[
"kubectl",
"--context",
kubernetes_context,
"apply",
"-k",
f"github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref={KFP_VERSION}&timeout=5m",
]
)
subprocess.check_call(
[
"kubectl",
"--context",
kubernetes_context,
"wait",
"--timeout=60s",
"--for",
"condition=established",
"crd/applications.app.k8s.io",
]
)
subprocess.check_call(
[
"kubectl",
"--context",
kubernetes_context,
"apply",
"-k",
f"github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref={KFP_VERSION}&timeout=5m",
]
)
wait_until_kubeflow_pipelines_ready(kubernetes_context=kubernetes_context)
logger.info("Finished Kubeflow Pipelines setup.")
k3d_cluster_exists(cluster_name)
Checks whether there exists a K3D cluster with the given name.
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def k3d_cluster_exists(cluster_name: str) -> bool:
"""Checks whether there exists a K3D cluster with the given name."""
output = subprocess.check_output(
["k3d", "cluster", "list", "--output", "json"]
)
clusters = json.loads(output)
for cluster in clusters:
if cluster["name"] == cluster_name:
return True
return False
k3d_cluster_running(cluster_name)
Checks whether the K3D cluster with the given name is running.
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def k3d_cluster_running(cluster_name: str) -> bool:
"""Checks whether the K3D cluster with the given name is running."""
output = subprocess.check_output(
["k3d", "cluster", "list", "--output", "json"]
)
clusters = json.loads(output)
for cluster in clusters:
if cluster["name"] == cluster_name:
server_count: int = cluster["serversCount"]
servers_running: int = cluster["serversRunning"]
return servers_running == server_count
return False
kubeflow_pipelines_ready(kubernetes_context)
Returns whether all Kubeflow Pipelines pods are ready.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
kubernetes_context |
str |
The kubernetes context in which the pods should be checked. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def kubeflow_pipelines_ready(kubernetes_context: str) -> bool:
"""Returns whether all Kubeflow Pipelines pods are ready.
Args:
kubernetes_context: The kubernetes context in which the pods
should be checked.
"""
try:
subprocess.check_call(
[
"kubectl",
"--context",
kubernetes_context,
"--namespace",
"kubeflow",
"wait",
"--for",
"condition=ready",
"--timeout=0s",
"pods",
"-l",
"application-crd-id=kubeflow-pipelines",
],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
return True
except subprocess.CalledProcessError:
return False
start_k3d_cluster(cluster_name)
Starts a K3D cluster with the given name.
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def start_k3d_cluster(cluster_name: str) -> None:
"""Starts a K3D cluster with the given name."""
subprocess.check_call(["k3d", "cluster", "start", cluster_name])
logger.info("Started local k3d cluster '%s'.", cluster_name)
start_kfp_ui_daemon(pid_file_path, log_file_path, port, kubernetes_context)
Starts a daemon process that forwards ports so the Kubeflow Pipelines UI is accessible in the browser.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pid_file_path |
str |
Path where the file with the daemons process ID should be written. |
required |
log_file_path |
str |
Path to a file where the daemon logs should be written. |
required |
port |
int |
Port on which the UI should be accessible. |
required |
kubernetes_context |
str |
The kubernetes context for the cluster where Kubeflow Pipelines is running. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def start_kfp_ui_daemon(
pid_file_path: str,
log_file_path: str,
port: int,
kubernetes_context: str,
) -> None:
"""Starts a daemon process that forwards ports so the Kubeflow Pipelines
UI is accessible in the browser.
Args:
pid_file_path: Path where the file with the daemons process ID should
be written.
log_file_path: Path to a file where the daemon logs should be written.
port: Port on which the UI should be accessible.
kubernetes_context: The kubernetes context for the cluster where
Kubeflow Pipelines is running.
"""
command = [
"kubectl",
"--context",
kubernetes_context,
"--namespace",
"kubeflow",
"port-forward",
"svc/ml-pipeline-ui",
f"{port}:80",
]
if not networking_utils.port_available(port):
modified_command = command.copy()
modified_command[-1] = "PORT:80"
logger.warning(
"Unable to port-forward Kubeflow Pipelines UI to local port %d "
"because the port is occupied. In order to access the Kubeflow "
"Pipelines UI at http://localhost:PORT/, please run '%s' in a "
"separate command line shell (replace PORT with a free port of "
"your choice).",
port,
" ".join(modified_command),
)
elif sys.platform == "win32":
logger.warning(
"Daemon functionality not supported on Windows. "
"In order to access the Kubeflow Pipelines UI at "
"http://localhost:%d/, please run '%s' in a separate command "
"line shell.",
port,
" ".join(command),
)
else:
from zenml.utils import daemon
def _daemon_function() -> None:
"""Port-forwards the Kubeflow Pipelines UI pod."""
subprocess.check_call(command)
daemon.run_as_daemon(
_daemon_function, pid_file=pid_file_path, log_file=log_file_path
)
logger.info(
"Started Kubeflow Pipelines UI daemon (check the daemon logs at %s "
"in case you're not able to view the UI). The Kubeflow Pipelines "
"UI should now be accessible at http://localhost:%d/.",
log_file_path,
port,
)
stop_k3d_cluster(cluster_name)
Stops a K3D cluster with the given name.
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def stop_k3d_cluster(cluster_name: str) -> None:
"""Stops a K3D cluster with the given name."""
subprocess.check_call(["k3d", "cluster", "stop", cluster_name])
logger.info("Stopped local k3d cluster '%s'.", cluster_name)
stop_kfp_ui_daemon(pid_file_path)
Stops the KFP UI daemon process if it is running.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pid_file_path |
str |
Path to the file with the daemons process ID. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def stop_kfp_ui_daemon(pid_file_path: str) -> None:
"""Stops the KFP UI daemon process if it is running.
Args:
pid_file_path: Path to the file with the daemons process ID.
"""
if fileio.exists(pid_file_path):
if sys.platform == "win32":
# Daemon functionality is not supported on Windows, so the PID
# file won't exist. This if clause exists just for mypy to not
# complain about missing functions
pass
else:
from zenml.utils import daemon
daemon.stop_daemon(pid_file_path)
fileio.remove(pid_file_path)
logger.info("Stopped Kubeflow Pipelines UI daemon.")
wait_until_kubeflow_pipelines_ready(kubernetes_context)
Waits until all Kubeflow Pipelines pods are ready.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
kubernetes_context |
str |
The kubernetes context in which the pods should be checked. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def wait_until_kubeflow_pipelines_ready(kubernetes_context: str) -> None:
"""Waits until all Kubeflow Pipelines pods are ready.
Args:
kubernetes_context: The kubernetes context in which the pods
should be checked.
"""
logger.info(
"Waiting for all Kubeflow Pipelines pods to be ready (this might "
"take a few minutes)."
)
while True:
logger.info("Current pod status:")
subprocess.check_call(
[
"kubectl",
"--context",
kubernetes_context,
"--namespace",
"kubeflow",
"get",
"pods",
]
)
if kubeflow_pipelines_ready(kubernetes_context=kubernetes_context):
break
logger.info("One or more pods not ready yet, waiting for 30 seconds...")
time.sleep(30)
write_local_registry_yaml(yaml_path, registry_name, registry_uri)
Writes a K3D registry config file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
yaml_path |
str |
Path where the config file should be written to. |
required |
registry_name |
str |
Name of the registry. |
required |
registry_uri |
str |
URI of the registry. |
required |
Source code in zenml/integrations/kubeflow/orchestrators/local_deployment_utils.py
def write_local_registry_yaml(
yaml_path: str, registry_name: str, registry_uri: str
) -> None:
"""Writes a K3D registry config file.
Args:
yaml_path: Path where the config file should be written to.
registry_name: Name of the registry.
registry_uri: URI of the registry.
"""
yaml_content = {
"mirrors": {registry_uri: {"endpoint": [f"http://{registry_name}"]}}
}
yaml_utils.write_yaml(yaml_path, yaml_content)
lightgbm
special
LightGBMIntegration (Integration)
Definition of lightgbm integration for ZenML.
Source code in zenml/integrations/lightgbm/__init__.py
class LightGBMIntegration(Integration):
"""Definition of lightgbm integration for ZenML."""
NAME = LIGHTGBM
REQUIREMENTS = ["lightgbm>=1.0.0"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.lightgbm import materializers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/lightgbm/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.lightgbm import materializers # noqa
materializers
special
lightgbm_booster_materializer
LightGBMBoosterMaterializer (BaseMaterializer)
Materializer to read data to and from lightgbm.Booster.
Source code in zenml/integrations/lightgbm/materializers/lightgbm_booster_materializer.py
class LightGBMBoosterMaterializer(BaseMaterializer):
"""Materializer to read data to and from lightgbm.Booster."""
ASSOCIATED_TYPES = (lgb.Booster,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> lgb.Booster:
"""Reads a lightgbm Booster model from a serialized JSON file."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
booster = lgb.Booster(model_file=temp_file)
# Cleanup and return
fileio.rmtree(temp_dir)
return booster
def handle_return(self, booster: lgb.Booster) -> None:
"""Creates a JSON serialization for a lightgbm Booster model.
Args:
booster: A lightgbm Booster model.
"""
super().handle_return(booster)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
with tempfile.NamedTemporaryFile(
mode="w", suffix=".txt", delete=True
) as f:
booster.save_model(f.name)
# Copy it into artifact store
fileio.copy(f.name, filepath)
handle_input(self, data_type)
Reads a lightgbm Booster model from a serialized JSON file.
Source code in zenml/integrations/lightgbm/materializers/lightgbm_booster_materializer.py
def handle_input(self, data_type: Type[Any]) -> lgb.Booster:
"""Reads a lightgbm Booster model from a serialized JSON file."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
booster = lgb.Booster(model_file=temp_file)
# Cleanup and return
fileio.rmtree(temp_dir)
return booster
handle_return(self, booster)
Creates a JSON serialization for a lightgbm Booster model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
booster |
Booster |
A lightgbm Booster model. |
required |
Source code in zenml/integrations/lightgbm/materializers/lightgbm_booster_materializer.py
def handle_return(self, booster: lgb.Booster) -> None:
"""Creates a JSON serialization for a lightgbm Booster model.
Args:
booster: A lightgbm Booster model.
"""
super().handle_return(booster)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
with tempfile.NamedTemporaryFile(
mode="w", suffix=".txt", delete=True
) as f:
booster.save_model(f.name)
# Copy it into artifact store
fileio.copy(f.name, filepath)
lightgbm_dataset_materializer
LightGBMDatasetMaterializer (BaseMaterializer)
Materializer to read data to and from lightgbm.Dataset
Source code in zenml/integrations/lightgbm/materializers/lightgbm_dataset_materializer.py
class LightGBMDatasetMaterializer(BaseMaterializer):
"""Materializer to read data to and from lightgbm.Dataset"""
ASSOCIATED_TYPES = (lgb.Dataset,)
ASSOCIATED_ARTIFACT_TYPES = (DataArtifact,)
def handle_input(self, data_type: Type[Any]) -> lgb.Dataset:
"""Reads a lightgbm.Dataset binary file and loads it."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
matrix = lgb.Dataset(temp_file, free_raw_data=False)
# No clean up this time because matrix is lazy loaded
return matrix
def handle_return(self, matrix: lgb.Dataset) -> None:
"""Creates a binary serialization for a lightgbm.Dataset object.
Args:
matrix: A lightgbm.Dataset object.
"""
super().handle_return(matrix)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
matrix.save_binary(temp_file)
# Copy it into artifact store
fileio.copy(temp_file, filepath)
fileio.rmtree(temp_dir)
handle_input(self, data_type)
Reads a lightgbm.Dataset binary file and loads it.
Source code in zenml/integrations/lightgbm/materializers/lightgbm_dataset_materializer.py
def handle_input(self, data_type: Type[Any]) -> lgb.Dataset:
"""Reads a lightgbm.Dataset binary file and loads it."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
matrix = lgb.Dataset(temp_file, free_raw_data=False)
# No clean up this time because matrix is lazy loaded
return matrix
handle_return(self, matrix)
Creates a binary serialization for a lightgbm.Dataset object.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
matrix |
Dataset |
A lightgbm.Dataset object. |
required |
Source code in zenml/integrations/lightgbm/materializers/lightgbm_dataset_materializer.py
def handle_return(self, matrix: lgb.Dataset) -> None:
"""Creates a binary serialization for a lightgbm.Dataset object.
Args:
matrix: A lightgbm.Dataset object.
"""
super().handle_return(matrix)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
matrix.save_binary(temp_file)
# Copy it into artifact store
fileio.copy(temp_file, filepath)
fileio.rmtree(temp_dir)
mlflow
special
The mlflow integrations currently enables you to use mlflow tracking as a convenient way to visualize your experiment runs within the mlflow ui
MlflowIntegration (Integration)
Definition of MLflow integration for ZenML.
Source code in zenml/integrations/mlflow/__init__.py
class MlflowIntegration(Integration):
"""Definition of MLflow integration for ZenML."""
NAME = MLFLOW
REQUIREMENTS = [
"mlflow>=1.2.0",
"mlserver>=0.5.3",
"mlserver-mlflow>=0.5.3",
]
@staticmethod
def activate() -> None:
"""Activate the MLflow integration."""
from zenml.integrations.mlflow import experiment_trackers # noqa
from zenml.integrations.mlflow import model_deployers # noqa
from zenml.integrations.mlflow import services # noqa
activate()
staticmethod
Activate the MLflow integration.
Source code in zenml/integrations/mlflow/__init__.py
@staticmethod
def activate() -> None:
"""Activate the MLflow integration."""
from zenml.integrations.mlflow import experiment_trackers # noqa
from zenml.integrations.mlflow import model_deployers # noqa
from zenml.integrations.mlflow import services # noqa
experiment_trackers
special
mlflow_experiment_tracker
MLFlowExperimentTracker (BaseExperimentTracker)
pydantic-model
Stores Mlflow configuration options.
ZenML should take care of configuring MLflow for you, but should you still need access to the configuration inside your step you can do it using a step context:
from zenml.steps import StepContext
@enable_mlflow
@step
def my_step(context: StepContext, ...)
context.stack.experiment_tracker # get the tracking_uri etc. from here
Attributes:
| Name | Type | Description |
|---|---|---|
tracking_uri |
Optional[str] |
The uri of the mlflow tracking server. If no uri is set,
your stack must contain a |
tracking_username |
Optional[str] |
Username for authenticating with the MLflow
tracking server. When a remote tracking uri is specified,
either |
tracking_password |
Optional[str] |
Password for authenticating with the MLflow
tracking server. When a remote tracking uri is specified,
either |
tracking_token |
Optional[str] |
Token for authenticating with the MLflow
tracking server. When a remote tracking uri is specified,
either |
tracking_insecure_tls |
bool |
Skips verification of TLS connection to the
MLflow tracking server if set to |
Source code in zenml/integrations/mlflow/experiment_trackers/mlflow_experiment_tracker.py
class MLFlowExperimentTracker(BaseExperimentTracker):
"""Stores Mlflow configuration options.
ZenML should take care of configuring MLflow for you, but should you still
need access to the configuration inside your step you can do it using a
step context:
```python
from zenml.steps import StepContext
@enable_mlflow
@step
def my_step(context: StepContext, ...)
context.stack.experiment_tracker # get the tracking_uri etc. from here
```
Attributes:
tracking_uri: The uri of the mlflow tracking server. If no uri is set,
your stack must contain a `LocalArtifactStore` and ZenML will
point MLflow to a subdirectory of your artifact store instead.
tracking_username: Username for authenticating with the MLflow
tracking server. When a remote tracking uri is specified,
either `tracking_token` or `tracking_username` and
`tracking_password` must be specified.
tracking_password: Password for authenticating with the MLflow
tracking server. When a remote tracking uri is specified,
either `tracking_token` or `tracking_username` and
`tracking_password` must be specified.
tracking_token: Token for authenticating with the MLflow
tracking server. When a remote tracking uri is specified,
either `tracking_token` or `tracking_username` and
`tracking_password` must be specified.
tracking_insecure_tls: Skips verification of TLS connection to the
MLflow tracking server if set to `True`.
"""
tracking_uri: Optional[str] = None
tracking_username: Optional[str] = None
tracking_password: Optional[str] = None
tracking_token: Optional[str] = None
tracking_insecure_tls: bool = False
# Class Configuration
FLAVOR: ClassVar[str] = MLFLOW
@validator("tracking_uri")
def _ensure_valid_tracking_uri(
cls, tracking_uri: Optional[str] = None
) -> Optional[str]:
"""Ensures that the tracking uri is a valid mlflow tracking uri."""
if tracking_uri:
valid_schemes = DATABASE_ENGINES + ["http", "https", "file"]
if not any(
tracking_uri.startswith(scheme) for scheme in valid_schemes
):
raise ValueError(
f"MLflow tracking uri does not start with one of the valid "
f"schemes {valid_schemes}. See "
f"https://www.mlflow.org/docs/latest/tracking.html#where-runs-are-recorded "
f"for more information."
)
return tracking_uri
@root_validator
def _ensure_authentication_if_necessary(
cls, values: Dict[str, Any]
) -> Dict[str, Any]:
"""Ensures that credentials or a token for authentication exist when
running MLflow tracking with a remote backend."""
tracking_uri = values.get("tracking_uri")
if tracking_uri and cls.is_remote_tracking_uri(tracking_uri):
# we need either username + password or a token to authenticate to
# the remote backend
basic_auth = values.get("tracking_username") and values.get(
"tracking_password"
)
token_auth = values.get("tracking_token")
if not (basic_auth or token_auth):
raise ValueError(
f"MLflow experiment tracking with a remote backend "
f"{tracking_uri} is only possible when specifying either "
f"username and password or an authentication token in your "
f"stack component. To update your component, run the "
f"following command: `zenml experiment-tracker update "
f"{values['name']} --tracking_username=MY_USERNAME "
f"--tracking_password=MY_PASSWORD "
f"--tracking_token=MY_TOKEN` and specify either your "
f"username and password or token."
)
return values
@staticmethod
def is_remote_tracking_uri(tracking_uri: str) -> bool:
"""Checks whether the given tracking uri is remote or not."""
return any(
tracking_uri.startswith(prefix)
for prefix in ["http://", "https://"]
)
@staticmethod
def _local_mlflow_backend() -> str:
"""Returns the local MLflow backend inside the ZenML artifact
repository directory
Returns:
The MLflow tracking URI for the local MLflow backend.
"""
repo = Repository(skip_repository_check=True) # type: ignore[call-arg]
artifact_store = repo.active_stack.artifact_store
local_mlflow_backend_uri = os.path.join(artifact_store.path, "mlruns")
if not os.path.exists(local_mlflow_backend_uri):
os.makedirs(local_mlflow_backend_uri)
return "file:" + local_mlflow_backend_uri
def get_tracking_uri(self) -> str:
"""Returns the configured tracking URI or a local fallback."""
return self.tracking_uri or self._local_mlflow_backend()
def configure_mlflow(self) -> None:
"""Configures the MLflow tracking URI and any additional credentials."""
mlflow.set_tracking_uri(self.get_tracking_uri())
if self.tracking_username:
os.environ[MLFLOW_TRACKING_USERNAME] = self.tracking_username
if self.tracking_password:
os.environ[MLFLOW_TRACKING_PASSWORD] = self.tracking_password
if self.tracking_token:
os.environ[MLFLOW_TRACKING_TOKEN] = self.tracking_token
os.environ[MLFLOW_TRACKING_INSECURE_TLS] = (
"true" if self.tracking_insecure_tls else "false"
)
def prepare_step_run(self) -> None:
"""Sets the MLflow tracking uri and credentials."""
self.configure_mlflow()
def cleanup_step_run(self) -> None:
"""Resets the MLflow tracking uri."""
mlflow.set_tracking_uri("")
@property
def validator(self) -> Optional["StackValidator"]:
"""Validates that the stack has a `LocalArtifactStore` if no tracking
uri was specified."""
if self.tracking_uri:
# user specified a tracking uri, do nothing
return None
else:
# try to fall back to a tracking uri inside the zenml artifact
# store. this only works in case of a local artifact store, so we
# make sure to prevent stack with other artifact stores for now
return StackValidator(
custom_validation_function=lambda stack: (
isinstance(stack.artifact_store, LocalArtifactStore),
"MLflow experiment tracker without a specified tracking "
"uri only works with a local artifact store.",
)
)
@property
def active_experiment(self) -> Optional[Experiment]:
"""Returns the currently active MLflow experiment."""
step_env = Environment().step_environment
if not step_env:
# we're not inside a step
return None
mlflow.set_experiment(experiment_name=step_env.pipeline_name)
return mlflow.get_experiment_by_name(step_env.pipeline_name)
@property
def active_run(self) -> Optional[mlflow.ActiveRun]:
"""Returns the currently active MLflow run."""
step_env = Environment().step_environment
if not self.active_experiment or not step_env:
return None
experiment_id = self.active_experiment.experiment_id
# TODO [ENG-458]: find a solution to avoid race-conditions while
# creating the same MLflow run from parallel steps
runs = mlflow.search_runs(
experiment_ids=[experiment_id],
filter_string=f'tags.mlflow.runName = "{step_env.pipeline_run_id}"',
output_format="list",
)
run_id = runs[0].info.run_id if runs else None
current_active_run = mlflow.active_run()
if current_active_run and current_active_run.info.run_id == run_id:
return current_active_run
return mlflow.start_run(
run_id=run_id,
run_name=step_env.pipeline_run_id,
experiment_id=experiment_id,
)
active_experiment: Optional[mlflow.entities.experiment.Experiment]
property
readonly
Returns the currently active MLflow experiment.
active_run: Optional[mlflow.tracking.fluent.ActiveRun]
property
readonly
Returns the currently active MLflow run.
validator: Optional[StackValidator]
property
readonly
Validates that the stack has a LocalArtifactStore if no tracking
uri was specified.
cleanup_step_run(self)
Resets the MLflow tracking uri.
Source code in zenml/integrations/mlflow/experiment_trackers/mlflow_experiment_tracker.py
def cleanup_step_run(self) -> None:
"""Resets the MLflow tracking uri."""
mlflow.set_tracking_uri("")
configure_mlflow(self)
Configures the MLflow tracking URI and any additional credentials.
Source code in zenml/integrations/mlflow/experiment_trackers/mlflow_experiment_tracker.py
def configure_mlflow(self) -> None:
"""Configures the MLflow tracking URI and any additional credentials."""
mlflow.set_tracking_uri(self.get_tracking_uri())
if self.tracking_username:
os.environ[MLFLOW_TRACKING_USERNAME] = self.tracking_username
if self.tracking_password:
os.environ[MLFLOW_TRACKING_PASSWORD] = self.tracking_password
if self.tracking_token:
os.environ[MLFLOW_TRACKING_TOKEN] = self.tracking_token
os.environ[MLFLOW_TRACKING_INSECURE_TLS] = (
"true" if self.tracking_insecure_tls else "false"
)
get_tracking_uri(self)
Returns the configured tracking URI or a local fallback.
Source code in zenml/integrations/mlflow/experiment_trackers/mlflow_experiment_tracker.py
def get_tracking_uri(self) -> str:
"""Returns the configured tracking URI or a local fallback."""
return self.tracking_uri or self._local_mlflow_backend()
is_remote_tracking_uri(tracking_uri)
staticmethod
Checks whether the given tracking uri is remote or not.
Source code in zenml/integrations/mlflow/experiment_trackers/mlflow_experiment_tracker.py
@staticmethod
def is_remote_tracking_uri(tracking_uri: str) -> bool:
"""Checks whether the given tracking uri is remote or not."""
return any(
tracking_uri.startswith(prefix)
for prefix in ["http://", "https://"]
)
prepare_step_run(self)
Sets the MLflow tracking uri and credentials.
Source code in zenml/integrations/mlflow/experiment_trackers/mlflow_experiment_tracker.py
def prepare_step_run(self) -> None:
"""Sets the MLflow tracking uri and credentials."""
self.configure_mlflow()
mlflow_step_decorator
enable_mlflow(_step=None)
Decorator to enable mlflow for a step function.
Apply this decorator to a ZenML pipeline step to enable MLflow experiment
tracking. The MLflow tracking configuration (tracking URI, experiment name,
run name) will be automatically configured before the step code is executed,
so the step can simply use the mlflow module to log metrics and artifacts,
like so:
@enable_mlflow
@step
def tf_evaluator(
x_test: np.ndarray,
y_test: np.ndarray,
model: tf.keras.Model,
) -> float:
_, test_acc = model.evaluate(x_test, y_test, verbose=2)
mlflow.log_metric("val_accuracy", test_acc)
return test_acc
All MLflow artifacts and metrics logged from all the steps in a pipeline
run are by default grouped under a single experiment named after the
pipeline. To log MLflow artifacts and metrics from a step in a separate
MLflow experiment, pass a custom experiment_name argument value to the
decorator.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
_step |
Optional[~S] |
The decorated step class. |
None |
Returns:
| Type | Description |
|---|---|
Union[~S, Callable[[~S], ~S]] |
The inner decorator which enhaces the input step class with mlflow tracking functionality |
Source code in zenml/integrations/mlflow/mlflow_step_decorator.py
def enable_mlflow(
_step: Optional[S] = None,
) -> Union[S, Callable[[S], S]]:
"""Decorator to enable mlflow for a step function.
Apply this decorator to a ZenML pipeline step to enable MLflow experiment
tracking. The MLflow tracking configuration (tracking URI, experiment name,
run name) will be automatically configured before the step code is executed,
so the step can simply use the `mlflow` module to log metrics and artifacts,
like so:
```python
@enable_mlflow
@step
def tf_evaluator(
x_test: np.ndarray,
y_test: np.ndarray,
model: tf.keras.Model,
) -> float:
_, test_acc = model.evaluate(x_test, y_test, verbose=2)
mlflow.log_metric("val_accuracy", test_acc)
return test_acc
```
All MLflow artifacts and metrics logged from all the steps in a pipeline
run are by default grouped under a single experiment named after the
pipeline. To log MLflow artifacts and metrics from a step in a separate
MLflow experiment, pass a custom `experiment_name` argument value to the
decorator.
Args:
_step: The decorated step class.
Returns:
The inner decorator which enhaces the input step class with mlflow
tracking functionality
"""
def inner_decorator(_step: S) -> S:
logger.debug(
"Applying 'enable_mlflow' decorator to step %s", _step.__name__
)
if not issubclass(_step, BaseStep):
raise RuntimeError(
"The `enable_mlflow` decorator can only be applied to a ZenML "
"`step` decorated function or a BaseStep subclass."
)
source_fn = getattr(_step, STEP_INNER_FUNC_NAME)
return cast(
S,
type( # noqa
_step.__name__,
(_step,),
{
STEP_INNER_FUNC_NAME: staticmethod(
mlflow_step_entrypoint()(source_fn)
),
"__module__": _step.__module__,
},
),
)
if _step is None:
return inner_decorator
else:
return inner_decorator(_step)
mlflow_step_entrypoint()
Decorator for a step entrypoint to enable mlflow.
Returns:
| Type | Description |
|---|---|
Callable[[~F], ~F] |
the input function enhanced with mlflow profiling functionality |
Source code in zenml/integrations/mlflow/mlflow_step_decorator.py
def mlflow_step_entrypoint() -> Callable[[F], F]:
"""Decorator for a step entrypoint to enable mlflow.
Returns:
the input function enhanced with mlflow profiling functionality
"""
def inner_decorator(func: F) -> F:
logger.debug(
"Applying 'mlflow_step_entrypoint' decorator to step entrypoint %s",
func.__name__,
)
@functools.wraps(func)
def wrapper(*args: Any, **kwargs: Any) -> Any: # noqa
logger.debug(
"Setting up MLflow backend before running step entrypoint %s",
func.__name__,
)
experiment_tracker = Repository( # type: ignore[call-arg]
skip_repository_check=True
).active_stack.experiment_tracker
if not isinstance(experiment_tracker, MLFlowExperimentTracker):
raise get_missing_mlflow_experiment_tracker_error()
active_run = experiment_tracker.active_run
if not active_run:
raise RuntimeError("No active mlflow run configued.")
with active_run:
return func(*args, **kwargs)
return cast(F, wrapper)
return inner_decorator
mlflow_utils
get_missing_mlflow_experiment_tracker_error()
Returns a detailed error that describes how to add an MLflow experiment tracker component to your stack.
Source code in zenml/integrations/mlflow/mlflow_utils.py
def get_missing_mlflow_experiment_tracker_error() -> ValueError:
"""Returns a detailed error that describes how to add an MLflow experiment
tracker component to your stack."""
return ValueError(
"The active stack needs to have a MLflow experiment tracker "
"component registered to be able to track experiments using "
"MLflow. You can create a new stack with a MLflow experiment "
"tracker component or update your existing stack to add this "
"component, e.g.:\n\n"
" 'zenml experiment-tracker register mlflow_tracker "
"--type=mlflow'\n"
" 'zenml stack register stack-name -e mlflow_tracker ...'\n"
)
get_tracking_uri()
Gets the MLflow tracking URI from the active experiment tracking stack component.
Returns:
| Type | Description |
|---|---|
str |
MLflow tracking URI. |
Exceptions:
| Type | Description |
|---|---|
ValueError |
If the active stack contains no MLflow experiment tracking component. |
Source code in zenml/integrations/mlflow/mlflow_utils.py
def get_tracking_uri() -> str:
"""Gets the MLflow tracking URI from the active experiment tracking stack
component.
Returns:
MLflow tracking URI.
Raises:
ValueError: If the active stack contains no MLflow experiment tracking
component.
"""
tracker = Repository().active_stack.experiment_tracker
if tracker is None or not isinstance(tracker, MLFlowExperimentTracker):
raise get_missing_mlflow_experiment_tracker_error()
return tracker.get_tracking_uri()
model_deployers
special
mlflow_model_deployer
MLFlowModelDeployer (BaseModelDeployer)
pydantic-model
MLflow implementation of the BaseModelDeployer
Attributes:
| Name | Type | Description |
|---|---|---|
service_path |
str |
the path where the local MLflow deployment service |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
class MLFlowModelDeployer(BaseModelDeployer):
"""MLflow implementation of the BaseModelDeployer
Attributes:
service_path: the path where the local MLflow deployment service
configuration, PID and log files are stored.
"""
service_path: str = ""
# Class Configuration
FLAVOR: ClassVar[str] = MLFLOW
@root_validator
def set_service_path(cls, values: Dict[str, Any]) -> Dict[str, Any]:
"""Sets the service_path attribute value according to the component
UUID."""
if values.get("service_path"):
return values
# not likely to happen, due to Pydantic validation, but mypy complains
assert "uuid" in values
values["service_path"] = cls.get_service_path(values["uuid"])
return values
@staticmethod
def get_service_path(uuid: uuid.UUID) -> str:
"""Get the path the path where the local MLflow deployment service
configuration, PID and log files are stored.
Args:
uuid: The UUID of the MLflow model deployer.
Returns:
The service path.
"""
service_path = os.path.join(
get_global_config_directory(),
LOCAL_STORES_DIRECTORY_NAME,
str(uuid),
)
create_dir_recursive_if_not_exists(service_path)
return service_path
@property
def local_path(self) -> str:
"""
Returns the path to the root directory where all configurations for
MLflow deployment daemon processes are stored.
Returns:
The path to the local service root directory.
"""
return self.service_path
@staticmethod
def get_model_server_info( # type: ignore[override]
service_instance: "MLFlowDeploymentService",
) -> Dict[str, Optional[str]]:
"""Return implementation specific information that might be relevant
to the user.
Args:
service_instance: Instance of a SeldonDeploymentService
"""
return {
"PREDICTION_URL": service_instance.endpoint.prediction_url,
"MODEL_URI": service_instance.config.model_uri,
"MODEL_NAME": service_instance.config.model_name,
"SERVICE_PATH": service_instance.status.runtime_path,
"DAEMON_PID": str(service_instance.status.pid),
}
@staticmethod
def get_active_model_deployer() -> "MLFlowModelDeployer":
"""
Returns the MLFlowModelDeployer component of the active stack.
Args:
None
Returns:
The MLFlowModelDeployer component of the active stack.
"""
model_deployer = Repository( # type: ignore[call-arg]
skip_repository_check=True
).active_stack.model_deployer
if not model_deployer or not isinstance(
model_deployer, MLFlowModelDeployer
):
raise TypeError(
"The active stack needs to have an MLflow model deployer "
"component registered to be able to deploy models with MLflow. "
"You can create a new stack with an MLflow model "
"deployer component or update your existing stack to add this "
"component, e.g.:\n\n"
" 'zenml model-deployer register mlflow --type=mlflow'\n"
" 'zenml stack create stack-name -d mlflow ...'\n"
)
return model_deployer
def deploy_model(
self,
config: ServiceConfig,
replace: bool = False,
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT,
) -> BaseService:
"""Create a new MLflow deployment service or update an existing one to
serve the supplied model and deployment configuration.
This method has two modes of operation, depending on the `replace`
argument value:
* if `replace` is False, calling this method will create a new MLflow
deployment server to reflect the model and other configuration
parameters specified in the supplied MLflow service `config`.
* if `replace` is True, this method will first attempt to find an
existing MLflow deployment service that is *equivalent* to the
supplied configuration parameters. Two or more MLflow deployment
services are considered equivalent if they have the same
`pipeline_name`, `pipeline_step_name` and `model_name` configuration
parameters. To put it differently, two MLflow deployment services
are equivalent if they serve versions of the same model deployed by
the same pipeline step. If an equivalent MLflow deployment is found,
it will be updated in place to reflect the new configuration
parameters.
Callers should set `replace` to True if they want a continuous model
deployment workflow that doesn't spin up a new MLflow deployment
server for each new model version. If multiple equivalent MLflow
deployment servers are found, one is selected at random to be updated
and the others are deleted.
Args:
config: the configuration of the model to be deployed with MLflow.
replace: set this flag to True to find and update an equivalent
MLflow deployment server with the new model instead of
creating and starting a new deployment server.
timeout: the timeout in seconds to wait for the MLflow server
to be provisioned and successfully started or updated. If set
to 0, the method will return immediately after the MLflow
server is provisioned, without waiting for it to fully start.
Returns:
The ZenML MLflow deployment service object that can be used to
interact with the MLflow model server.
Raises:
RuntimeError: if `timeout` is set to a positive value that is
exceeded while waiting for the MLflow deployment server
to start, or if an operational failure is encountered before
it reaches a ready state.
"""
config = cast(MLFlowDeploymentConfig, config)
service = None
# if replace is True, remove all existing services
if replace is True:
existing_services = self.find_model_server(
pipeline_name=config.pipeline_name,
pipeline_step_name=config.pipeline_step_name,
model_name=config.model_name,
)
for existing_service in existing_services:
if service is None:
# keep the most recently created service
service = cast(MLFlowDeploymentService, existing_service)
try:
# delete the older services and don't wait for them to
# be deprovisioned
self._clean_up_existing_service(
existing_service=cast(
MLFlowDeploymentService, existing_service
),
timeout=timeout,
force=True,
)
except RuntimeError:
# ignore errors encountered while stopping old services
pass
if service:
logger.info(
f"Updating an existing MLflow deployment service: {service}"
)
# set the root runtime path with the stack component's UUID
config.root_runtime_path = self.local_path
service.stop(timeout=timeout, force=True)
service.update(config)
service.start(timeout=timeout)
else:
# create a new MLFlowDeploymentService instance
service = self._create_new_service(timeout, config)
logger.info(f"Created a new MLflow deployment service: {service}")
return cast(BaseService, service)
def _clean_up_existing_service(
self,
timeout: int,
force: bool,
existing_service: MLFlowDeploymentService,
) -> None:
# stop the older service
existing_service.stop(timeout=timeout, force=force)
# delete the old configuration file
service_directory_path = existing_service.status.runtime_path or ""
shutil.rmtree(service_directory_path)
# the step will receive a config from the user that mentions the number
# of workers etc.the step implementation will create a new config using
# all values from the user and add values like pipeline name, model_uri
def _create_new_service(
self, timeout: int, config: MLFlowDeploymentConfig
) -> MLFlowDeploymentService:
"""Creates a new MLFlowDeploymentService."""
# set the root runtime path with the stack component's UUID
config.root_runtime_path = self.local_path
# create a new service for the new model
service = MLFlowDeploymentService(config)
service.start(timeout=timeout)
return service
def find_model_server(
self,
running: bool = False,
service_uuid: Optional[UUID] = None,
pipeline_name: Optional[str] = None,
pipeline_run_id: Optional[str] = None,
pipeline_step_name: Optional[str] = None,
model_name: Optional[str] = None,
model_uri: Optional[str] = None,
model_type: Optional[str] = None,
) -> List[BaseService]:
"""Method to find one or more model servers that match the
given criteria.
Args:
running: If true, only running services will be returned.
service_uuid: The UUID of the service that was originally used
to deploy the model.
pipeline_name: Name of the pipeline that the deployed model was part
of.
pipeline_run_id: ID of the pipeline run which the deployed model was part of.
pipeline_step_name: The name of the pipeline model deployment step that
deployed the model.
model_name: Name of the deployed model.
model_uri: URI of the deployed model.
model_type: Type/format of the deployed model. Not used in this MLflow
case.
Returns:
One or more Service objects representing model servers that match
the input search criteria.
"""
services = []
config = MLFlowDeploymentConfig(
model_name=model_name or "",
model_uri=model_uri or "",
pipeline_name=pipeline_name or "",
pipeline_run_id=pipeline_run_id or "",
pipeline_step_name=pipeline_step_name or "",
)
# find all services that match the input criteria
for root, _, files in os.walk(self.local_path):
if service_uuid and Path(root).name != str(service_uuid):
continue
for file in files:
if file == SERVICE_DAEMON_CONFIG_FILE_NAME:
service_config_path = os.path.join(root, file)
logger.debug(
"Loading service daemon configuration from %s",
service_config_path,
)
existing_service_config = None
with open(service_config_path, "r") as f:
existing_service_config = f.read()
existing_service = ServiceRegistry().load_service_from_json(
existing_service_config
)
if not isinstance(
existing_service, MLFlowDeploymentService
):
raise TypeError(
f"Expected service type MLFlowDeploymentService but got "
f"{type(existing_service)} instead"
)
existing_service.update_status()
if self._matches_search_criteria(existing_service, config):
if not running or existing_service.is_running:
services.append(cast(BaseService, existing_service))
return services
def _matches_search_criteria(
self,
existing_service: MLFlowDeploymentService,
config: MLFlowDeploymentConfig,
) -> bool:
"""Returns true if a service matches the input criteria. If any of the values in
the input criteria are None, they are ignored. This allows listing services
just by common pipeline names or step names, etc.
Args:
existing_service: The materialized Service instance derived from the config
of the older (existing) service
config: The MLFlowDeploymentConfig object passed to the deploy_model function holding
parameters of the new service to be created."""
existing_service_config = existing_service.config
# check if the existing service matches the input criteria
if (
(
not config.pipeline_name
or existing_service_config.pipeline_name == config.pipeline_name
)
and (
not config.model_name
or existing_service_config.model_name == config.model_name
)
and (
not config.pipeline_step_name
or existing_service_config.pipeline_step_name
== config.pipeline_step_name
)
and (
not config.pipeline_run_id
or existing_service_config.pipeline_run_id
== config.pipeline_run_id
)
):
return True
return False
def stop_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Method to stop a model server.
Args:
uuid: UUID of the model server to stop.
timeout: Timeout in seconds to wait for the service to stop.
force: If True, force the service to stop.
"""
# get list of all services
existing_services = self.find_model_server(service_uuid=uuid)
# if the service exists, stop it
if existing_services:
existing_services[0].stop(timeout=timeout, force=force)
def start_model_server(
self, uuid: UUID, timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT
) -> None:
"""Method to start a model server.
Args:
uuid: UUID of the model server to start.
"""
# get list of all services
existing_services = self.find_model_server(service_uuid=uuid)
# if the service exists, start it
if existing_services:
existing_services[0].start(timeout=timeout)
def delete_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Method to delete all configuration of a model server.
Args:
uuid: UUID of the model server to delete.
"""
# get list of all services
existing_services = self.find_model_server(service_uuid=uuid)
# if the service exists, clean it up
if existing_services:
service = cast(MLFlowDeploymentService, existing_services[0])
self._clean_up_existing_service(
existing_service=service, timeout=timeout, force=force
)
local_path: str
property
readonly
Returns the path to the root directory where all configurations for MLflow deployment daemon processes are stored.
Returns:
| Type | Description |
|---|---|
str |
The path to the local service root directory. |
delete_model_server(self, uuid, timeout=10, force=False)
Method to delete all configuration of a model server.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
uuid |
UUID |
UUID of the model server to delete. |
required |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
def delete_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Method to delete all configuration of a model server.
Args:
uuid: UUID of the model server to delete.
"""
# get list of all services
existing_services = self.find_model_server(service_uuid=uuid)
# if the service exists, clean it up
if existing_services:
service = cast(MLFlowDeploymentService, existing_services[0])
self._clean_up_existing_service(
existing_service=service, timeout=timeout, force=force
)
deploy_model(self, config, replace=False, timeout=10)
Create a new MLflow deployment service or update an existing one to serve the supplied model and deployment configuration.
This method has two modes of operation, depending on the replace
argument value:
-
if
replaceis False, calling this method will create a new MLflow deployment server to reflect the model and other configuration parameters specified in the supplied MLflow serviceconfig. -
if
replaceis True, this method will first attempt to find an existing MLflow deployment service that is equivalent to the supplied configuration parameters. Two or more MLflow deployment services are considered equivalent if they have the samepipeline_name,pipeline_step_nameandmodel_nameconfiguration parameters. To put it differently, two MLflow deployment services are equivalent if they serve versions of the same model deployed by the same pipeline step. If an equivalent MLflow deployment is found, it will be updated in place to reflect the new configuration parameters.
Callers should set replace to True if they want a continuous model
deployment workflow that doesn't spin up a new MLflow deployment
server for each new model version. If multiple equivalent MLflow
deployment servers are found, one is selected at random to be updated
and the others are deleted.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
config |
ServiceConfig |
the configuration of the model to be deployed with MLflow. |
required |
replace |
bool |
set this flag to True to find and update an equivalent MLflow deployment server with the new model instead of creating and starting a new deployment server. |
False |
timeout |
int |
the timeout in seconds to wait for the MLflow server to be provisioned and successfully started or updated. If set to 0, the method will return immediately after the MLflow server is provisioned, without waiting for it to fully start. |
10 |
Returns:
| Type | Description |
|---|---|
BaseService |
The ZenML MLflow deployment service object that can be used to interact with the MLflow model server. |
Exceptions:
| Type | Description |
|---|---|
RuntimeError |
if |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
def deploy_model(
self,
config: ServiceConfig,
replace: bool = False,
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT,
) -> BaseService:
"""Create a new MLflow deployment service or update an existing one to
serve the supplied model and deployment configuration.
This method has two modes of operation, depending on the `replace`
argument value:
* if `replace` is False, calling this method will create a new MLflow
deployment server to reflect the model and other configuration
parameters specified in the supplied MLflow service `config`.
* if `replace` is True, this method will first attempt to find an
existing MLflow deployment service that is *equivalent* to the
supplied configuration parameters. Two or more MLflow deployment
services are considered equivalent if they have the same
`pipeline_name`, `pipeline_step_name` and `model_name` configuration
parameters. To put it differently, two MLflow deployment services
are equivalent if they serve versions of the same model deployed by
the same pipeline step. If an equivalent MLflow deployment is found,
it will be updated in place to reflect the new configuration
parameters.
Callers should set `replace` to True if they want a continuous model
deployment workflow that doesn't spin up a new MLflow deployment
server for each new model version. If multiple equivalent MLflow
deployment servers are found, one is selected at random to be updated
and the others are deleted.
Args:
config: the configuration of the model to be deployed with MLflow.
replace: set this flag to True to find and update an equivalent
MLflow deployment server with the new model instead of
creating and starting a new deployment server.
timeout: the timeout in seconds to wait for the MLflow server
to be provisioned and successfully started or updated. If set
to 0, the method will return immediately after the MLflow
server is provisioned, without waiting for it to fully start.
Returns:
The ZenML MLflow deployment service object that can be used to
interact with the MLflow model server.
Raises:
RuntimeError: if `timeout` is set to a positive value that is
exceeded while waiting for the MLflow deployment server
to start, or if an operational failure is encountered before
it reaches a ready state.
"""
config = cast(MLFlowDeploymentConfig, config)
service = None
# if replace is True, remove all existing services
if replace is True:
existing_services = self.find_model_server(
pipeline_name=config.pipeline_name,
pipeline_step_name=config.pipeline_step_name,
model_name=config.model_name,
)
for existing_service in existing_services:
if service is None:
# keep the most recently created service
service = cast(MLFlowDeploymentService, existing_service)
try:
# delete the older services and don't wait for them to
# be deprovisioned
self._clean_up_existing_service(
existing_service=cast(
MLFlowDeploymentService, existing_service
),
timeout=timeout,
force=True,
)
except RuntimeError:
# ignore errors encountered while stopping old services
pass
if service:
logger.info(
f"Updating an existing MLflow deployment service: {service}"
)
# set the root runtime path with the stack component's UUID
config.root_runtime_path = self.local_path
service.stop(timeout=timeout, force=True)
service.update(config)
service.start(timeout=timeout)
else:
# create a new MLFlowDeploymentService instance
service = self._create_new_service(timeout, config)
logger.info(f"Created a new MLflow deployment service: {service}")
return cast(BaseService, service)
find_model_server(self, running=False, service_uuid=None, pipeline_name=None, pipeline_run_id=None, pipeline_step_name=None, model_name=None, model_uri=None, model_type=None)
Method to find one or more model servers that match the given criteria.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
running |
bool |
If true, only running services will be returned. |
False |
service_uuid |
Optional[uuid.UUID] |
The UUID of the service that was originally used to deploy the model. |
None |
pipeline_name |
Optional[str] |
Name of the pipeline that the deployed model was part |
None |
pipeline_run_id |
Optional[str] |
ID of the pipeline run which the deployed model was part of. |
None |
pipeline_step_name |
Optional[str] |
The name of the pipeline model deployment step that deployed the model. |
None |
model_name |
Optional[str] |
Name of the deployed model. |
None |
model_uri |
Optional[str] |
URI of the deployed model. |
None |
model_type |
Optional[str] |
Type/format of the deployed model. Not used in this MLflow case. |
None |
Returns:
| Type | Description |
|---|---|
List[zenml.services.service.BaseService] |
One or more Service objects representing model servers that match the input search criteria. |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
def find_model_server(
self,
running: bool = False,
service_uuid: Optional[UUID] = None,
pipeline_name: Optional[str] = None,
pipeline_run_id: Optional[str] = None,
pipeline_step_name: Optional[str] = None,
model_name: Optional[str] = None,
model_uri: Optional[str] = None,
model_type: Optional[str] = None,
) -> List[BaseService]:
"""Method to find one or more model servers that match the
given criteria.
Args:
running: If true, only running services will be returned.
service_uuid: The UUID of the service that was originally used
to deploy the model.
pipeline_name: Name of the pipeline that the deployed model was part
of.
pipeline_run_id: ID of the pipeline run which the deployed model was part of.
pipeline_step_name: The name of the pipeline model deployment step that
deployed the model.
model_name: Name of the deployed model.
model_uri: URI of the deployed model.
model_type: Type/format of the deployed model. Not used in this MLflow
case.
Returns:
One or more Service objects representing model servers that match
the input search criteria.
"""
services = []
config = MLFlowDeploymentConfig(
model_name=model_name or "",
model_uri=model_uri or "",
pipeline_name=pipeline_name or "",
pipeline_run_id=pipeline_run_id or "",
pipeline_step_name=pipeline_step_name or "",
)
# find all services that match the input criteria
for root, _, files in os.walk(self.local_path):
if service_uuid and Path(root).name != str(service_uuid):
continue
for file in files:
if file == SERVICE_DAEMON_CONFIG_FILE_NAME:
service_config_path = os.path.join(root, file)
logger.debug(
"Loading service daemon configuration from %s",
service_config_path,
)
existing_service_config = None
with open(service_config_path, "r") as f:
existing_service_config = f.read()
existing_service = ServiceRegistry().load_service_from_json(
existing_service_config
)
if not isinstance(
existing_service, MLFlowDeploymentService
):
raise TypeError(
f"Expected service type MLFlowDeploymentService but got "
f"{type(existing_service)} instead"
)
existing_service.update_status()
if self._matches_search_criteria(existing_service, config):
if not running or existing_service.is_running:
services.append(cast(BaseService, existing_service))
return services
get_active_model_deployer()
staticmethod
Returns the MLFlowModelDeployer component of the active stack.
Returns:
| Type | Description |
|---|---|
MLFlowModelDeployer |
The MLFlowModelDeployer component of the active stack. |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
@staticmethod
def get_active_model_deployer() -> "MLFlowModelDeployer":
"""
Returns the MLFlowModelDeployer component of the active stack.
Args:
None
Returns:
The MLFlowModelDeployer component of the active stack.
"""
model_deployer = Repository( # type: ignore[call-arg]
skip_repository_check=True
).active_stack.model_deployer
if not model_deployer or not isinstance(
model_deployer, MLFlowModelDeployer
):
raise TypeError(
"The active stack needs to have an MLflow model deployer "
"component registered to be able to deploy models with MLflow. "
"You can create a new stack with an MLflow model "
"deployer component or update your existing stack to add this "
"component, e.g.:\n\n"
" 'zenml model-deployer register mlflow --type=mlflow'\n"
" 'zenml stack create stack-name -d mlflow ...'\n"
)
return model_deployer
get_model_server_info(service_instance)
staticmethod
Return implementation specific information that might be relevant to the user.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
service_instance |
MLFlowDeploymentService |
Instance of a SeldonDeploymentService |
required |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
@staticmethod
def get_model_server_info( # type: ignore[override]
service_instance: "MLFlowDeploymentService",
) -> Dict[str, Optional[str]]:
"""Return implementation specific information that might be relevant
to the user.
Args:
service_instance: Instance of a SeldonDeploymentService
"""
return {
"PREDICTION_URL": service_instance.endpoint.prediction_url,
"MODEL_URI": service_instance.config.model_uri,
"MODEL_NAME": service_instance.config.model_name,
"SERVICE_PATH": service_instance.status.runtime_path,
"DAEMON_PID": str(service_instance.status.pid),
}
get_service_path(uuid)
staticmethod
Get the path the path where the local MLflow deployment service configuration, PID and log files are stored.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
uuid |
UUID |
The UUID of the MLflow model deployer. |
required |
Returns:
| Type | Description |
|---|---|
str |
The service path. |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
@staticmethod
def get_service_path(uuid: uuid.UUID) -> str:
"""Get the path the path where the local MLflow deployment service
configuration, PID and log files are stored.
Args:
uuid: The UUID of the MLflow model deployer.
Returns:
The service path.
"""
service_path = os.path.join(
get_global_config_directory(),
LOCAL_STORES_DIRECTORY_NAME,
str(uuid),
)
create_dir_recursive_if_not_exists(service_path)
return service_path
set_service_path(values)
classmethod
Sets the service_path attribute value according to the component UUID.
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
@root_validator
def set_service_path(cls, values: Dict[str, Any]) -> Dict[str, Any]:
"""Sets the service_path attribute value according to the component
UUID."""
if values.get("service_path"):
return values
# not likely to happen, due to Pydantic validation, but mypy complains
assert "uuid" in values
values["service_path"] = cls.get_service_path(values["uuid"])
return values
start_model_server(self, uuid, timeout=10)
Method to start a model server.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
uuid |
UUID |
UUID of the model server to start. |
required |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
def start_model_server(
self, uuid: UUID, timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT
) -> None:
"""Method to start a model server.
Args:
uuid: UUID of the model server to start.
"""
# get list of all services
existing_services = self.find_model_server(service_uuid=uuid)
# if the service exists, start it
if existing_services:
existing_services[0].start(timeout=timeout)
stop_model_server(self, uuid, timeout=10, force=False)
Method to stop a model server.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
uuid |
UUID |
UUID of the model server to stop. |
required |
timeout |
int |
Timeout in seconds to wait for the service to stop. |
10 |
force |
bool |
If True, force the service to stop. |
False |
Source code in zenml/integrations/mlflow/model_deployers/mlflow_model_deployer.py
def stop_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Method to stop a model server.
Args:
uuid: UUID of the model server to stop.
timeout: Timeout in seconds to wait for the service to stop.
force: If True, force the service to stop.
"""
# get list of all services
existing_services = self.find_model_server(service_uuid=uuid)
# if the service exists, stop it
if existing_services:
existing_services[0].stop(timeout=timeout, force=force)
services
special
mlflow_deployment
MLFlowDeploymentConfig (LocalDaemonServiceConfig)
pydantic-model
MLflow model deployment configuration.
Attributes:
| Name | Type | Description |
|---|---|---|
model_uri |
str |
URI of the MLflow model to serve |
model_name |
str |
the name of the model |
workers |
int |
number of workers to use for the prediction service |
mlserver |
bool |
set to True to use the MLflow MLServer backend (see https://github.com/SeldonIO/MLServer). If False, the MLflow built-in scoring server will be used. |
Source code in zenml/integrations/mlflow/services/mlflow_deployment.py
class MLFlowDeploymentConfig(LocalDaemonServiceConfig):
"""MLflow model deployment configuration.
Attributes:
model_uri: URI of the MLflow model to serve
model_name: the name of the model
workers: number of workers to use for the prediction service
mlserver: set to True to use the MLflow MLServer backend (see
https://github.com/SeldonIO/MLServer). If False, the
MLflow built-in scoring server will be used.
"""
model_uri: str
model_name: str
workers: int = 1
mlserver: bool = False
MLFlowDeploymentEndpoint (LocalDaemonServiceEndpoint)
pydantic-model
A service endpoint exposed by the MLflow deployment daemon.
Attributes:
| Name | Type | Description |
|---|---|---|
config |
MLFlowDeploymentEndpointConfig |
service endpoint configuration |
monitor |
HTTPEndpointHealthMonitor |
optional service endpoint health monitor |
Source code in zenml/integrations/mlflow/services/mlflow_deployment.py
class MLFlowDeploymentEndpoint(LocalDaemonServiceEndpoint):
"""A service endpoint exposed by the MLflow deployment daemon.
Attributes:
config: service endpoint configuration
monitor: optional service endpoint health monitor
"""
config: MLFlowDeploymentEndpointConfig
monitor: HTTPEndpointHealthMonitor
@property
def prediction_url(self) -> Optional[str]:
uri = self.status.uri
if not uri:
return None
return f"{uri}{self.config.prediction_url_path}"
MLFlowDeploymentEndpointConfig (LocalDaemonServiceEndpointConfig)
pydantic-model
MLflow daemon service endpoint configuration.
Attributes:
| Name | Type | Description |
|---|---|---|
prediction_url_path |
str |
URI subpath for prediction requests |
Source code in zenml/integrations/mlflow/services/mlflow_deployment.py
class MLFlowDeploymentEndpointConfig(LocalDaemonServiceEndpointConfig):
"""MLflow daemon service endpoint configuration.
Attributes:
prediction_url_path: URI subpath for prediction requests
"""
prediction_url_path: str
MLFlowDeploymentService (LocalDaemonService)
pydantic-model
MLFlow deployment service that can be used to start a local prediction server for MLflow models.
Attributes:
| Name | Type | Description |
|---|---|---|
SERVICE_TYPE |
ClassVar[zenml.services.service_type.ServiceType] |
a service type descriptor with information describing the MLflow deployment service class |
config |
MLFlowDeploymentConfig |
service configuration |
endpoint |
MLFlowDeploymentEndpoint |
optional service endpoint |
Source code in zenml/integrations/mlflow/services/mlflow_deployment.py
class MLFlowDeploymentService(LocalDaemonService):
"""MLFlow deployment service that can be used to start a local prediction
server for MLflow models.
Attributes:
SERVICE_TYPE: a service type descriptor with information describing
the MLflow deployment service class
config: service configuration
endpoint: optional service endpoint
"""
SERVICE_TYPE = ServiceType(
name="mlflow-deployment",
type="model-serving",
flavor="mlflow",
description="MLflow prediction service",
)
config: MLFlowDeploymentConfig
endpoint: MLFlowDeploymentEndpoint
def __init__(
self,
config: Union[MLFlowDeploymentConfig, Dict[str, Any]],
**attrs: Any,
) -> None:
# ensure that the endpoint is created before the service is initialized
# TODO [ENG-700]: implement a service factory or builder for MLflow
# deployment services
if (
isinstance(config, MLFlowDeploymentConfig)
and "endpoint" not in attrs
):
if config.mlserver:
prediction_url_path = MLSERVER_PREDICTION_URL_PATH
healthcheck_uri_path = MLSERVER_HEALTHCHECK_URL_PATH
use_head_request = False
else:
prediction_url_path = MLFLOW_PREDICTION_URL_PATH
healthcheck_uri_path = MLFLOW_HEALTHCHECK_URL_PATH
use_head_request = True
endpoint = MLFlowDeploymentEndpoint(
config=MLFlowDeploymentEndpointConfig(
protocol=ServiceEndpointProtocol.HTTP,
prediction_url_path=prediction_url_path,
),
monitor=HTTPEndpointHealthMonitor(
config=HTTPEndpointHealthMonitorConfig(
healthcheck_uri_path=healthcheck_uri_path,
use_head_request=use_head_request,
)
),
)
attrs["endpoint"] = endpoint
super().__init__(config=config, **attrs)
def run(self) -> None:
logger.info(
"Starting MLflow prediction service as blocking "
"process... press CTRL+C once to stop it."
)
self.endpoint.prepare_for_start()
try:
backend = PyFuncBackend(
config={},
no_conda=True,
workers=self.config.workers,
install_mlflow=False,
)
backend.serve(
model_uri=self.config.model_uri,
port=self.endpoint.status.port,
host="localhost",
enable_mlserver=self.config.mlserver,
)
except KeyboardInterrupt:
logger.info(
"MLflow prediction service stopped. Resuming normal execution."
)
@property
def prediction_url(self) -> Optional[str]:
"""Get the URI where the prediction service is answering requests.
Returns:
The URI where the prediction service can be contacted to process
HTTP/REST inference requests, or None, if the service isn't running.
"""
if not self.is_running:
return None
return self.endpoint.prediction_url
def predict(self, request: "NDArray[Any]") -> "NDArray[Any]":
"""Make a prediction using the service.
Args:
request: a numpy array representing the request
Returns:
A numpy array representing the prediction returned by the service.
"""
if not self.is_running:
raise Exception(
"MLflow prediction service is not running. "
"Please start the service before making predictions."
)
if self.endpoint.prediction_url is not None:
response = requests.post(
self.endpoint.prediction_url,
json={"instances": request.tolist()},
)
else:
raise ValueError("No endpoint known for prediction.")
response.raise_for_status()
return np.array(response.json())
prediction_url: Optional[str]
property
readonly
Get the URI where the prediction service is answering requests.
Returns:
| Type | Description |
|---|---|
Optional[str] |
The URI where the prediction service can be contacted to process HTTP/REST inference requests, or None, if the service isn't running. |
predict(self, request)
Make a prediction using the service.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
request |
NDArray[Any] |
a numpy array representing the request |
required |
Returns:
| Type | Description |
|---|---|
NDArray[Any] |
A numpy array representing the prediction returned by the service. |
Source code in zenml/integrations/mlflow/services/mlflow_deployment.py
def predict(self, request: "NDArray[Any]") -> "NDArray[Any]":
"""Make a prediction using the service.
Args:
request: a numpy array representing the request
Returns:
A numpy array representing the prediction returned by the service.
"""
if not self.is_running:
raise Exception(
"MLflow prediction service is not running. "
"Please start the service before making predictions."
)
if self.endpoint.prediction_url is not None:
response = requests.post(
self.endpoint.prediction_url,
json={"instances": request.tolist()},
)
else:
raise ValueError("No endpoint known for prediction.")
response.raise_for_status()
return np.array(response.json())
run(self)
Run the service daemon process associated with this service.
Subclasses must implement this method to provide the service daemon
functionality. This method will be executed in the context of the
running daemon, not in the context of the process that calls the
start method.
Source code in zenml/integrations/mlflow/services/mlflow_deployment.py
def run(self) -> None:
logger.info(
"Starting MLflow prediction service as blocking "
"process... press CTRL+C once to stop it."
)
self.endpoint.prepare_for_start()
try:
backend = PyFuncBackend(
config={},
no_conda=True,
workers=self.config.workers,
install_mlflow=False,
)
backend.serve(
model_uri=self.config.model_uri,
port=self.endpoint.status.port,
host="localhost",
enable_mlserver=self.config.mlserver,
)
except KeyboardInterrupt:
logger.info(
"MLflow prediction service stopped. Resuming normal execution."
)
steps
special
mlflow_deployer
MLFlowDeployerConfig (BaseStepConfig)
pydantic-model
MLflow model deployer step configuration
Attributes:
| Name | Type | Description |
|---|---|---|
model_name |
str |
the name of the MLflow model logged in the MLflow artifact store for the current pipeline. |
workers |
int |
number of workers to use for the prediction service |
mlserver |
bool |
set to True to use the MLflow MLServer backend (see https://github.com/SeldonIO/MLServer). If False, the MLflow built-in scoring server will be used. |
timeout |
int |
the number of seconds to wait for the service to start/stop. |
Source code in zenml/integrations/mlflow/steps/mlflow_deployer.py
class MLFlowDeployerConfig(BaseStepConfig):
"""MLflow model deployer step configuration
Attributes:
model_name: the name of the MLflow model logged in the MLflow artifact
store for the current pipeline.
workers: number of workers to use for the prediction service
mlserver: set to True to use the MLflow MLServer backend (see
https://github.com/SeldonIO/MLServer). If False, the
MLflow built-in scoring server will be used.
timeout: the number of seconds to wait for the service to start/stop.
"""
model_name: str = "model"
model_uri: str = ""
workers: int = 1
mlserver: bool = False
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT
mlflow_model_deployer_step (mlflow_model_deployer_step)
CONFIG_CLASS (BaseStepConfig)
pydantic-model
MLflow model deployer step configuration
Attributes:
| Name | Type | Description |
|---|---|---|
model_name |
str |
the name of the MLflow model logged in the MLflow artifact store for the current pipeline. |
workers |
int |
number of workers to use for the prediction service |
mlserver |
bool |
set to True to use the MLflow MLServer backend (see https://github.com/SeldonIO/MLServer). If False, the MLflow built-in scoring server will be used. |
timeout |
int |
the number of seconds to wait for the service to start/stop. |
Source code in zenml/integrations/mlflow/steps/mlflow_deployer.py
class MLFlowDeployerConfig(BaseStepConfig):
"""MLflow model deployer step configuration
Attributes:
model_name: the name of the MLflow model logged in the MLflow artifact
store for the current pipeline.
workers: number of workers to use for the prediction service
mlserver: set to True to use the MLflow MLServer backend (see
https://github.com/SeldonIO/MLServer). If False, the
MLflow built-in scoring server will be used.
timeout: the number of seconds to wait for the service to start/stop.
"""
model_name: str = "model"
model_uri: str = ""
workers: int = 1
mlserver: bool = False
timeout: int = DEFAULT_SERVICE_START_STOP_TIMEOUT
entrypoint(deploy_decision, model, config)
staticmethod
MLflow model deployer pipeline step
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
deploy_decision |
bool |
whether to deploy the model or not |
required |
model |
ModelArtifact |
the model artifact to deploy |
required |
config |
MLFlowDeployerConfig |
configuration for the deployer step |
required |
Returns:
| Type | Description |
|---|---|
MLFlowDeploymentService |
MLflow deployment service |
Source code in zenml/integrations/mlflow/steps/mlflow_deployer.py
@enable_mlflow
@step(enable_cache=True)
def mlflow_model_deployer_step(
deploy_decision: bool,
model: ModelArtifact,
config: MLFlowDeployerConfig,
) -> MLFlowDeploymentService:
"""MLflow model deployer pipeline step
Args:
deploy_decision: whether to deploy the model or not
model: the model artifact to deploy
config: configuration for the deployer step
Returns:
MLflow deployment service
"""
model_deployer = MLFlowModelDeployer.get_active_model_deployer()
if not config.model_uri:
# fetch the MLflow artifacts logged during the pipeline run
experiment_tracker = Repository( # type: ignore[call-arg]
skip_repository_check=True
).active_stack.experiment_tracker
if not isinstance(experiment_tracker, MLFlowExperimentTracker):
raise get_missing_mlflow_experiment_tracker_error()
client = MlflowClient()
model_uri = ""
mlflow_run = experiment_tracker.active_run
if mlflow_run and client.list_artifacts(
mlflow_run.info.run_id, config.model_name
):
model_uri = get_artifact_uri(config.model_name)
config.model_uri = model_uri
# get pipeline name, step name and run id
step_env = cast(StepEnvironment, Environment()[STEP_ENVIRONMENT_NAME])
pipeline_name = step_env.pipeline_name
run_id = step_env.pipeline_run_id
step_name = step_env.step_name
# fetch existing services with same pipeline name, step name and model name
existing_services = model_deployer.find_model_server(
pipeline_name=pipeline_name,
pipeline_step_name=step_name,
model_name=config.model_name,
)
# create a config for the new model service
predictor_cfg = MLFlowDeploymentConfig(
model_name=config.model_name or "",
model_uri=config.model_uri or "",
workers=config.workers,
mlserver=config.mlserver,
pipeline_name=pipeline_name,
pipeline_run_id=run_id,
pipeline_step_name=step_name,
)
# Creating a new service with inactive state and status by default
service = MLFlowDeploymentService(predictor_cfg)
if existing_services:
service = cast(MLFlowDeploymentService, existing_services[0])
# check for conditions to deploy the model
if not config.model_uri:
# an MLflow model was not found in the current run, so we simply reuse
# the service created during the previous step run
if not existing_services:
raise RuntimeError(
f"An MLflow model with name `{config.model_name}` was not "
f"trained in the current pipeline run and no previous "
f"service was found."
)
return service
if not deploy_decision:
logger.info(
f"Skipping model deployment because the model quality does not "
f"meet the criteria. Loading last service deployed by step "
f"'{step_name}' and pipeline '{pipeline_name}' for model "
f"'{config.model_name}'..."
)
# no new deployment is required, so we reuse the existing service
return service
# create a new model deployment and replace an old one if it exists
new_service = cast(
MLFlowDeploymentService,
model_deployer.deploy_model(
replace=True,
config=predictor_cfg,
timeout=config.timeout,
),
)
logger.info(
f"MLflow deployment service started and reachable at:\n"
f" {new_service.prediction_url}\n"
)
return new_service
mlflow_deployer_step(enable_cache=True, name=None)
Shortcut function to create a pipeline step to deploy a given ML model with a local MLflow prediction server.
The returned step can be used in a pipeline to implement continuous deployment for an MLflow model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
enable_cache |
bool |
Specify whether caching is enabled for this step. If no value is passed, caching is enabled by default |
True |
Returns:
| Type | Description |
|---|---|
Type[zenml.steps.base_step.BaseStep] |
an MLflow model deployer pipeline step |
Source code in zenml/integrations/mlflow/steps/mlflow_deployer.py
def mlflow_deployer_step(
enable_cache: bool = True,
name: Optional[str] = None,
) -> Type[BaseStep]:
"""Shortcut function to create a pipeline step to deploy a given ML model
with a local MLflow prediction server.
The returned step can be used in a pipeline to implement continuous
deployment for an MLflow model.
Args:
enable_cache: Specify whether caching is enabled for this step. If no
value is passed, caching is enabled by default
Returns:
an MLflow model deployer pipeline step
"""
logger.warning(
"The `mlflow_deployer_step` function is deprecated. Please "
"use the built-in `mlflow_model_deployer_step` step instead."
)
return mlflow_model_deployer_step
neural_prophet
special
NeuralProphetIntegration (Integration)
Definition of NeuralProphet integration for ZenML.
Source code in zenml/integrations/neural_prophet/__init__.py
class NeuralProphetIntegration(Integration):
"""Definition of NeuralProphet integration for ZenML."""
NAME = NEURAL_PROPHET
REQUIREMENTS = ["neuralprophet>=0.3.2"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.neural_prophet import materializers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/neural_prophet/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.neural_prophet import materializers # noqa
materializers
special
neural_prophet_materializer
NeuralProphetMaterializer (BaseMaterializer)
Materializer to read/write NeuralProphet models.
Source code in zenml/integrations/neural_prophet/materializers/neural_prophet_materializer.py
class NeuralProphetMaterializer(BaseMaterializer):
"""Materializer to read/write NeuralProphet models."""
ASSOCIATED_TYPES = (NeuralProphet,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> NeuralProphet:
"""Reads and returns a NeuralProphet model.
Returns:
A loaded NeuralProphet model.
"""
super().handle_input(data_type)
return torch.load( # type: ignore[no-untyped-call]
os.path.join(self.artifact.uri, DEFAULT_FILENAME)
) # noqa
def handle_return(self, model: NeuralProphet) -> None:
"""Writes a NeuralProphet model.
Args:
model: A NeuralProphet model object.
"""
super().handle_return(model)
torch.save(model, os.path.join(self.artifact.uri, DEFAULT_FILENAME))
handle_input(self, data_type)
Reads and returns a NeuralProphet model.
Returns:
| Type | Description |
|---|---|
NeuralProphet |
A loaded NeuralProphet model. |
Source code in zenml/integrations/neural_prophet/materializers/neural_prophet_materializer.py
def handle_input(self, data_type: Type[Any]) -> NeuralProphet:
"""Reads and returns a NeuralProphet model.
Returns:
A loaded NeuralProphet model.
"""
super().handle_input(data_type)
return torch.load( # type: ignore[no-untyped-call]
os.path.join(self.artifact.uri, DEFAULT_FILENAME)
) # noqa
handle_return(self, model)
Writes a NeuralProphet model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
model |
NeuralProphet |
A NeuralProphet model object. |
required |
Source code in zenml/integrations/neural_prophet/materializers/neural_prophet_materializer.py
def handle_return(self, model: NeuralProphet) -> None:
"""Writes a NeuralProphet model.
Args:
model: A NeuralProphet model object.
"""
super().handle_return(model)
torch.save(model, os.path.join(self.artifact.uri, DEFAULT_FILENAME))
plotly
special
PlotlyIntegration (Integration)
Definition of Plotly integration for ZenML.
Source code in zenml/integrations/plotly/__init__.py
class PlotlyIntegration(Integration):
"""Definition of Plotly integration for ZenML."""
NAME = PLOTLY
REQUIREMENTS = ["plotly>=5.4.0"]
visualizers
special
pipeline_lineage_visualizer
PipelineLineageVisualizer (BasePipelineVisualizer)
Visualize the lineage of runs in a pipeline using plotly.
Source code in zenml/integrations/plotly/visualizers/pipeline_lineage_visualizer.py
class PipelineLineageVisualizer(BasePipelineVisualizer):
"""Visualize the lineage of runs in a pipeline using plotly."""
@abstractmethod
def visualize(
self, object: PipelineView, *args: Any, **kwargs: Any
) -> Figure:
"""Creates a pipeline lineage diagram using plotly."""
logger.warning(
"This integration is not completed yet. Results might be unexpected."
)
category_dict = {}
dimensions = ["run"]
for run in object.runs:
category_dict[run.name] = {"run": run.name}
for step in run.steps:
category_dict[run.name].update(
{
step.entrypoint_name: str(step.id),
}
)
if step.entrypoint_name not in dimensions:
dimensions.append(f"{step.entrypoint_name}")
category_df = pd.DataFrame.from_dict(category_dict, orient="index")
category_df = category_df.reset_index()
fig = px.parallel_categories(
category_df,
dimensions,
color=None,
labels="status",
)
fig.show()
return fig
visualize(self, object, *args, **kwargs)
Creates a pipeline lineage diagram using plotly.
Source code in zenml/integrations/plotly/visualizers/pipeline_lineage_visualizer.py
@abstractmethod
def visualize(
self, object: PipelineView, *args: Any, **kwargs: Any
) -> Figure:
"""Creates a pipeline lineage diagram using plotly."""
logger.warning(
"This integration is not completed yet. Results might be unexpected."
)
category_dict = {}
dimensions = ["run"]
for run in object.runs:
category_dict[run.name] = {"run": run.name}
for step in run.steps:
category_dict[run.name].update(
{
step.entrypoint_name: str(step.id),
}
)
if step.entrypoint_name not in dimensions:
dimensions.append(f"{step.entrypoint_name}")
category_df = pd.DataFrame.from_dict(category_dict, orient="index")
category_df = category_df.reset_index()
fig = px.parallel_categories(
category_df,
dimensions,
color=None,
labels="status",
)
fig.show()
return fig
pytorch
special
PytorchIntegration (Integration)
Definition of PyTorch integration for ZenML.
Source code in zenml/integrations/pytorch/__init__.py
class PytorchIntegration(Integration):
"""Definition of PyTorch integration for ZenML."""
NAME = PYTORCH
REQUIREMENTS = ["torch"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.pytorch import materializers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/pytorch/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.pytorch import materializers # noqa
materializers
special
pytorch_materializer
PyTorchMaterializer (BaseMaterializer)
Materializer to read/write Pytorch models.
Source code in zenml/integrations/pytorch/materializers/pytorch_materializer.py
class PyTorchMaterializer(BaseMaterializer):
"""Materializer to read/write Pytorch models."""
ASSOCIATED_TYPES = (Module, TorchDict)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> Union[Module, TorchDict]:
"""Reads and returns a PyTorch model.
Returns:
A loaded pytorch model.
"""
super().handle_input(data_type)
with fileio.open(
os.path.join(self.artifact.uri, DEFAULT_FILENAME), "rb"
) as f:
return torch.load(f) # type: ignore[no-untyped-call] # noqa
def handle_return(self, model: Union[Module, TorchDict]) -> None:
"""Writes a PyTorch model.
Args:
model: A torch.nn.Module or a dict to pass into model.save
"""
super().handle_return(model)
# Save entire model to artifact directory, This is the default behavior for loading model in development phase (training, evaluation)
with fileio.open(
os.path.join(self.artifact.uri, DEFAULT_FILENAME), "wb"
) as f:
torch.save(model, f)
# Save model checkpoint to artifact directory, This is the default behavior for loading model in production phase (inference)
if isinstance(model, Module):
with fileio.open(
os.path.join(self.artifact.uri, CHECKPOINT_FILENAME), "wb"
) as f:
torch.save(model.state_dict(), f)
handle_input(self, data_type)
Reads and returns a PyTorch model.
Returns:
| Type | Description |
|---|---|
Union[torch.nn.modules.module.Module, zenml.integrations.pytorch.materializers.pytorch_types.TorchDict] |
A loaded pytorch model. |
Source code in zenml/integrations/pytorch/materializers/pytorch_materializer.py
def handle_input(self, data_type: Type[Any]) -> Union[Module, TorchDict]:
"""Reads and returns a PyTorch model.
Returns:
A loaded pytorch model.
"""
super().handle_input(data_type)
with fileio.open(
os.path.join(self.artifact.uri, DEFAULT_FILENAME), "rb"
) as f:
return torch.load(f) # type: ignore[no-untyped-call] # noqa
handle_return(self, model)
Writes a PyTorch model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
model |
Union[torch.nn.modules.module.Module, zenml.integrations.pytorch.materializers.pytorch_types.TorchDict] |
A torch.nn.Module or a dict to pass into model.save |
required |
Source code in zenml/integrations/pytorch/materializers/pytorch_materializer.py
def handle_return(self, model: Union[Module, TorchDict]) -> None:
"""Writes a PyTorch model.
Args:
model: A torch.nn.Module or a dict to pass into model.save
"""
super().handle_return(model)
# Save entire model to artifact directory, This is the default behavior for loading model in development phase (training, evaluation)
with fileio.open(
os.path.join(self.artifact.uri, DEFAULT_FILENAME), "wb"
) as f:
torch.save(model, f)
# Save model checkpoint to artifact directory, This is the default behavior for loading model in production phase (inference)
if isinstance(model, Module):
with fileio.open(
os.path.join(self.artifact.uri, CHECKPOINT_FILENAME), "wb"
) as f:
torch.save(model.state_dict(), f)
pytorch_types
TorchDict (dict, Generic)
A type of dict that represents saving a model.
Source code in zenml/integrations/pytorch/materializers/pytorch_types.py
class TorchDict(Dict[str, Any]):
"""A type of dict that represents saving a model."""
pytorch_lightning
special
PytorchLightningIntegration (Integration)
Definition of PyTorch Lightning integration for ZenML.
Source code in zenml/integrations/pytorch_lightning/__init__.py
class PytorchLightningIntegration(Integration):
"""Definition of PyTorch Lightning integration for ZenML."""
NAME = PYTORCH_L
REQUIREMENTS = ["pytorch_lightning"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.pytorch_lightning import materializers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/pytorch_lightning/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.pytorch_lightning import materializers # noqa
materializers
special
pytorch_lightning_materializer
PyTorchLightningMaterializer (BaseMaterializer)
Materializer to read/write Pytorch models.
Source code in zenml/integrations/pytorch_lightning/materializers/pytorch_lightning_materializer.py
class PyTorchLightningMaterializer(BaseMaterializer):
"""Materializer to read/write Pytorch models."""
ASSOCIATED_TYPES = (Trainer,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> Trainer:
"""Reads and returns a PyTorch Lightning trainer.
Returns:
A PyTorch Lightning trainer object.
"""
super().handle_input(data_type)
return Trainer(
resume_from_checkpoint=os.path.join(
self.artifact.uri, CHECKPOINT_NAME
)
)
def handle_return(self, trainer: Trainer) -> None:
"""Writes a PyTorch Lightning trainer.
Args:
trainer: A PyTorch Lightning trainer object.
"""
super().handle_return(trainer)
trainer.save_checkpoint(
os.path.join(self.artifact.uri, CHECKPOINT_NAME)
)
handle_input(self, data_type)
Reads and returns a PyTorch Lightning trainer.
Returns:
| Type | Description |
|---|---|
Trainer |
A PyTorch Lightning trainer object. |
Source code in zenml/integrations/pytorch_lightning/materializers/pytorch_lightning_materializer.py
def handle_input(self, data_type: Type[Any]) -> Trainer:
"""Reads and returns a PyTorch Lightning trainer.
Returns:
A PyTorch Lightning trainer object.
"""
super().handle_input(data_type)
return Trainer(
resume_from_checkpoint=os.path.join(
self.artifact.uri, CHECKPOINT_NAME
)
)
handle_return(self, trainer)
Writes a PyTorch Lightning trainer.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
trainer |
Trainer |
A PyTorch Lightning trainer object. |
required |
Source code in zenml/integrations/pytorch_lightning/materializers/pytorch_lightning_materializer.py
def handle_return(self, trainer: Trainer) -> None:
"""Writes a PyTorch Lightning trainer.
Args:
trainer: A PyTorch Lightning trainer object.
"""
super().handle_return(trainer)
trainer.save_checkpoint(
os.path.join(self.artifact.uri, CHECKPOINT_NAME)
)
registry
IntegrationRegistry
Registry to keep track of ZenML Integrations
Source code in zenml/integrations/registry.py
class IntegrationRegistry(object):
"""Registry to keep track of ZenML Integrations"""
def __init__(self) -> None:
"""Initializing the integration registry"""
self._integrations: Dict[str, Type["Integration"]] = {}
@property
def integrations(self) -> Dict[str, Type["Integration"]]:
"""Method to get integrations dictionary.
Returns:
A dict of integration key to type of `Integration`.
"""
return self._integrations
@integrations.setter
def integrations(self, i: Any) -> None:
"""Setter method for the integrations property"""
raise IntegrationError(
"Please do not manually change the integrations within the "
"registry. If you would like to register a new integration "
"manually, please use "
"`integration_registry.register_integration()`."
)
def register_integration(
self, key: str, type_: Type["Integration"]
) -> None:
"""Method to register an integration with a given name"""
self._integrations[key] = type_
def activate_integrations(self) -> None:
"""Method to activate the integrations with are registered in the
registry"""
for name, integration in self._integrations.items():
if integration.check_installation():
integration.activate()
logger.debug(f"Integration `{name}` is activated.")
else:
logger.debug(f"Integration `{name}` could not be activated.")
@property
def list_integration_names(self) -> List[str]:
"""Get a list of all possible integrations"""
return [name for name in self._integrations]
def select_integration_requirements(
self, integration_name: Optional[str] = None
) -> List[str]:
"""Select the requirements for a given integration
or all integrations"""
if integration_name:
if integration_name in self.list_integration_names:
return self._integrations[integration_name].REQUIREMENTS
else:
raise KeyError(
f"Version {integration_name} does not exist. "
f"Currently the following integrations are implemented. "
f"{self.list_integration_names}"
)
else:
return [
requirement
for name in self.list_integration_names
for requirement in self._integrations[name].REQUIREMENTS
]
def is_installed(self, integration_name: Optional[str] = None) -> bool:
"""Checks if all requirements for an integration are installed"""
if integration_name in self.list_integration_names:
return self._integrations[integration_name].check_installation()
elif not integration_name:
all_installed = [
self._integrations[item].check_installation()
for item in self.list_integration_names
]
return all(all_installed)
else:
raise KeyError(
f"Integration '{integration_name}' not found. "
f"Currently the following integrations are available: "
f"{self.list_integration_names}"
)
integrations: Dict[str, Type[Integration]]
property
writable
Method to get integrations dictionary.
Returns:
| Type | Description |
|---|---|
Dict[str, Type[Integration]] |
A dict of integration key to type of |
list_integration_names: List[str]
property
readonly
Get a list of all possible integrations
__init__(self)
special
Initializing the integration registry
Source code in zenml/integrations/registry.py
def __init__(self) -> None:
"""Initializing the integration registry"""
self._integrations: Dict[str, Type["Integration"]] = {}
activate_integrations(self)
Method to activate the integrations with are registered in the registry
Source code in zenml/integrations/registry.py
def activate_integrations(self) -> None:
"""Method to activate the integrations with are registered in the
registry"""
for name, integration in self._integrations.items():
if integration.check_installation():
integration.activate()
logger.debug(f"Integration `{name}` is activated.")
else:
logger.debug(f"Integration `{name}` could not be activated.")
is_installed(self, integration_name=None)
Checks if all requirements for an integration are installed
Source code in zenml/integrations/registry.py
def is_installed(self, integration_name: Optional[str] = None) -> bool:
"""Checks if all requirements for an integration are installed"""
if integration_name in self.list_integration_names:
return self._integrations[integration_name].check_installation()
elif not integration_name:
all_installed = [
self._integrations[item].check_installation()
for item in self.list_integration_names
]
return all(all_installed)
else:
raise KeyError(
f"Integration '{integration_name}' not found. "
f"Currently the following integrations are available: "
f"{self.list_integration_names}"
)
register_integration(self, key, type_)
Method to register an integration with a given name
Source code in zenml/integrations/registry.py
def register_integration(
self, key: str, type_: Type["Integration"]
) -> None:
"""Method to register an integration with a given name"""
self._integrations[key] = type_
select_integration_requirements(self, integration_name=None)
Select the requirements for a given integration or all integrations
Source code in zenml/integrations/registry.py
def select_integration_requirements(
self, integration_name: Optional[str] = None
) -> List[str]:
"""Select the requirements for a given integration
or all integrations"""
if integration_name:
if integration_name in self.list_integration_names:
return self._integrations[integration_name].REQUIREMENTS
else:
raise KeyError(
f"Version {integration_name} does not exist. "
f"Currently the following integrations are implemented. "
f"{self.list_integration_names}"
)
else:
return [
requirement
for name in self.list_integration_names
for requirement in self._integrations[name].REQUIREMENTS
]
s3
special
The S3 integration allows the use of cloud artifact stores and file operations on S3 buckets.
S3Integration (Integration)
Definition of S3 integration for ZenML.
Source code in zenml/integrations/s3/__init__.py
class S3Integration(Integration):
"""Definition of S3 integration for ZenML."""
NAME = S3
REQUIREMENTS = ["s3fs==2022.3.0"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.s3 import artifact_stores # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/s3/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.s3 import artifact_stores # noqa
artifact_stores
special
s3_artifact_store
S3ArtifactStore (BaseArtifactStore)
pydantic-model
Artifact Store for S3 based artifacts.
All attributes of this class except path will be passed to the
s3fs.S3FileSystem initialization. See
here for more information on how
to use those configuration options to connect to any S3-compatible storage.
When you want to register an S3ArtifactStore from the CLI and need to pass
client_kwargs, config_kwargs or s3_additional_kwargs, you should pass
them as a json string:
zenml artifact-store register my_s3_store --type=s3 --path=s3://my_bucket --client_kwargs='{"endpoint_url": "http://my-s3-endpoint"}'
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
class S3ArtifactStore(BaseArtifactStore):
"""Artifact Store for S3 based artifacts.
All attributes of this class except `path` will be passed to the
`s3fs.S3FileSystem` initialization. See
[here](https://s3fs.readthedocs.io/en/latest/) for more information on how
to use those configuration options to connect to any S3-compatible storage.
When you want to register an S3ArtifactStore from the CLI and need to pass
`client_kwargs`, `config_kwargs` or `s3_additional_kwargs`, you should pass
them as a json string:
```
zenml artifact-store register my_s3_store --type=s3 --path=s3://my_bucket \
--client_kwargs='{"endpoint_url": "http://my-s3-endpoint"}'
```
"""
key: Optional[str] = None
secret: Optional[str] = None
token: Optional[str] = None
client_kwargs: Optional[Dict[str, Any]] = None
config_kwargs: Optional[Dict[str, Any]] = None
s3_additional_kwargs: Optional[Dict[str, Any]] = None
_filesystem: Optional[s3fs.S3FileSystem] = None
# Class variables
FLAVOR: ClassVar[str] = "s3"
SUPPORTED_SCHEMES: ClassVar[Set[str]] = {"s3://"}
@validator(
"client_kwargs", "config_kwargs", "s3_additional_kwargs", pre=True
)
def _convert_json_string(
cls, value: Union[None, str, Dict[str, Any]]
) -> Optional[Dict[str, Any]]:
"""Converts potential JSON strings passed via the CLI to
dictionaries.
Raises:
TypeError: If the value is not a `str`, `Dict` or `None`.
ValueError: If the value is an invalid json string or a json string
that does not decode into a dictionary.
"""
if isinstance(value, str):
try:
dict_ = json.loads(value)
except json.JSONDecodeError as e:
raise ValueError(f"Invalid json string '{value}'") from e
if not isinstance(dict_, Dict):
raise ValueError(
f"Json string '{value}' did not decode into a dictionary."
)
return dict_
elif isinstance(value, Dict) or value is None:
return value
else:
raise TypeError(f"{value} is not a json string or a dictionary.")
@property
def filesystem(self) -> s3fs.S3FileSystem:
"""The s3 filesystem to access this artifact store."""
if not self._filesystem:
self._filesystem = s3fs.S3FileSystem(
key=self.key,
secret=self.secret,
token=self.token,
client_kwargs=self.client_kwargs,
config_kwargs=self.config_kwargs,
s3_additional_kwargs=self.s3_additional_kwargs,
)
return self._filesystem
def open(self, path: PathType, mode: str = "r") -> Any:
"""Open a file at the given path.
Args:
path: Path of the file to open.
mode: Mode in which to open the file. Currently, only
'rb' and 'wb' to read and write binary files are supported.
"""
return self.filesystem.open(path=path, mode=mode)
def copyfile(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Copy a file.
Args:
src: The path to copy from.
dst: The path to copy to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to copy to destination '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to copy anyway."
)
# TODO [ENG-151]: Check if it works with overwrite=True or if we need to
# manually remove it first
self.filesystem.copy(path1=src, path2=dst)
def exists(self, path: PathType) -> bool:
"""Check whether a path exists."""
return self.filesystem.exists(path=path) # type: ignore[no-any-return]
def glob(self, pattern: PathType) -> List[PathType]:
"""Return all paths that match the given glob pattern.
The glob pattern may include:
- '*' to match any number of characters
- '?' to match a single character
- '[...]' to match one of the characters inside the brackets
- '**' as the full name of a path component to match to search
in subdirectories of any depth (e.g. '/some_dir/**/some_file)
Args:
pattern: The glob pattern to match, see details above.
Returns:
A list of paths that match the given glob pattern.
"""
return [f"s3://{path}" for path in self.filesystem.glob(path=pattern)]
def isdir(self, path: PathType) -> bool:
"""Check whether a path is a directory."""
return self.filesystem.isdir(path=path) # type: ignore[no-any-return]
def listdir(self, path: PathType) -> List[PathType]:
"""Return a list of files in a directory."""
# remove s3 prefix if given, so we can remove the directory later as
# this method is expected to only return filenames
path = convert_to_str(path)
if path.startswith("s3://"):
path = path[5:]
def _extract_basename(file_dict: Dict[str, Any]) -> str:
"""Extracts the basename from a file info dict returned by the S3
filesystem."""
file_path = cast(str, file_dict["Key"])
base_name = file_path[len(path) :]
return base_name.lstrip("/")
return [
_extract_basename(dict_)
for dict_ in self.filesystem.listdir(path=path)
# s3fs.listdir also returns the root directory, so we filter
# it out here
if _extract_basename(dict_)
]
def makedirs(self, path: PathType) -> None:
"""Create a directory at the given path. If needed also
create missing parent directories."""
self.filesystem.makedirs(path=path, exist_ok=True)
def mkdir(self, path: PathType) -> None:
"""Create a directory at the given path."""
self.filesystem.makedir(path=path)
def remove(self, path: PathType) -> None:
"""Remove the file at the given path."""
self.filesystem.rm_file(path=path)
def rename(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Rename source file to destination file.
Args:
src: The path of the file to rename.
dst: The path to rename the source file to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to rename file to '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to rename anyway."
)
# TODO [ENG-152]: Check if it works with overwrite=True or if we need
# to manually remove it first
self.filesystem.rename(path1=src, path2=dst)
def rmtree(self, path: PathType) -> None:
"""Remove the given directory."""
self.filesystem.delete(path=path, recursive=True)
def stat(self, path: PathType) -> Dict[str, Any]:
"""Return stat info for the given path."""
return self.filesystem.stat(path=path) # type: ignore[no-any-return]
def walk(
self,
top: PathType,
topdown: bool = True,
onerror: Optional[Callable[..., None]] = None,
) -> Iterable[Tuple[PathType, List[PathType], List[PathType]]]:
"""Return an iterator that walks the contents of the given directory.
Args:
top: Path of directory to walk.
topdown: Unused argument to conform to interface.
onerror: Unused argument to conform to interface.
Returns:
An Iterable of Tuples, each of which contain the path of the current
directory path, a list of directories inside the current directory
and a list of files inside the current directory.
"""
# TODO [ENG-153]: Additional params
for directory, subdirectories, files in self.filesystem.walk(path=top):
yield f"s3://{directory}", subdirectories, files
filesystem: S3FileSystem
property
readonly
The s3 filesystem to access this artifact store.
copyfile(self, src, dst, overwrite=False)
Copy a file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
src |
Union[bytes, str] |
The path to copy from. |
required |
dst |
Union[bytes, str] |
The path to copy to. |
required |
overwrite |
bool |
If a file already exists at the destination, this
method will overwrite it if overwrite= |
False |
Exceptions:
| Type | Description |
|---|---|
FileNotFoundError |
If the source file does not exist. |
FileExistsError |
If a file already exists at the destination
and overwrite is not set to |
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def copyfile(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Copy a file.
Args:
src: The path to copy from.
dst: The path to copy to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to copy to destination '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to copy anyway."
)
# TODO [ENG-151]: Check if it works with overwrite=True or if we need to
# manually remove it first
self.filesystem.copy(path1=src, path2=dst)
exists(self, path)
Check whether a path exists.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def exists(self, path: PathType) -> bool:
"""Check whether a path exists."""
return self.filesystem.exists(path=path) # type: ignore[no-any-return]
glob(self, pattern)
Return all paths that match the given glob pattern. The glob pattern may include: - '' to match any number of characters - '?' to match a single character - '[...]' to match one of the characters inside the brackets - '' as the full name of a path component to match to search in subdirectories of any depth (e.g. '/some_dir/*/some_file)
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pattern |
Union[bytes, str] |
The glob pattern to match, see details above. |
required |
Returns:
| Type | Description |
|---|---|
List[Union[bytes, str]] |
A list of paths that match the given glob pattern. |
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def glob(self, pattern: PathType) -> List[PathType]:
"""Return all paths that match the given glob pattern.
The glob pattern may include:
- '*' to match any number of characters
- '?' to match a single character
- '[...]' to match one of the characters inside the brackets
- '**' as the full name of a path component to match to search
in subdirectories of any depth (e.g. '/some_dir/**/some_file)
Args:
pattern: The glob pattern to match, see details above.
Returns:
A list of paths that match the given glob pattern.
"""
return [f"s3://{path}" for path in self.filesystem.glob(path=pattern)]
isdir(self, path)
Check whether a path is a directory.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def isdir(self, path: PathType) -> bool:
"""Check whether a path is a directory."""
return self.filesystem.isdir(path=path) # type: ignore[no-any-return]
listdir(self, path)
Return a list of files in a directory.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def listdir(self, path: PathType) -> List[PathType]:
"""Return a list of files in a directory."""
# remove s3 prefix if given, so we can remove the directory later as
# this method is expected to only return filenames
path = convert_to_str(path)
if path.startswith("s3://"):
path = path[5:]
def _extract_basename(file_dict: Dict[str, Any]) -> str:
"""Extracts the basename from a file info dict returned by the S3
filesystem."""
file_path = cast(str, file_dict["Key"])
base_name = file_path[len(path) :]
return base_name.lstrip("/")
return [
_extract_basename(dict_)
for dict_ in self.filesystem.listdir(path=path)
# s3fs.listdir also returns the root directory, so we filter
# it out here
if _extract_basename(dict_)
]
makedirs(self, path)
Create a directory at the given path. If needed also create missing parent directories.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def makedirs(self, path: PathType) -> None:
"""Create a directory at the given path. If needed also
create missing parent directories."""
self.filesystem.makedirs(path=path, exist_ok=True)
mkdir(self, path)
Create a directory at the given path.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def mkdir(self, path: PathType) -> None:
"""Create a directory at the given path."""
self.filesystem.makedir(path=path)
open(self, path, mode='r')
Open a file at the given path.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
path |
Union[bytes, str] |
Path of the file to open. |
required |
mode |
str |
Mode in which to open the file. Currently, only 'rb' and 'wb' to read and write binary files are supported. |
'r' |
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def open(self, path: PathType, mode: str = "r") -> Any:
"""Open a file at the given path.
Args:
path: Path of the file to open.
mode: Mode in which to open the file. Currently, only
'rb' and 'wb' to read and write binary files are supported.
"""
return self.filesystem.open(path=path, mode=mode)
remove(self, path)
Remove the file at the given path.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def remove(self, path: PathType) -> None:
"""Remove the file at the given path."""
self.filesystem.rm_file(path=path)
rename(self, src, dst, overwrite=False)
Rename source file to destination file.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
src |
Union[bytes, str] |
The path of the file to rename. |
required |
dst |
Union[bytes, str] |
The path to rename the source file to. |
required |
overwrite |
bool |
If a file already exists at the destination, this
method will overwrite it if overwrite= |
False |
Exceptions:
| Type | Description |
|---|---|
FileNotFoundError |
If the source file does not exist. |
FileExistsError |
If a file already exists at the destination
and overwrite is not set to |
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def rename(
self, src: PathType, dst: PathType, overwrite: bool = False
) -> None:
"""Rename source file to destination file.
Args:
src: The path of the file to rename.
dst: The path to rename the source file to.
overwrite: If a file already exists at the destination, this
method will overwrite it if overwrite=`True` and
raise a FileExistsError otherwise.
Raises:
FileNotFoundError: If the source file does not exist.
FileExistsError: If a file already exists at the destination
and overwrite is not set to `True`.
"""
if not overwrite and self.filesystem.exists(dst):
raise FileExistsError(
f"Unable to rename file to '{convert_to_str(dst)}', "
f"file already exists. Set `overwrite=True` to rename anyway."
)
# TODO [ENG-152]: Check if it works with overwrite=True or if we need
# to manually remove it first
self.filesystem.rename(path1=src, path2=dst)
rmtree(self, path)
Remove the given directory.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def rmtree(self, path: PathType) -> None:
"""Remove the given directory."""
self.filesystem.delete(path=path, recursive=True)
stat(self, path)
Return stat info for the given path.
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def stat(self, path: PathType) -> Dict[str, Any]:
"""Return stat info for the given path."""
return self.filesystem.stat(path=path) # type: ignore[no-any-return]
walk(self, top, topdown=True, onerror=None)
Return an iterator that walks the contents of the given directory.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
top |
Union[bytes, str] |
Path of directory to walk. |
required |
topdown |
bool |
Unused argument to conform to interface. |
True |
onerror |
Optional[Callable[..., NoneType]] |
Unused argument to conform to interface. |
None |
Returns:
| Type | Description |
|---|---|
Iterable[Tuple[Union[bytes, str], List[Union[bytes, str]], List[Union[bytes, str]]]] |
An Iterable of Tuples, each of which contain the path of the current directory path, a list of directories inside the current directory and a list of files inside the current directory. |
Source code in zenml/integrations/s3/artifact_stores/s3_artifact_store.py
def walk(
self,
top: PathType,
topdown: bool = True,
onerror: Optional[Callable[..., None]] = None,
) -> Iterable[Tuple[PathType, List[PathType], List[PathType]]]:
"""Return an iterator that walks the contents of the given directory.
Args:
top: Path of directory to walk.
topdown: Unused argument to conform to interface.
onerror: Unused argument to conform to interface.
Returns:
An Iterable of Tuples, each of which contain the path of the current
directory path, a list of directories inside the current directory
and a list of files inside the current directory.
"""
# TODO [ENG-153]: Additional params
for directory, subdirectories, files in self.filesystem.walk(path=top):
yield f"s3://{directory}", subdirectories, files
sagemaker
special
The Sagemaker integration submodule provides a way to run ZenML steps in Sagemaker.
SagemakerIntegration (Integration)
Definition of Sagemaker integration for ZenML.
Source code in zenml/integrations/sagemaker/__init__.py
class SagemakerIntegration(Integration):
"""Definition of Sagemaker integration for ZenML."""
NAME = SAGEMAKER
REQUIREMENTS = ["sagemaker==2.82.2"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.sagemaker import step_operators # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/sagemaker/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.sagemaker import step_operators # noqa
step_operators
special
sagemaker_step_operator
SagemakerStepOperator (BaseStepOperator)
pydantic-model
Step operator to run a step on Sagemaker.
This class defines code that builds an image with the ZenML entrypoint to run using Sagemaker's Estimator.
Attributes:
| Name | Type | Description |
|---|---|---|
role |
str |
The role that has to be assigned to the jobs which are running in Sagemaker. |
instance_type |
str |
The type of the compute instance where jobs will run. |
base_image |
Optional[str] |
[Optional] The base image to use for building the docker image that will be executed. |
bucket |
Optional[str] |
[Optional] Name of the S3 bucket to use for storing artifacts from the job run. If not provided, a default bucket will be created based on the following format: "sagemaker-{region}-{aws-account-id}". |
experiment_name |
Optional[str] |
[Optional] The name for the experiment to which the job will be associated. If not provided, the job runs would be independent. |
Source code in zenml/integrations/sagemaker/step_operators/sagemaker_step_operator.py
class SagemakerStepOperator(BaseStepOperator):
"""Step operator to run a step on Sagemaker.
This class defines code that builds an image with the ZenML entrypoint
to run using Sagemaker's Estimator.
Attributes:
role: The role that has to be assigned to the jobs which are
running in Sagemaker.
instance_type: The type of the compute instance where jobs will run.
base_image: [Optional] The base image to use for building the docker
image that will be executed.
bucket: [Optional] Name of the S3 bucket to use for storing artifacts
from the job run. If not provided, a default bucket will be created
based on the following format: "sagemaker-{region}-{aws-account-id}".
experiment_name: [Optional] The name for the experiment to which the job
will be associated. If not provided, the job runs would be
independent.
"""
role: str
instance_type: str
base_image: Optional[str] = None
bucket: Optional[str] = None
experiment_name: Optional[str] = None
# Class Configuration
FLAVOR: ClassVar[str] = "sagemaker"
@property
def validator(self) -> Optional[StackValidator]:
"""Validates that the stack contains a container registry."""
def _ensure_local_orchestrator(stack: Stack) -> Tuple[bool, str]:
return (
stack.orchestrator.FLAVOR == "local",
"Local orchestrator is required",
)
return StackValidator(
required_components={StackComponentType.CONTAINER_REGISTRY},
custom_validation_function=_ensure_local_orchestrator,
)
def _build_docker_image(
self,
pipeline_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> str:
repo = Repository()
container_registry = repo.active_stack.container_registry
if not container_registry:
raise RuntimeError("Missing container registry")
registry_uri = container_registry.uri.rstrip("/")
image_name = f"{registry_uri}/zenml-sagemaker:{pipeline_name}"
docker_utils.build_docker_image(
build_context_path=get_source_root_path(),
image_name=image_name,
entrypoint=" ".join(entrypoint_command),
requirements=set(requirements),
base_image=self.base_image,
)
docker_utils.push_docker_image(image_name)
return docker_utils.get_image_digest(image_name) or image_name
def launch(
self,
pipeline_name: str,
run_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> None:
"""Launches a step on Sagemaker.
Args:
pipeline_name: Name of the pipeline which the step to be executed
is part of.
run_name: Name of the pipeline run which the step to be executed
is part of.
entrypoint_command: Command that executes the step.
requirements: List of pip requirements that must be installed
inside the step operator environment.
"""
image_name = self._build_docker_image(
pipeline_name=pipeline_name,
requirements=requirements,
entrypoint_command=entrypoint_command,
)
session = sagemaker.Session(default_bucket=self.bucket)
estimator = sagemaker.estimator.Estimator(
image_name,
self.role,
instance_count=1,
instance_type=self.instance_type,
sagemaker_session=session,
)
# Sagemaker doesn't allow any underscores in job/experiment/trial names
sanitized_run_name = run_name.replace("_", "-")
experiment_config = {}
if self.experiment_name:
experiment_config = {
"ExperimentName": self.experiment_name,
"TrialName": sanitized_run_name,
}
estimator.fit(
wait=True,
experiment_config=experiment_config,
job_name=sanitized_run_name,
)
validator: Optional[zenml.stack.stack_validator.StackValidator]
property
readonly
Validates that the stack contains a container registry.
launch(self, pipeline_name, run_name, requirements, entrypoint_command)
Launches a step on Sagemaker.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline_name |
str |
Name of the pipeline which the step to be executed is part of. |
required |
run_name |
str |
Name of the pipeline run which the step to be executed is part of. |
required |
entrypoint_command |
List[str] |
Command that executes the step. |
required |
requirements |
List[str] |
List of pip requirements that must be installed inside the step operator environment. |
required |
Source code in zenml/integrations/sagemaker/step_operators/sagemaker_step_operator.py
def launch(
self,
pipeline_name: str,
run_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> None:
"""Launches a step on Sagemaker.
Args:
pipeline_name: Name of the pipeline which the step to be executed
is part of.
run_name: Name of the pipeline run which the step to be executed
is part of.
entrypoint_command: Command that executes the step.
requirements: List of pip requirements that must be installed
inside the step operator environment.
"""
image_name = self._build_docker_image(
pipeline_name=pipeline_name,
requirements=requirements,
entrypoint_command=entrypoint_command,
)
session = sagemaker.Session(default_bucket=self.bucket)
estimator = sagemaker.estimator.Estimator(
image_name,
self.role,
instance_count=1,
instance_type=self.instance_type,
sagemaker_session=session,
)
# Sagemaker doesn't allow any underscores in job/experiment/trial names
sanitized_run_name = run_name.replace("_", "-")
experiment_config = {}
if self.experiment_name:
experiment_config = {
"ExperimentName": self.experiment_name,
"TrialName": sanitized_run_name,
}
estimator.fit(
wait=True,
experiment_config=experiment_config,
job_name=sanitized_run_name,
)
seldon
special
The Seldon Core integration allows you to use the Seldon Core model serving platform to implement continuous model deployment.
SeldonIntegration (Integration)
Definition of Seldon Core integration for ZenML.
Source code in zenml/integrations/seldon/__init__.py
class SeldonIntegration(Integration):
"""Definition of Seldon Core integration for ZenML."""
NAME = SELDON
REQUIREMENTS = [
"kubernetes==18.20.0",
]
@staticmethod
def activate() -> None:
"""Activate the Seldon Core integration."""
from zenml.integrations.seldon import model_deployers # noqa
from zenml.integrations.seldon import secret_schemas # noqa
from zenml.integrations.seldon import services # noqa
activate()
staticmethod
Activate the Seldon Core integration.
Source code in zenml/integrations/seldon/__init__.py
@staticmethod
def activate() -> None:
"""Activate the Seldon Core integration."""
from zenml.integrations.seldon import model_deployers # noqa
from zenml.integrations.seldon import secret_schemas # noqa
from zenml.integrations.seldon import services # noqa
model_deployers
special
seldon_model_deployer
SeldonModelDeployer (BaseModelDeployer)
pydantic-model
Seldon Core model deployer stack component implementation.
Attributes:
| Name | Type | Description |
|---|---|---|
kubernetes_context |
Optional[str] |
the Kubernetes context to use to contact the remote Seldon Core installation. If not specified, the current configuration is used. Depending on where the Seldon model deployer is being used, this can be either a locally active context or an in-cluster Kubernetes configuration (if running inside a pod). |
kubernetes_namespace |
Optional[str] |
the Kubernetes namespace where the Seldon Core deployment servers are provisioned and managed by ZenML. If not specified, the namespace set in the current configuration is used. Depending on where the Seldon model deployer is being used, this can be either the current namespace configured in the locally active context or the namespace in the context of which the pod is running (if running inside a pod). |
base_url |
str |
the base URL of the Kubernetes ingress used to expose the Seldon Core deployment servers. |
secret |
Optional[str] |
the name of a ZenML secret containing the credentials used by Seldon Core storage initializers to authenticate to the Artifact Store (i.e. the storage backend where models are stored - see https://docs.seldon.io/projects/seldon-core/en/latest/servers/overview.html#handling-credentials). |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
class SeldonModelDeployer(BaseModelDeployer):
"""Seldon Core model deployer stack component implementation.
Attributes:
kubernetes_context: the Kubernetes context to use to contact the remote
Seldon Core installation. If not specified, the current
configuration is used. Depending on where the Seldon model deployer
is being used, this can be either a locally active context or an
in-cluster Kubernetes configuration (if running inside a pod).
kubernetes_namespace: the Kubernetes namespace where the Seldon Core
deployment servers are provisioned and managed by ZenML. If not
specified, the namespace set in the current configuration is used.
Depending on where the Seldon model deployer is being used, this can
be either the current namespace configured in the locally active
context or the namespace in the context of which the pod is running
(if running inside a pod).
base_url: the base URL of the Kubernetes ingress used to expose the
Seldon Core deployment servers.
secret: the name of a ZenML secret containing the credentials used by
Seldon Core storage initializers to authenticate to the Artifact
Store (i.e. the storage backend where models are stored - see
https://docs.seldon.io/projects/seldon-core/en/latest/servers/overview.html#handling-credentials).
"""
# Class Configuration
FLAVOR: ClassVar[str] = SELDON
kubernetes_context: Optional[str]
kubernetes_namespace: Optional[str]
base_url: str
secret: Optional[str]
# private attributes
_client: Optional[SeldonClient] = None
@staticmethod
def get_model_server_info( # type: ignore[override]
service_instance: "SeldonDeploymentService",
) -> Dict[str, Optional[str]]:
""" "Return implementation specific information that might be relevant
to the user.
Args:
service_instance: Instance of a SeldonDeploymentService
"""
return {
"PREDICTION_URL": service_instance.prediction_url,
"MODEL_URI": service_instance.config.model_uri,
"MODEL_NAME": service_instance.config.model_name,
"SELDON_DEPLOYMENT": service_instance.seldon_deployment_name,
}
@staticmethod
def get_active_model_deployer() -> "SeldonModelDeployer":
"""Get the Seldon Core model deployer registered in the active stack.
Returns:
The Seldon Core model deployer registered in the active stack.
Raises:
TypeError: if the Seldon Core model deployer is not available.
"""
model_deployer = Repository( # type: ignore [call-arg]
skip_repository_check=True
).active_stack.model_deployer
if not model_deployer or not isinstance(
model_deployer, SeldonModelDeployer
):
raise TypeError(
"The active stack needs to have a Seldon Core model deployer "
"component registered to be able to deploy models with Seldon "
"Core. You can create a new stack with a Seldon Core model "
"deployer component or update your existing stack to add this "
"component, e.g.:\n\n"
" 'zenml model-deployer register seldon --type=seldon "
"--kubernetes_context=context-name --kubernetes_namespace="
"namespace-name --base_url=https://ingress.cluster.kubernetes'\n"
" 'zenml stack create stack-name -d seldon ...'\n"
)
return model_deployer
@property
def seldon_client(self) -> SeldonClient:
"""Get the Seldon Core client associated with this model deployer.
Returns:
The Seldon Core client.
Raises:
SeldonClientError: if the Kubernetes client configuration cannot be
found.
"""
if not self._client:
self._client = SeldonClient(
context=self.kubernetes_context,
namespace=self.kubernetes_namespace,
)
return self._client
@property
def kubernetes_secret_name(self) -> Optional[str]:
"""Get the Kubernetes secret name associated with this model deployer.
If a secret is configured for this model deployer, a corresponding
Kubernetes secret is created in the remote cluster to be used
by Seldon Core storage initializers to authenticate to the Artifact
Store. This method returns the unique name that is used for this secret.
Returns:
The Seldon Core Kubernetes secret name, or None if no secret is
configured.
"""
if not self.secret:
return None
return (
re.sub(r"[^0-9a-zA-Z-]+", "-", f"zenml-seldon-core-{self.secret}")
.strip("-")
.lower()
)
def _create_or_update_kubernetes_secret(self) -> Optional[str]:
"""Create or update a Kubernetes secret with the information stored in
the ZenML secret configured for the model deployer.
Returns:
The name of the Kubernetes secret that was created or updated, or
None if no secret was configured.
Raises:
SeldonClientError: if the secret cannot be created or updated.
"""
# if a ZenML secret was configured in the model deployer,
# create a Kubernetes secret as a means to pass this information
# to the Seldon Core deployment
if self.secret:
secret_manager = Repository( # type: ignore [call-arg]
skip_repository_check=True
).active_stack.secrets_manager
if not secret_manager or not isinstance(
secret_manager, BaseSecretsManager
):
raise RuntimeError(
f"The active stack doesn't have a secret manager component. "
f"The ZenML secret specified in the Seldon Core Model "
f"Deployer configuration cannot be fetched: {self.secret}."
)
try:
zenml_secret = secret_manager.get_secret(self.secret)
except KeyError:
raise RuntimeError(
f"The ZenML secret '{self.secret}' specified in the "
f"Seldon Core Model Deployer configuration was not found "
f"in the active stack's secret manager."
)
# should never happen, just making mypy happy
assert self.kubernetes_secret_name is not None
self.seldon_client.create_or_update_secret(
self.kubernetes_secret_name, zenml_secret
)
return self.kubernetes_secret_name
def _delete_kubernetes_secret(self) -> None:
"""Delete the Kubernetes secret associated with this model deployer
if no Seldon Core deployments are using it.
Raises:
SeldonClientError: if the secret cannot be deleted.
"""
if self.kubernetes_secret_name:
# fetch all the Seldon Core deployments that currently
# configured to use this secret
services = self.find_model_server()
for service in services:
config = cast(SeldonDeploymentConfig, service.config)
if config.secret_name == self.kubernetes_secret_name:
return
self.seldon_client.delete_secret(self.kubernetes_secret_name)
def deploy_model(
self,
config: ServiceConfig,
replace: bool = False,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
) -> BaseService:
"""Create a new Seldon Core deployment or update an existing one to
serve the supplied model and deployment configuration.
This method has two modes of operation, depending on the `replace`
argument value:
* if `replace` is False, calling this method will create a new Seldon
Core deployment server to reflect the model and other configuration
parameters specified in the supplied Seldon deployment `config`.
* if `replace` is True, this method will first attempt to find an
existing Seldon Core deployment that is *equivalent* to the supplied
configuration parameters. Two or more Seldon Core deployments are
considered equivalent if they have the same `pipeline_name`,
`pipeline_step_name` and `model_name` configuration parameters. To
put it differently, two Seldon Core deployments are equivalent if
they serve versions of the same model deployed by the same pipeline
step. If an equivalent Seldon Core deployment is found, it will be
updated in place to reflect the new configuration parameters. This
allows an existing Seldon Core deployment to retain its prediction
URL while performing a rolling update to serve a new model version.
Callers should set `replace` to True if they want a continuous model
deployment workflow that doesn't spin up a new Seldon Core deployment
server for each new model version. If multiple equivalent Seldon Core
deployments are found, the most recently created deployment is selected
to be updated and the others are deleted.
Args:
config: the configuration of the model to be deployed with Seldon.
Core
replace: set this flag to True to find and update an equivalent
Seldon Core deployment server with the new model instead of
starting a new deployment server.
timeout: the timeout in seconds to wait for the Seldon Core server
to be provisioned and successfully started or updated. If set
to 0, the method will return immediately after the Seldon Core
server is provisioned, without waiting for it to fully start.
Returns:
The ZenML Seldon Core deployment service object that can be used to
interact with the remote Seldon Core server.
Raises:
SeldonClientError: if a Seldon Core client error is encountered
while provisioning the Seldon Core deployment server.
RuntimeError: if `timeout` is set to a positive value that is
exceeded while waiting for the Seldon Core deployment server
to start, or if an operational failure is encountered before
it reaches a ready state.
"""
config = cast(SeldonDeploymentConfig, config)
service = None
# if a custom Kubernetes secret is not explicitly specified in the
# SeldonDeploymentConfig, try to create one from the ZenML secret
# configured for the model deployer
config.secret_name = (
config.secret_name or self._create_or_update_kubernetes_secret()
)
# if replace is True, find equivalent Seldon Core deployments
if replace is True:
equivalent_services = self.find_model_server(
running=False,
pipeline_name=config.pipeline_name,
pipeline_step_name=config.pipeline_step_name,
model_name=config.model_name,
)
for equivalent_service in equivalent_services:
if service is None:
# keep the most recently created service
service = equivalent_service
else:
try:
# delete the older services and don't wait for them to
# be deprovisioned
service.stop()
except RuntimeError:
# ignore errors encountered while stopping old services
pass
if service:
# update an equivalent service in place
service.update(config)
logger.info(
f"Updating an existing Seldon deployment service: {service}"
)
else:
# create a new service
service = SeldonDeploymentService(config=config)
logger.info(f"Creating a new Seldon deployment service: {service}")
# start the service which in turn provisions the Seldon Core
# deployment server and waits for it to reach a ready state
service.start(timeout=timeout)
return service
def find_model_server(
self,
running: bool = False,
service_uuid: Optional[UUID] = None,
pipeline_name: Optional[str] = None,
pipeline_run_id: Optional[str] = None,
pipeline_step_name: Optional[str] = None,
model_name: Optional[str] = None,
model_uri: Optional[str] = None,
model_type: Optional[str] = None,
) -> List[BaseService]:
"""Find one or more Seldon Core model services that match th given
criteria.
The Seldon Core deployment services that meet the search criteria are
returned sorted in descending order of their creation time (i.e. more
recent deployments first).
Args:
running: if true, only running services will be returned.
service_uuid: the UUID of the Seldon Core service that was originally used
to create the Seldon Core deployment resource.
pipeline_name: name of the pipeline that the deployed model was part
of.
pipeline_run_id: ID of the pipeline run which the deployed model was
part of.
pipeline_step_name: the name of the pipeline model deployment step
that deployed the model.
model_name: the name of the deployed model.
model_uri: URI of the deployed model.
model_type: the Seldon Core server implementation used to serve
the model
Returns:
One or more Seldon Core service objects representing Seldon Core
model servers that match the input search criteria.
"""
# Use a Seldon deployment service configuration to compute the labels
config = SeldonDeploymentConfig(
pipeline_name=pipeline_name or "",
pipeline_run_id=pipeline_run_id or "",
pipeline_step_name=pipeline_step_name or "",
model_name=model_name or "",
model_uri=model_uri or "",
implementation=model_type or "",
)
labels = config.get_seldon_deployment_labels()
if service_uuid:
# the service UUID is not a label covered by the Seldon
# deployment service configuration, so we need to add it
# separately
labels["zenml.service_uuid"] = str(service_uuid)
deployments = self.seldon_client.find_deployments(labels=labels)
# sort the deployments in descending order of their creation time
deployments.sort(
key=lambda deployment: datetime.strptime(
deployment.metadata.creationTimestamp,
"%Y-%m-%dT%H:%M:%SZ",
)
if deployment.metadata.creationTimestamp
else datetime.min,
reverse=True,
)
services: List[BaseService] = []
for deployment in deployments:
# recreate the Seldon deployment service object from the Seldon
# deployment resource
service = SeldonDeploymentService.create_from_deployment(
deployment=deployment
)
if running and not service.is_running:
# skip non-running services
continue
services.append(service)
return services
def stop_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Stop a Seldon Core model server.
Args:
uuid: UUID of the model server to stop.
timeout: timeout in seconds to wait for the service to stop.
force: if True, force the service to stop.
"""
raise NotImplementedError(
"Stopping Seldon Core model servers is not implemented. Try "
"deleting the Seldon Core model server instead."
)
def start_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
) -> None:
"""Start a Seldon Core model deployment server.
Args:
uuid: UUID of the model server to start.
timeout: timeout in seconds to wait for the service to become
active. . If set to 0, the method will return immediately after
provisioning the service, without waiting for it to become
active.
"""
raise NotImplementedError(
"Starting Seldon Core model servers is not implemented"
)
def delete_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Delete a Seldon Core model deployment server.
Args:
uuid: UUID of the model server to delete.
timeout: timeout in seconds to wait for the service to stop. If
set to 0, the method will return immediately after
deprovisioning the service, without waiting for it to stop.
force: if True, force the service to stop.
"""
services = self.find_model_server(service_uuid=uuid)
if len(services) == 0:
return
services[0].stop(timeout=timeout, force=force)
# if this is the last Seldon Core model server, delete the Kubernetes
# secret used to store the authentication information for the Seldon
# Core model server storage initializer
self._delete_kubernetes_secret()
kubernetes_secret_name: Optional[str]
property
readonly
Get the Kubernetes secret name associated with this model deployer.
If a secret is configured for this model deployer, a corresponding Kubernetes secret is created in the remote cluster to be used by Seldon Core storage initializers to authenticate to the Artifact Store. This method returns the unique name that is used for this secret.
Returns:
| Type | Description |
|---|---|
Optional[str] |
The Seldon Core Kubernetes secret name, or None if no secret is configured. |
seldon_client: SeldonClient
property
readonly
Get the Seldon Core client associated with this model deployer.
Returns:
| Type | Description |
|---|---|
SeldonClient |
The Seldon Core client. |
Exceptions:
| Type | Description |
|---|---|
SeldonClientError |
if the Kubernetes client configuration cannot be found. |
delete_model_server(self, uuid, timeout=300, force=False)
Delete a Seldon Core model deployment server.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
uuid |
UUID |
UUID of the model server to delete. |
required |
timeout |
int |
timeout in seconds to wait for the service to stop. If set to 0, the method will return immediately after deprovisioning the service, without waiting for it to stop. |
300 |
force |
bool |
if True, force the service to stop. |
False |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
def delete_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Delete a Seldon Core model deployment server.
Args:
uuid: UUID of the model server to delete.
timeout: timeout in seconds to wait for the service to stop. If
set to 0, the method will return immediately after
deprovisioning the service, without waiting for it to stop.
force: if True, force the service to stop.
"""
services = self.find_model_server(service_uuid=uuid)
if len(services) == 0:
return
services[0].stop(timeout=timeout, force=force)
# if this is the last Seldon Core model server, delete the Kubernetes
# secret used to store the authentication information for the Seldon
# Core model server storage initializer
self._delete_kubernetes_secret()
deploy_model(self, config, replace=False, timeout=300)
Create a new Seldon Core deployment or update an existing one to serve the supplied model and deployment configuration.
This method has two modes of operation, depending on the replace
argument value:
-
if
replaceis False, calling this method will create a new Seldon Core deployment server to reflect the model and other configuration parameters specified in the supplied Seldon deploymentconfig. -
if
replaceis True, this method will first attempt to find an existing Seldon Core deployment that is equivalent to the supplied configuration parameters. Two or more Seldon Core deployments are considered equivalent if they have the samepipeline_name,pipeline_step_nameandmodel_nameconfiguration parameters. To put it differently, two Seldon Core deployments are equivalent if they serve versions of the same model deployed by the same pipeline step. If an equivalent Seldon Core deployment is found, it will be updated in place to reflect the new configuration parameters. This allows an existing Seldon Core deployment to retain its prediction URL while performing a rolling update to serve a new model version.
Callers should set replace to True if they want a continuous model
deployment workflow that doesn't spin up a new Seldon Core deployment
server for each new model version. If multiple equivalent Seldon Core
deployments are found, the most recently created deployment is selected
to be updated and the others are deleted.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
config |
ServiceConfig |
the configuration of the model to be deployed with Seldon. Core |
required |
replace |
bool |
set this flag to True to find and update an equivalent Seldon Core deployment server with the new model instead of starting a new deployment server. |
False |
timeout |
int |
the timeout in seconds to wait for the Seldon Core server to be provisioned and successfully started or updated. If set to 0, the method will return immediately after the Seldon Core server is provisioned, without waiting for it to fully start. |
300 |
Returns:
| Type | Description |
|---|---|
BaseService |
The ZenML Seldon Core deployment service object that can be used to interact with the remote Seldon Core server. |
Exceptions:
| Type | Description |
|---|---|
SeldonClientError |
if a Seldon Core client error is encountered while provisioning the Seldon Core deployment server. |
RuntimeError |
if |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
def deploy_model(
self,
config: ServiceConfig,
replace: bool = False,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
) -> BaseService:
"""Create a new Seldon Core deployment or update an existing one to
serve the supplied model and deployment configuration.
This method has two modes of operation, depending on the `replace`
argument value:
* if `replace` is False, calling this method will create a new Seldon
Core deployment server to reflect the model and other configuration
parameters specified in the supplied Seldon deployment `config`.
* if `replace` is True, this method will first attempt to find an
existing Seldon Core deployment that is *equivalent* to the supplied
configuration parameters. Two or more Seldon Core deployments are
considered equivalent if they have the same `pipeline_name`,
`pipeline_step_name` and `model_name` configuration parameters. To
put it differently, two Seldon Core deployments are equivalent if
they serve versions of the same model deployed by the same pipeline
step. If an equivalent Seldon Core deployment is found, it will be
updated in place to reflect the new configuration parameters. This
allows an existing Seldon Core deployment to retain its prediction
URL while performing a rolling update to serve a new model version.
Callers should set `replace` to True if they want a continuous model
deployment workflow that doesn't spin up a new Seldon Core deployment
server for each new model version. If multiple equivalent Seldon Core
deployments are found, the most recently created deployment is selected
to be updated and the others are deleted.
Args:
config: the configuration of the model to be deployed with Seldon.
Core
replace: set this flag to True to find and update an equivalent
Seldon Core deployment server with the new model instead of
starting a new deployment server.
timeout: the timeout in seconds to wait for the Seldon Core server
to be provisioned and successfully started or updated. If set
to 0, the method will return immediately after the Seldon Core
server is provisioned, without waiting for it to fully start.
Returns:
The ZenML Seldon Core deployment service object that can be used to
interact with the remote Seldon Core server.
Raises:
SeldonClientError: if a Seldon Core client error is encountered
while provisioning the Seldon Core deployment server.
RuntimeError: if `timeout` is set to a positive value that is
exceeded while waiting for the Seldon Core deployment server
to start, or if an operational failure is encountered before
it reaches a ready state.
"""
config = cast(SeldonDeploymentConfig, config)
service = None
# if a custom Kubernetes secret is not explicitly specified in the
# SeldonDeploymentConfig, try to create one from the ZenML secret
# configured for the model deployer
config.secret_name = (
config.secret_name or self._create_or_update_kubernetes_secret()
)
# if replace is True, find equivalent Seldon Core deployments
if replace is True:
equivalent_services = self.find_model_server(
running=False,
pipeline_name=config.pipeline_name,
pipeline_step_name=config.pipeline_step_name,
model_name=config.model_name,
)
for equivalent_service in equivalent_services:
if service is None:
# keep the most recently created service
service = equivalent_service
else:
try:
# delete the older services and don't wait for them to
# be deprovisioned
service.stop()
except RuntimeError:
# ignore errors encountered while stopping old services
pass
if service:
# update an equivalent service in place
service.update(config)
logger.info(
f"Updating an existing Seldon deployment service: {service}"
)
else:
# create a new service
service = SeldonDeploymentService(config=config)
logger.info(f"Creating a new Seldon deployment service: {service}")
# start the service which in turn provisions the Seldon Core
# deployment server and waits for it to reach a ready state
service.start(timeout=timeout)
return service
find_model_server(self, running=False, service_uuid=None, pipeline_name=None, pipeline_run_id=None, pipeline_step_name=None, model_name=None, model_uri=None, model_type=None)
Find one or more Seldon Core model services that match th given criteria.
The Seldon Core deployment services that meet the search criteria are returned sorted in descending order of their creation time (i.e. more recent deployments first).
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
running |
bool |
if true, only running services will be returned. |
False |
service_uuid |
Optional[uuid.UUID] |
the UUID of the Seldon Core service that was originally used to create the Seldon Core deployment resource. |
None |
pipeline_name |
Optional[str] |
name of the pipeline that the deployed model was part of. |
None |
pipeline_run_id |
Optional[str] |
ID of the pipeline run which the deployed model was part of. |
None |
pipeline_step_name |
Optional[str] |
the name of the pipeline model deployment step that deployed the model. |
None |
model_name |
Optional[str] |
the name of the deployed model. |
None |
model_uri |
Optional[str] |
URI of the deployed model. |
None |
model_type |
Optional[str] |
the Seldon Core server implementation used to serve the model |
None |
Returns:
| Type | Description |
|---|---|
List[zenml.services.service.BaseService] |
One or more Seldon Core service objects representing Seldon Core model servers that match the input search criteria. |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
def find_model_server(
self,
running: bool = False,
service_uuid: Optional[UUID] = None,
pipeline_name: Optional[str] = None,
pipeline_run_id: Optional[str] = None,
pipeline_step_name: Optional[str] = None,
model_name: Optional[str] = None,
model_uri: Optional[str] = None,
model_type: Optional[str] = None,
) -> List[BaseService]:
"""Find one or more Seldon Core model services that match th given
criteria.
The Seldon Core deployment services that meet the search criteria are
returned sorted in descending order of their creation time (i.e. more
recent deployments first).
Args:
running: if true, only running services will be returned.
service_uuid: the UUID of the Seldon Core service that was originally used
to create the Seldon Core deployment resource.
pipeline_name: name of the pipeline that the deployed model was part
of.
pipeline_run_id: ID of the pipeline run which the deployed model was
part of.
pipeline_step_name: the name of the pipeline model deployment step
that deployed the model.
model_name: the name of the deployed model.
model_uri: URI of the deployed model.
model_type: the Seldon Core server implementation used to serve
the model
Returns:
One or more Seldon Core service objects representing Seldon Core
model servers that match the input search criteria.
"""
# Use a Seldon deployment service configuration to compute the labels
config = SeldonDeploymentConfig(
pipeline_name=pipeline_name or "",
pipeline_run_id=pipeline_run_id or "",
pipeline_step_name=pipeline_step_name or "",
model_name=model_name or "",
model_uri=model_uri or "",
implementation=model_type or "",
)
labels = config.get_seldon_deployment_labels()
if service_uuid:
# the service UUID is not a label covered by the Seldon
# deployment service configuration, so we need to add it
# separately
labels["zenml.service_uuid"] = str(service_uuid)
deployments = self.seldon_client.find_deployments(labels=labels)
# sort the deployments in descending order of their creation time
deployments.sort(
key=lambda deployment: datetime.strptime(
deployment.metadata.creationTimestamp,
"%Y-%m-%dT%H:%M:%SZ",
)
if deployment.metadata.creationTimestamp
else datetime.min,
reverse=True,
)
services: List[BaseService] = []
for deployment in deployments:
# recreate the Seldon deployment service object from the Seldon
# deployment resource
service = SeldonDeploymentService.create_from_deployment(
deployment=deployment
)
if running and not service.is_running:
# skip non-running services
continue
services.append(service)
return services
get_active_model_deployer()
staticmethod
Get the Seldon Core model deployer registered in the active stack.
Returns:
| Type | Description |
|---|---|
SeldonModelDeployer |
The Seldon Core model deployer registered in the active stack. |
Exceptions:
| Type | Description |
|---|---|
TypeError |
if the Seldon Core model deployer is not available. |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
@staticmethod
def get_active_model_deployer() -> "SeldonModelDeployer":
"""Get the Seldon Core model deployer registered in the active stack.
Returns:
The Seldon Core model deployer registered in the active stack.
Raises:
TypeError: if the Seldon Core model deployer is not available.
"""
model_deployer = Repository( # type: ignore [call-arg]
skip_repository_check=True
).active_stack.model_deployer
if not model_deployer or not isinstance(
model_deployer, SeldonModelDeployer
):
raise TypeError(
"The active stack needs to have a Seldon Core model deployer "
"component registered to be able to deploy models with Seldon "
"Core. You can create a new stack with a Seldon Core model "
"deployer component or update your existing stack to add this "
"component, e.g.:\n\n"
" 'zenml model-deployer register seldon --type=seldon "
"--kubernetes_context=context-name --kubernetes_namespace="
"namespace-name --base_url=https://ingress.cluster.kubernetes'\n"
" 'zenml stack create stack-name -d seldon ...'\n"
)
return model_deployer
get_model_server_info(service_instance)
staticmethod
"Return implementation specific information that might be relevant to the user.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
service_instance |
SeldonDeploymentService |
Instance of a SeldonDeploymentService |
required |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
@staticmethod
def get_model_server_info( # type: ignore[override]
service_instance: "SeldonDeploymentService",
) -> Dict[str, Optional[str]]:
""" "Return implementation specific information that might be relevant
to the user.
Args:
service_instance: Instance of a SeldonDeploymentService
"""
return {
"PREDICTION_URL": service_instance.prediction_url,
"MODEL_URI": service_instance.config.model_uri,
"MODEL_NAME": service_instance.config.model_name,
"SELDON_DEPLOYMENT": service_instance.seldon_deployment_name,
}
start_model_server(self, uuid, timeout=300)
Start a Seldon Core model deployment server.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
uuid |
UUID |
UUID of the model server to start. |
required |
timeout |
int |
timeout in seconds to wait for the service to become active. . If set to 0, the method will return immediately after provisioning the service, without waiting for it to become active. |
300 |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
def start_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
) -> None:
"""Start a Seldon Core model deployment server.
Args:
uuid: UUID of the model server to start.
timeout: timeout in seconds to wait for the service to become
active. . If set to 0, the method will return immediately after
provisioning the service, without waiting for it to become
active.
"""
raise NotImplementedError(
"Starting Seldon Core model servers is not implemented"
)
stop_model_server(self, uuid, timeout=300, force=False)
Stop a Seldon Core model server.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
uuid |
UUID |
UUID of the model server to stop. |
required |
timeout |
int |
timeout in seconds to wait for the service to stop. |
300 |
force |
bool |
if True, force the service to stop. |
False |
Source code in zenml/integrations/seldon/model_deployers/seldon_model_deployer.py
def stop_model_server(
self,
uuid: UUID,
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT,
force: bool = False,
) -> None:
"""Stop a Seldon Core model server.
Args:
uuid: UUID of the model server to stop.
timeout: timeout in seconds to wait for the service to stop.
force: if True, force the service to stop.
"""
raise NotImplementedError(
"Stopping Seldon Core model servers is not implemented. Try "
"deleting the Seldon Core model server instead."
)
secret_schemas
special
Seldon Secret Schemas
These are secret schemas that can be used to authenticate Seldon to the Artifact Store used to store served ML models.
secret_schemas
SeldonAzureSecretSchema (BaseSecretSchema)
pydantic-model
Seldon Azure Blob Storage credentials.
Based on: https://rclone.org/azureblob/
Attributes:
| Name | Type | Description |
|---|---|---|
rclone_config_azureblob_type |
Literal['azureblob'] |
the rclone config type. Must be set to "azureblob" for this schema. |
rclone_config_azureblob_account |
Optional[str] |
storage Account Name. Leave blank to use SAS URL or MSI. |
rclone_config_azureblob_key |
Optional[str] |
storage Account Key. Leave blank to use SAS URL or MSI. |
rclone_config_azureblob_sas_url |
Optional[str] |
SAS URL for container level access only. Leave blank if using account/key or MSI. |
rclone_config_azureblob_use_msi |
bool |
use a managed service identity to authenticate (only works in Azure). |
Source code in zenml/integrations/seldon/secret_schemas/secret_schemas.py
class SeldonAzureSecretSchema(BaseSecretSchema):
"""Seldon Azure Blob Storage credentials.
Based on: https://rclone.org/azureblob/
Attributes:
rclone_config_azureblob_type: the rclone config type. Must be set to
"azureblob" for this schema.
rclone_config_azureblob_account: storage Account Name. Leave blank to
use SAS URL or MSI.
rclone_config_azureblob_key: storage Account Key. Leave blank to
use SAS URL or MSI.
rclone_config_azureblob_sas_url: SAS URL for container level access
only. Leave blank if using account/key or MSI.
rclone_config_azureblob_use_msi: use a managed service identity to
authenticate (only works in Azure).
"""
TYPE: ClassVar[str] = SELDON_AZUREBLOB_SECRET_SCHEMA_TYPE
rclone_config_azureblob_type: Literal["azureblob"] = "azureblob"
rclone_config_azureblob_account: Optional[str]
rclone_config_azureblob_key: Optional[str]
rclone_config_azureblob_sas_url: Optional[str]
rclone_config_azureblob_use_msi: bool = False
SeldonGSSecretSchema (BaseSecretSchema)
pydantic-model
Seldon GCS credentials.
Based on: https://rclone.org/googlecloudstorage/
Attributes:
| Name | Type | Description |
|---|---|---|
rclone_config_gs_type |
Literal['google cloud storage'] |
the rclone config type. Must be set to "google cloud storage" for this schema. |
rclone_config_gs_client_id |
Optional[str] |
OAuth client id. |
rclone_config_gs_client_secret |
Optional[str] |
OAuth client secret. |
rclone_config_gs_token |
Optional[str] |
OAuth Access Token as a JSON blob. |
rclone_config_gs_project_number |
Optional[str] |
project number. |
rclone_config_gs_service_account_credentials |
Optional[str] |
service account credentials JSON blob. |
rclone_config_gs_anonymous |
bool |
access public buckets and objects without credentials. Set to True if you just want to download files and don't configure credentials. |
rclone_config_gs_auth_url |
Optional[str] |
auth server URL. |
Source code in zenml/integrations/seldon/secret_schemas/secret_schemas.py
class SeldonGSSecretSchema(BaseSecretSchema):
"""Seldon GCS credentials.
Based on: https://rclone.org/googlecloudstorage/
Attributes:
rclone_config_gs_type: the rclone config type. Must be set to "google
cloud storage" for this schema.
rclone_config_gs_client_id: OAuth client id.
rclone_config_gs_client_secret: OAuth client secret.
rclone_config_gs_token: OAuth Access Token as a JSON blob.
rclone_config_gs_project_number: project number.
rclone_config_gs_service_account_credentials: service account
credentials JSON blob.
rclone_config_gs_anonymous: access public buckets and objects without
credentials. Set to True if you just want to download files and
don't configure credentials.
rclone_config_gs_auth_url: auth server URL.
"""
TYPE: ClassVar[str] = SELDON_GS_SECRET_SCHEMA_TYPE
rclone_config_gs_type: Literal[
"google cloud storage"
] = "google cloud storage"
rclone_config_gs_client_id: Optional[str]
rclone_config_gs_client_secret: Optional[str]
rclone_config_gs_project_number: Optional[str]
rclone_config_gs_service_account_credentials: Optional[str]
rclone_config_gs_anonymous: bool = False
rclone_config_gs_token: Optional[str]
rclone_config_gs_auth_url: Optional[str]
rclone_config_gs_token_url: Optional[str]
SeldonS3SecretSchema (BaseSecretSchema)
pydantic-model
Seldon S3 credentials.
Based on: https://rclone.org/s3/#amazon-s3
Attributes:
| Name | Type | Description |
|---|---|---|
rclone_config_s3_type |
Literal['s3'] |
the rclone config type. Must be set to "s3" for this schema. |
rclone_config_s3_provider |
str |
the S3 provider (e.g. aws, ceph, minio). |
rclone_config_s3_env_auth |
bool |
get AWS credentials from EC2/ECS meta data (i.e. with IAM roles configuration). Only applies if access_key_id and secret_access_key are blank. |
rclone_config_s3_access_key_id |
Optional[str] |
AWS Access Key ID. |
rclone_config_s3_secret_access_key |
Optional[str] |
AWS Secret Access Key. |
rclone_config_s3_session_token |
Optional[str] |
AWS Session Token. |
rclone_config_s3_region |
Optional[str] |
region to connect to. |
rclone_config_s3_endpoint |
Optional[str] |
S3 API endpoint. |
Source code in zenml/integrations/seldon/secret_schemas/secret_schemas.py
class SeldonS3SecretSchema(BaseSecretSchema):
"""Seldon S3 credentials.
Based on: https://rclone.org/s3/#amazon-s3
Attributes:
rclone_config_s3_type: the rclone config type. Must be set to "s3" for
this schema.
rclone_config_s3_provider: the S3 provider (e.g. aws, ceph, minio).
rclone_config_s3_env_auth: get AWS credentials from EC2/ECS meta data
(i.e. with IAM roles configuration). Only applies if access_key_id
and secret_access_key are blank.
rclone_config_s3_access_key_id: AWS Access Key ID.
rclone_config_s3_secret_access_key: AWS Secret Access Key.
rclone_config_s3_session_token: AWS Session Token.
rclone_config_s3_region: region to connect to.
rclone_config_s3_endpoint: S3 API endpoint.
"""
TYPE: ClassVar[str] = SELDON_S3_SECRET_SCHEMA_TYPE
rclone_config_s3_type: Literal["s3"] = "s3"
rclone_config_s3_provider: str = "aws"
rclone_config_s3_env_auth: bool = False
rclone_config_s3_access_key_id: Optional[str]
rclone_config_s3_secret_access_key: Optional[str]
rclone_config_s3_session_token: Optional[str]
rclone_config_s3_region: Optional[str]
rclone_config_s3_endpoint: Optional[str]
seldon_client
SeldonClient
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonClient:
def __init__(self, context: Optional[str], namespace: Optional[str]):
"""Initialize a Seldon Core client.
Args:
context: the Kubernetes context to use.
namespace: the Kubernetes namespace to use.
"""
self._context = context
self._namespace = namespace
self._initialize_k8s_clients()
def _initialize_k8s_clients(self) -> None:
"""Initialize the Kubernetes clients."""
try:
k8s_config.load_incluster_config()
if not self._namespace:
# load the namespace in the context of which the
# current pod is running
self._namespace = open(
"/var/run/secrets/kubernetes.io/serviceaccount/namespace"
).read()
except k8s_config.config_exception.ConfigException:
if not self._namespace:
raise SeldonClientError(
"The Kubernetes namespace must be explicitly "
"configured when running outside of a cluster."
)
try:
k8s_config.load_kube_config(
context=self._context, persist_config=False
)
except k8s_config.config_exception.ConfigException as e:
raise SeldonClientError(
"Could not load the Kubernetes configuration"
) from e
self._core_api = k8s_client.CoreV1Api()
self._custom_objects_api = k8s_client.CustomObjectsApi()
@staticmethod
def sanitize_labels(labels: Dict[str, str]) -> None:
"""Update the label values to be valid Kubernetes labels.
See: https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/#syntax-and-character-set
"""
for key, value in labels.items():
# Kubernetes labels must be alphanumeric, no longer than
# 63 characters, and must begin and end with an alphanumeric
# character ([a-z0-9A-Z])
labels[key] = re.sub(r"[^0-9a-zA-Z-_\.]+", "_", value)[:63].strip(
"-_."
)
@property
def namespace(self) -> str:
"""Returns the Kubernetes namespace in use by the client.
Returns:
The Kubernetes namespace in use by the client.
"""
if not self._namespace:
# shouldn't happen if the client is initialized, but we need to
# appease the mypy type checker
raise RuntimeError("The Kubernetes namespace is not configured")
return self._namespace
def create_deployment(
self,
deployment: SeldonDeployment,
poll_timeout: int = 0,
) -> SeldonDeployment:
"""Create a Seldon Core deployment resource.
Args:
deployment: the Seldon Core deployment resource to create
poll_timeout: the maximum time to wait for the deployment to become
available or to fail. If set to 0, the function will return
immediately without checking the deployment status. If a timeout
occurs and the deployment is still pending creation, it will
be returned anyway and no exception will be raised.
Returns:
the created Seldon Core deployment resource with updated status.
Raises:
SeldonDeploymentExistsError: if a deployment with the same name
already exists.
SeldonDeploymentNotFoundError: if the newly created deployment
resource cannot be found.
SeldonClientError: if an unknown error occurs during the creation of
the deployment.
"""
try:
logger.debug(f"Creating SeldonDeployment resource: {deployment}")
# mark the deployment as managed by ZenML, to differentiate
# between deployments that are created by ZenML and those that
# are not
deployment.mark_as_managed_by_zenml()
response = self._custom_objects_api.create_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
body=deployment.dict(exclude_none=True),
_request_timeout=poll_timeout or None,
)
logger.debug("Seldon Core API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when creating SeldonDeployment resource: %s", str(e)
)
if e.status == 409:
raise SeldonDeploymentExistsError(
f"A deployment with the name {deployment.name} "
f"already exists in namespace {self._namespace}"
)
raise SeldonClientError(
"Exception when creating SeldonDeployment resource"
) from e
created_deployment = self.get_deployment(name=deployment.name)
while poll_timeout > 0 and created_deployment.is_pending():
time.sleep(5)
poll_timeout -= 5
created_deployment = self.get_deployment(name=deployment.name)
return created_deployment
def delete_deployment(
self,
name: str,
force: bool = False,
poll_timeout: int = 0,
) -> None:
"""Delete a Seldon Core deployment resource managed by ZenML.
Args:
name: the name of the Seldon Core deployment resource to delete.
force: if True, the deployment deletion will be forced (the graceful
period will be set to zero).
poll_timeout: the maximum time to wait for the deployment to be
deleted. If set to 0, the function will return immediately
without checking the deployment status. If a timeout
occurs and the deployment still exists, this method will
return and no exception will be raised.
Raises:
SeldonDeploymentNotFoundError: if the deployment resource cannot be
found or is not managed by ZenML.
SeldonClientError: if an unknown error occurs during the deployment
removal.
"""
try:
logger.debug(f"Deleting SeldonDeployment resource: {name}")
# call `get_deployment` to check that the deployment exists
# and is managed by ZenML. It will raise
# a SeldonDeploymentNotFoundError otherwise
self.get_deployment(name=name)
response = self._custom_objects_api.delete_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
name=name,
_request_timeout=poll_timeout or None,
grace_period_seconds=0 if force else None,
)
logger.debug("Seldon Core API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when deleting SeldonDeployment resource %s: %s",
name,
str(e),
)
raise SeldonClientError(
f"Exception when deleting SeldonDeployment resource {name}"
) from e
while poll_timeout > 0:
try:
self.get_deployment(name=name)
except SeldonDeploymentNotFoundError:
return
time.sleep(5)
poll_timeout -= 5
def update_deployment(
self,
deployment: SeldonDeployment,
poll_timeout: int = 0,
) -> SeldonDeployment:
"""Update a Seldon Core deployment resource.
Args:
deployment: the Seldon Core deployment resource to update
poll_timeout: the maximum time to wait for the deployment to become
available or to fail. If set to 0, the function will return
immediately without checking the deployment status. If a timeout
occurs and the deployment is still pending creation, it will
be returned anyway and no exception will be raised.
Returns:
the updated Seldon Core deployment resource with updated status.
Raises:
SeldonDeploymentNotFoundError: if deployment resource with the given
name cannot be found.
SeldonClientError: if an unknown error occurs while updating the
deployment.
"""
try:
logger.debug(
f"Updating SeldonDeployment resource: {deployment.name}"
)
# mark the deployment as managed by ZenML, to differentiate
# between deployments that are created by ZenML and those that
# are not
deployment.mark_as_managed_by_zenml()
# call `get_deployment` to check that the deployment exists
# and is managed by ZenML. It will raise
# a SeldonDeploymentNotFoundError otherwise
self.get_deployment(name=deployment.name)
response = self._custom_objects_api.patch_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
name=deployment.name,
body=deployment.dict(exclude_none=True),
_request_timeout=poll_timeout or None,
)
logger.debug("Seldon Core API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when updating SeldonDeployment resource: %s", str(e)
)
raise SeldonClientError(
"Exception when creating SeldonDeployment resource"
) from e
updated_deployment = self.get_deployment(name=deployment.name)
while poll_timeout > 0 and updated_deployment.is_pending():
time.sleep(5)
poll_timeout -= 5
updated_deployment = self.get_deployment(name=deployment.name)
return updated_deployment
def get_deployment(self, name: str) -> SeldonDeployment:
"""Get a ZenML managed Seldon Core deployment resource by name.
Args:
name: the name of the Seldon Core deployment resource to fetch.
Returns:
The Seldon Core deployment resource.
Raises:
SeldonDeploymentNotFoundError: if the deployment resource cannot
be found or is not managed by ZenML.
SeldonClientError: if an unknown error occurs while fetching
the deployment.
"""
try:
logger.debug(f"Retrieving SeldonDeployment resource: {name}")
response = self._custom_objects_api.get_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
name=name,
)
logger.debug("Seldon Core API response: %s", response)
try:
deployment = SeldonDeployment(**response)
except ValidationError as e:
logger.error(
"Invalid Seldon Core deployment resource: %s\n%s",
str(e),
str(response),
)
raise SeldonDeploymentNotFoundError(
f"SeldonDeployment resource {name} could not be parsed"
)
# Only Seldon deployments managed by ZenML are returned
if not deployment.is_managed_by_zenml():
raise SeldonDeploymentNotFoundError(
f"Seldon Deployment {name} is not managed by ZenML"
)
return deployment
except k8s_client.rest.ApiException as e:
if e.status == 404:
raise SeldonDeploymentNotFoundError(
f"SeldonDeployment resource not found: {name}"
) from e
logger.error(
"Exception when fetching SeldonDeployment resource %s: %s",
name,
str(e),
)
raise SeldonClientError(
f"Unexpected exception when fetching SeldonDeployment "
f"resource: {name}"
) from e
def find_deployments(
self,
name: Optional[str] = None,
labels: Optional[Dict[str, str]] = None,
fields: Optional[Dict[str, str]] = None,
) -> List[SeldonDeployment]:
"""Find all ZenML managed Seldon Core deployment resources that match
the given criteria.
Args:
name: optional name of the deployment resource to find.
fields: optional selector to restrict the list of returned
Seldon deployments by their fields. Defaults to everything.
labels: optional selector to restrict the list of returned
Seldon deployments by their labels. Defaults to everything.
Returns:
List of Seldon Core deployments that match the given criteria.
Raises:
SeldonClientError: if an unknown error occurs while fetching
the deployments.
"""
fields = fields or {}
labels = labels or {}
# always filter results to only include Seldon deployments managed
# by ZenML
labels["app"] = "zenml"
if name:
fields = {"metadata.name": name}
field_selector = (
",".join(f"{k}={v}" for k, v in fields.items()) if fields else None
)
label_selector = (
",".join(f"{k}={v}" for k, v in labels.items()) if labels else None
)
try:
logger.debug(
f"Searching SeldonDeployment resources with label selector "
f"'{labels or ''}' and field selector '{fields or ''}'"
)
response = self._custom_objects_api.list_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
field_selector=field_selector,
label_selector=label_selector,
)
logger.debug(
"Seldon Core API returned %s items", len(response["items"])
)
deployments = []
for item in response.get("items") or []:
try:
deployments.append(SeldonDeployment(**item))
except ValidationError as e:
logger.error(
"Invalid Seldon Core deployment resource: %s\n%s",
str(e),
str(item),
)
return deployments
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when searching SeldonDeployment resources with "
"label selector '%s' and field selector '%s': %s",
label_selector or "",
field_selector or "",
)
raise SeldonClientError(
f"Unexpected exception when searching SeldonDeployment "
f"with labels '{labels or ''}' and field '{fields or ''}'"
) from e
def create_or_update_secret(
self,
name: str,
secret: BaseSecretSchema,
) -> None:
"""Create or update a Kubernetes Secret resource with the information
contained in a ZenML secret.
Args:
name: the name of the Secret resource to create.
secret: a ZenML secret with key-values that should be
stored in the Secret resource.
Raises:
SeldonClientError: if an unknown error occurs during the creation of
the secret.
"""
try:
logger.debug(f"Creating Secret resource: {name}")
secret_data = {
k.upper(): base64.b64encode(str(v).encode("utf-8")).decode(
"ascii"
)
for k, v in secret.content.items()
if v is not None
}
secret = k8s_client.V1Secret(
metadata=k8s_client.V1ObjectMeta(
name=name,
labels={"app": "zenml"},
),
type="Opaque",
data=secret_data,
)
try:
# check if the secret is already present
self._core_api.read_namespaced_secret(
name=name,
namespace=self._namespace,
)
# if we got this far, the secret is already present, update it
# in place
response = self._core_api.replace_namespaced_secret(
name=name,
namespace=self._namespace,
body=secret,
)
except k8s_client.rest.ApiException as e:
if e.status != 404:
# if an error other than 404 is raised here, treat it
# as an unexpected error
raise
response = self._core_api.create_namespaced_secret(
namespace=self._namespace,
body=secret,
)
logger.debug("Kubernetes API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error("Exception when creating Secret resource: %s", str(e))
raise SeldonClientError(
"Exception when creating Secret resource"
) from e
def delete_secret(
self,
name: str,
) -> None:
"""Delete a Kubernetes Secret resource managed by ZenML.
Args:
name: the name of the Kubernetes Secret resource to delete.
Raises:
SeldonClientError: if an unknown error occurs during the removal
of the secret.
"""
try:
logger.debug(f"Deleting Secret resource: {name}")
response = self._core_api.delete_namespaced_secret(
name=name,
namespace=self._namespace,
)
logger.debug("Kubernetes API response: %s", response)
except k8s_client.rest.ApiException as e:
if e.status == 404:
# the secret is no longer present, nothing to do
return
logger.error(
"Exception when deleting Secret resource %s: %s",
name,
str(e),
)
raise SeldonClientError(
f"Exception when deleting Secret resource {name}"
) from e
namespace: str
property
readonly
Returns the Kubernetes namespace in use by the client.
Returns:
| Type | Description |
|---|---|
str |
The Kubernetes namespace in use by the client. |
__init__(self, context, namespace)
special
Initialize a Seldon Core client.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
context |
Optional[str] |
the Kubernetes context to use. |
required |
namespace |
Optional[str] |
the Kubernetes namespace to use. |
required |
Source code in zenml/integrations/seldon/seldon_client.py
def __init__(self, context: Optional[str], namespace: Optional[str]):
"""Initialize a Seldon Core client.
Args:
context: the Kubernetes context to use.
namespace: the Kubernetes namespace to use.
"""
self._context = context
self._namespace = namespace
self._initialize_k8s_clients()
create_deployment(self, deployment, poll_timeout=0)
Create a Seldon Core deployment resource.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
deployment |
SeldonDeployment |
the Seldon Core deployment resource to create |
required |
poll_timeout |
int |
the maximum time to wait for the deployment to become available or to fail. If set to 0, the function will return immediately without checking the deployment status. If a timeout occurs and the deployment is still pending creation, it will be returned anyway and no exception will be raised. |
0 |
Returns:
| Type | Description |
|---|---|
SeldonDeployment |
the created Seldon Core deployment resource with updated status. |
Exceptions:
| Type | Description |
|---|---|
SeldonDeploymentExistsError |
if a deployment with the same name already exists. |
SeldonDeploymentNotFoundError |
if the newly created deployment resource cannot be found. |
SeldonClientError |
if an unknown error occurs during the creation of the deployment. |
Source code in zenml/integrations/seldon/seldon_client.py
def create_deployment(
self,
deployment: SeldonDeployment,
poll_timeout: int = 0,
) -> SeldonDeployment:
"""Create a Seldon Core deployment resource.
Args:
deployment: the Seldon Core deployment resource to create
poll_timeout: the maximum time to wait for the deployment to become
available or to fail. If set to 0, the function will return
immediately without checking the deployment status. If a timeout
occurs and the deployment is still pending creation, it will
be returned anyway and no exception will be raised.
Returns:
the created Seldon Core deployment resource with updated status.
Raises:
SeldonDeploymentExistsError: if a deployment with the same name
already exists.
SeldonDeploymentNotFoundError: if the newly created deployment
resource cannot be found.
SeldonClientError: if an unknown error occurs during the creation of
the deployment.
"""
try:
logger.debug(f"Creating SeldonDeployment resource: {deployment}")
# mark the deployment as managed by ZenML, to differentiate
# between deployments that are created by ZenML and those that
# are not
deployment.mark_as_managed_by_zenml()
response = self._custom_objects_api.create_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
body=deployment.dict(exclude_none=True),
_request_timeout=poll_timeout or None,
)
logger.debug("Seldon Core API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when creating SeldonDeployment resource: %s", str(e)
)
if e.status == 409:
raise SeldonDeploymentExistsError(
f"A deployment with the name {deployment.name} "
f"already exists in namespace {self._namespace}"
)
raise SeldonClientError(
"Exception when creating SeldonDeployment resource"
) from e
created_deployment = self.get_deployment(name=deployment.name)
while poll_timeout > 0 and created_deployment.is_pending():
time.sleep(5)
poll_timeout -= 5
created_deployment = self.get_deployment(name=deployment.name)
return created_deployment
create_or_update_secret(self, name, secret)
Create or update a Kubernetes Secret resource with the information contained in a ZenML secret.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name |
str |
the name of the Secret resource to create. |
required |
secret |
BaseSecretSchema |
a ZenML secret with key-values that should be stored in the Secret resource. |
required |
Exceptions:
| Type | Description |
|---|---|
SeldonClientError |
if an unknown error occurs during the creation of the secret. |
Source code in zenml/integrations/seldon/seldon_client.py
def create_or_update_secret(
self,
name: str,
secret: BaseSecretSchema,
) -> None:
"""Create or update a Kubernetes Secret resource with the information
contained in a ZenML secret.
Args:
name: the name of the Secret resource to create.
secret: a ZenML secret with key-values that should be
stored in the Secret resource.
Raises:
SeldonClientError: if an unknown error occurs during the creation of
the secret.
"""
try:
logger.debug(f"Creating Secret resource: {name}")
secret_data = {
k.upper(): base64.b64encode(str(v).encode("utf-8")).decode(
"ascii"
)
for k, v in secret.content.items()
if v is not None
}
secret = k8s_client.V1Secret(
metadata=k8s_client.V1ObjectMeta(
name=name,
labels={"app": "zenml"},
),
type="Opaque",
data=secret_data,
)
try:
# check if the secret is already present
self._core_api.read_namespaced_secret(
name=name,
namespace=self._namespace,
)
# if we got this far, the secret is already present, update it
# in place
response = self._core_api.replace_namespaced_secret(
name=name,
namespace=self._namespace,
body=secret,
)
except k8s_client.rest.ApiException as e:
if e.status != 404:
# if an error other than 404 is raised here, treat it
# as an unexpected error
raise
response = self._core_api.create_namespaced_secret(
namespace=self._namespace,
body=secret,
)
logger.debug("Kubernetes API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error("Exception when creating Secret resource: %s", str(e))
raise SeldonClientError(
"Exception when creating Secret resource"
) from e
delete_deployment(self, name, force=False, poll_timeout=0)
Delete a Seldon Core deployment resource managed by ZenML.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name |
str |
the name of the Seldon Core deployment resource to delete. |
required |
force |
bool |
if True, the deployment deletion will be forced (the graceful period will be set to zero). |
False |
poll_timeout |
int |
the maximum time to wait for the deployment to be deleted. If set to 0, the function will return immediately without checking the deployment status. If a timeout occurs and the deployment still exists, this method will return and no exception will be raised. |
0 |
Exceptions:
| Type | Description |
|---|---|
SeldonDeploymentNotFoundError |
if the deployment resource cannot be found or is not managed by ZenML. |
SeldonClientError |
if an unknown error occurs during the deployment removal. |
Source code in zenml/integrations/seldon/seldon_client.py
def delete_deployment(
self,
name: str,
force: bool = False,
poll_timeout: int = 0,
) -> None:
"""Delete a Seldon Core deployment resource managed by ZenML.
Args:
name: the name of the Seldon Core deployment resource to delete.
force: if True, the deployment deletion will be forced (the graceful
period will be set to zero).
poll_timeout: the maximum time to wait for the deployment to be
deleted. If set to 0, the function will return immediately
without checking the deployment status. If a timeout
occurs and the deployment still exists, this method will
return and no exception will be raised.
Raises:
SeldonDeploymentNotFoundError: if the deployment resource cannot be
found or is not managed by ZenML.
SeldonClientError: if an unknown error occurs during the deployment
removal.
"""
try:
logger.debug(f"Deleting SeldonDeployment resource: {name}")
# call `get_deployment` to check that the deployment exists
# and is managed by ZenML. It will raise
# a SeldonDeploymentNotFoundError otherwise
self.get_deployment(name=name)
response = self._custom_objects_api.delete_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
name=name,
_request_timeout=poll_timeout or None,
grace_period_seconds=0 if force else None,
)
logger.debug("Seldon Core API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when deleting SeldonDeployment resource %s: %s",
name,
str(e),
)
raise SeldonClientError(
f"Exception when deleting SeldonDeployment resource {name}"
) from e
while poll_timeout > 0:
try:
self.get_deployment(name=name)
except SeldonDeploymentNotFoundError:
return
time.sleep(5)
poll_timeout -= 5
delete_secret(self, name)
Delete a Kubernetes Secret resource managed by ZenML.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name |
str |
the name of the Kubernetes Secret resource to delete. |
required |
Exceptions:
| Type | Description |
|---|---|
SeldonClientError |
if an unknown error occurs during the removal of the secret. |
Source code in zenml/integrations/seldon/seldon_client.py
def delete_secret(
self,
name: str,
) -> None:
"""Delete a Kubernetes Secret resource managed by ZenML.
Args:
name: the name of the Kubernetes Secret resource to delete.
Raises:
SeldonClientError: if an unknown error occurs during the removal
of the secret.
"""
try:
logger.debug(f"Deleting Secret resource: {name}")
response = self._core_api.delete_namespaced_secret(
name=name,
namespace=self._namespace,
)
logger.debug("Kubernetes API response: %s", response)
except k8s_client.rest.ApiException as e:
if e.status == 404:
# the secret is no longer present, nothing to do
return
logger.error(
"Exception when deleting Secret resource %s: %s",
name,
str(e),
)
raise SeldonClientError(
f"Exception when deleting Secret resource {name}"
) from e
find_deployments(self, name=None, labels=None, fields=None)
Find all ZenML managed Seldon Core deployment resources that match the given criteria.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name |
Optional[str] |
optional name of the deployment resource to find. |
None |
fields |
Optional[Dict[str, str]] |
optional selector to restrict the list of returned Seldon deployments by their fields. Defaults to everything. |
None |
labels |
Optional[Dict[str, str]] |
optional selector to restrict the list of returned Seldon deployments by their labels. Defaults to everything. |
None |
Returns:
| Type | Description |
|---|---|
List[zenml.integrations.seldon.seldon_client.SeldonDeployment] |
List of Seldon Core deployments that match the given criteria. |
Exceptions:
| Type | Description |
|---|---|
SeldonClientError |
if an unknown error occurs while fetching the deployments. |
Source code in zenml/integrations/seldon/seldon_client.py
def find_deployments(
self,
name: Optional[str] = None,
labels: Optional[Dict[str, str]] = None,
fields: Optional[Dict[str, str]] = None,
) -> List[SeldonDeployment]:
"""Find all ZenML managed Seldon Core deployment resources that match
the given criteria.
Args:
name: optional name of the deployment resource to find.
fields: optional selector to restrict the list of returned
Seldon deployments by their fields. Defaults to everything.
labels: optional selector to restrict the list of returned
Seldon deployments by their labels. Defaults to everything.
Returns:
List of Seldon Core deployments that match the given criteria.
Raises:
SeldonClientError: if an unknown error occurs while fetching
the deployments.
"""
fields = fields or {}
labels = labels or {}
# always filter results to only include Seldon deployments managed
# by ZenML
labels["app"] = "zenml"
if name:
fields = {"metadata.name": name}
field_selector = (
",".join(f"{k}={v}" for k, v in fields.items()) if fields else None
)
label_selector = (
",".join(f"{k}={v}" for k, v in labels.items()) if labels else None
)
try:
logger.debug(
f"Searching SeldonDeployment resources with label selector "
f"'{labels or ''}' and field selector '{fields or ''}'"
)
response = self._custom_objects_api.list_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
field_selector=field_selector,
label_selector=label_selector,
)
logger.debug(
"Seldon Core API returned %s items", len(response["items"])
)
deployments = []
for item in response.get("items") or []:
try:
deployments.append(SeldonDeployment(**item))
except ValidationError as e:
logger.error(
"Invalid Seldon Core deployment resource: %s\n%s",
str(e),
str(item),
)
return deployments
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when searching SeldonDeployment resources with "
"label selector '%s' and field selector '%s': %s",
label_selector or "",
field_selector or "",
)
raise SeldonClientError(
f"Unexpected exception when searching SeldonDeployment "
f"with labels '{labels or ''}' and field '{fields or ''}'"
) from e
get_deployment(self, name)
Get a ZenML managed Seldon Core deployment resource by name.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name |
str |
the name of the Seldon Core deployment resource to fetch. |
required |
Returns:
| Type | Description |
|---|---|
SeldonDeployment |
The Seldon Core deployment resource. |
Exceptions:
| Type | Description |
|---|---|
SeldonDeploymentNotFoundError |
if the deployment resource cannot be found or is not managed by ZenML. |
SeldonClientError |
if an unknown error occurs while fetching the deployment. |
Source code in zenml/integrations/seldon/seldon_client.py
def get_deployment(self, name: str) -> SeldonDeployment:
"""Get a ZenML managed Seldon Core deployment resource by name.
Args:
name: the name of the Seldon Core deployment resource to fetch.
Returns:
The Seldon Core deployment resource.
Raises:
SeldonDeploymentNotFoundError: if the deployment resource cannot
be found or is not managed by ZenML.
SeldonClientError: if an unknown error occurs while fetching
the deployment.
"""
try:
logger.debug(f"Retrieving SeldonDeployment resource: {name}")
response = self._custom_objects_api.get_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
name=name,
)
logger.debug("Seldon Core API response: %s", response)
try:
deployment = SeldonDeployment(**response)
except ValidationError as e:
logger.error(
"Invalid Seldon Core deployment resource: %s\n%s",
str(e),
str(response),
)
raise SeldonDeploymentNotFoundError(
f"SeldonDeployment resource {name} could not be parsed"
)
# Only Seldon deployments managed by ZenML are returned
if not deployment.is_managed_by_zenml():
raise SeldonDeploymentNotFoundError(
f"Seldon Deployment {name} is not managed by ZenML"
)
return deployment
except k8s_client.rest.ApiException as e:
if e.status == 404:
raise SeldonDeploymentNotFoundError(
f"SeldonDeployment resource not found: {name}"
) from e
logger.error(
"Exception when fetching SeldonDeployment resource %s: %s",
name,
str(e),
)
raise SeldonClientError(
f"Unexpected exception when fetching SeldonDeployment "
f"resource: {name}"
) from e
sanitize_labels(labels)
staticmethod
Update the label values to be valid Kubernetes labels.
See: https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/#syntax-and-character-set
Source code in zenml/integrations/seldon/seldon_client.py
@staticmethod
def sanitize_labels(labels: Dict[str, str]) -> None:
"""Update the label values to be valid Kubernetes labels.
See: https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/#syntax-and-character-set
"""
for key, value in labels.items():
# Kubernetes labels must be alphanumeric, no longer than
# 63 characters, and must begin and end with an alphanumeric
# character ([a-z0-9A-Z])
labels[key] = re.sub(r"[^0-9a-zA-Z-_\.]+", "_", value)[:63].strip(
"-_."
)
update_deployment(self, deployment, poll_timeout=0)
Update a Seldon Core deployment resource.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
deployment |
SeldonDeployment |
the Seldon Core deployment resource to update |
required |
poll_timeout |
int |
the maximum time to wait for the deployment to become available or to fail. If set to 0, the function will return immediately without checking the deployment status. If a timeout occurs and the deployment is still pending creation, it will be returned anyway and no exception will be raised. |
0 |
Returns:
| Type | Description |
|---|---|
SeldonDeployment |
the updated Seldon Core deployment resource with updated status. |
Exceptions:
| Type | Description |
|---|---|
SeldonDeploymentNotFoundError |
if deployment resource with the given name cannot be found. |
SeldonClientError |
if an unknown error occurs while updating the deployment. |
Source code in zenml/integrations/seldon/seldon_client.py
def update_deployment(
self,
deployment: SeldonDeployment,
poll_timeout: int = 0,
) -> SeldonDeployment:
"""Update a Seldon Core deployment resource.
Args:
deployment: the Seldon Core deployment resource to update
poll_timeout: the maximum time to wait for the deployment to become
available or to fail. If set to 0, the function will return
immediately without checking the deployment status. If a timeout
occurs and the deployment is still pending creation, it will
be returned anyway and no exception will be raised.
Returns:
the updated Seldon Core deployment resource with updated status.
Raises:
SeldonDeploymentNotFoundError: if deployment resource with the given
name cannot be found.
SeldonClientError: if an unknown error occurs while updating the
deployment.
"""
try:
logger.debug(
f"Updating SeldonDeployment resource: {deployment.name}"
)
# mark the deployment as managed by ZenML, to differentiate
# between deployments that are created by ZenML and those that
# are not
deployment.mark_as_managed_by_zenml()
# call `get_deployment` to check that the deployment exists
# and is managed by ZenML. It will raise
# a SeldonDeploymentNotFoundError otherwise
self.get_deployment(name=deployment.name)
response = self._custom_objects_api.patch_namespaced_custom_object(
group="machinelearning.seldon.io",
version="v1",
namespace=self._namespace,
plural="seldondeployments",
name=deployment.name,
body=deployment.dict(exclude_none=True),
_request_timeout=poll_timeout or None,
)
logger.debug("Seldon Core API response: %s", response)
except k8s_client.rest.ApiException as e:
logger.error(
"Exception when updating SeldonDeployment resource: %s", str(e)
)
raise SeldonClientError(
"Exception when creating SeldonDeployment resource"
) from e
updated_deployment = self.get_deployment(name=deployment.name)
while poll_timeout > 0 and updated_deployment.is_pending():
time.sleep(5)
poll_timeout -= 5
updated_deployment = self.get_deployment(name=deployment.name)
return updated_deployment
SeldonClientError (Exception)
Base exception class for all exceptions raised by the SeldonClient.
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonClientError(Exception):
"""Base exception class for all exceptions raised by the SeldonClient."""
SeldonClientTimeout (SeldonClientError)
Raised when the Seldon client timed out while waiting for a resource to reach the expected status.
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonClientTimeout(SeldonClientError):
"""Raised when the Seldon client timed out while waiting for a resource
to reach the expected status."""
SeldonDeployment (BaseModel)
pydantic-model
A Seldon Core deployment CRD.
This is a Pydantic representation of some of the fields in the Seldon Core CRD (documented here: https://docs.seldon.io/projects/seldon-core/en/latest/reference/seldon-deployment.html).
Note that not all fields are represented, only those that are relevant to the ZenML integration. The fields that are not represented are silently ignored when the Seldon Deployment is created or updated from an external SeldonDeployment CRD representation.
Attributes:
| Name | Type | Description |
|---|---|---|
kind |
str |
Kubernetes kind field. |
apiVersion |
str |
Kubernetes apiVersion field. |
metadata |
SeldonDeploymentMetadata |
Kubernetes metadata field. |
spec |
SeldonDeploymentSpec |
Seldon Deployment spec entry. |
status |
Optional[zenml.integrations.seldon.seldon_client.SeldonDeploymentStatus] |
Seldon Deployment status. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeployment(BaseModel):
"""A Seldon Core deployment CRD.
This is a Pydantic representation of some of the fields in the Seldon Core
CRD (documented here:
https://docs.seldon.io/projects/seldon-core/en/latest/reference/seldon-deployment.html).
Note that not all fields are represented, only those that are relevant to
the ZenML integration. The fields that are not represented are silently
ignored when the Seldon Deployment is created or updated from an external
SeldonDeployment CRD representation.
Attributes:
kind: Kubernetes kind field.
apiVersion: Kubernetes apiVersion field.
metadata: Kubernetes metadata field.
spec: Seldon Deployment spec entry.
status: Seldon Deployment status.
"""
kind: str = Field(SELDON_DEPLOYMENT_KIND, const=True)
apiVersion: str = Field(SELDON_DEPLOYMENT_API_VERSION, const=True)
metadata: SeldonDeploymentMetadata = Field(
default_factory=SeldonDeploymentMetadata
)
spec: SeldonDeploymentSpec = Field(default_factory=SeldonDeploymentSpec)
status: Optional[SeldonDeploymentStatus]
def __str__(self) -> str:
return json.dumps(self.dict(exclude_none=True), indent=4)
@classmethod
def build(
cls,
name: Optional[str] = None,
model_uri: Optional[str] = None,
model_name: Optional[str] = None,
implementation: Optional[str] = None,
secret_name: Optional[str] = None,
labels: Optional[Dict[str, str]] = None,
annotations: Optional[Dict[str, str]] = None,
) -> "SeldonDeployment":
"""Build a basic Seldon Deployment object.
Args:
name: The name of the Seldon Deployment. If not explicitly passed,
a unique name is autogenerated.
model_uri: The URI of the model.
model_name: The name of the model.
implementation: The implementation of the model.
secret_name: The name of the Kubernetes secret containing
environment variable values (e.g. with credentials for the
artifact store) to use with the deployment service.
labels: A dictionary of labels to apply to the Seldon Deployment.
annotations: A dictionary of annotations to apply to the Seldon
Deployment.
Returns:
A minimal SeldonDeployment object built from the provided
parameters.
"""
if not name:
name = f"zenml-{time.time()}"
if labels is None:
labels = {}
if annotations is None:
annotations = {}
return SeldonDeployment(
metadata=SeldonDeploymentMetadata(
name=name, labels=labels, annotations=annotations
),
spec=SeldonDeploymentSpec(
name=name,
predictors=[
SeldonDeploymentPredictor(
name=model_name or "",
graph=SeldonDeploymentPredictiveUnit(
name="default",
type=SeldonDeploymentPredictiveUnitType.MODEL,
modelUri=model_uri or "",
implementation=implementation or "",
envSecretRefName=secret_name,
),
)
],
),
)
def is_managed_by_zenml(self) -> bool:
"""Checks if this Seldon Deployment is managed by ZenML.
The convention used to differentiate between SeldonDeployment instances
that are managed by ZenML and those that are not is to set the `app`
label value to `zenml`.
Returns:
True if the Seldon Deployment is managed by ZenML, False
otherwise.
"""
return self.metadata.labels.get("app") == "zenml"
def mark_as_managed_by_zenml(self) -> None:
"""Marks this Seldon Deployment as managed by ZenML.
The convention used to differentiate between SeldonDeployment instances
that are managed by ZenML and those that are not is to set the `app`
label value to `zenml`.
"""
self.metadata.labels["app"] = "zenml"
@property
def name(self) -> str:
"""Returns the name of this Seldon Deployment.
This is just a shortcut for `self.metadata.name`.
Returns:
The name of this Seldon Deployment.
"""
return self.metadata.name
@property
def state(self) -> SeldonDeploymentStatusState:
"""The state of the Seldon Deployment.
Returns:
The state of the Seldon Deployment.
"""
if not self.status:
return SeldonDeploymentStatusState.UNKNOWN
return self.status.state
def is_pending(self) -> bool:
"""Checks if the Seldon Deployment is in a pending state.
Returns:
True if the Seldon Deployment is pending, False otherwise.
"""
return self.state == SeldonDeploymentStatusState.CREATING
def is_available(self) -> bool:
"""Checks if the Seldon Deployment is in an available state.
Returns:
True if the Seldon Deployment is available, False otherwise.
"""
return self.state == SeldonDeploymentStatusState.AVAILABLE
def is_failed(self) -> bool:
"""Checks if the Seldon Deployment is in a failed state.
Returns:
True if the Seldon Deployment is failed, False otherwise.
"""
return self.state == SeldonDeploymentStatusState.FAILED
def get_error(self) -> Optional[str]:
"""Get a message describing the error, if in an error state.
Returns:
A message describing the error, if in an error state, otherwise
None.
"""
if self.status and self.is_failed():
return self.status.description
return None
def get_pending_message(self) -> Optional[str]:
"""Get a message describing the pending conditions of the Seldon
Deployment.
Returns:
A message describing the pending condition of the Seldon
Deployment, or None, if no conditions are pending.
"""
if not self.status or not self.status.conditions:
return None
ready_condition_message = [
c.message
for c in self.status.conditions
if c.type == "Ready" and not c.status
]
if not ready_condition_message:
return None
return ready_condition_message[0]
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
name: str
property
readonly
Returns the name of this Seldon Deployment.
This is just a shortcut for self.metadata.name.
Returns:
| Type | Description |
|---|---|
str |
The name of this Seldon Deployment. |
state: SeldonDeploymentStatusState
property
readonly
The state of the Seldon Deployment.
Returns:
| Type | Description |
|---|---|
SeldonDeploymentStatusState |
The state of the Seldon Deployment. |
Config
Pydantic configuration class.
Source code in zenml/integrations/seldon/seldon_client.py
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
__str__(self)
special
Return str(self).
Source code in zenml/integrations/seldon/seldon_client.py
def __str__(self) -> str:
return json.dumps(self.dict(exclude_none=True), indent=4)
build(name=None, model_uri=None, model_name=None, implementation=None, secret_name=None, labels=None, annotations=None)
classmethod
Build a basic Seldon Deployment object.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name |
Optional[str] |
The name of the Seldon Deployment. If not explicitly passed, a unique name is autogenerated. |
None |
model_uri |
Optional[str] |
The URI of the model. |
None |
model_name |
Optional[str] |
The name of the model. |
None |
implementation |
Optional[str] |
The implementation of the model. |
None |
secret_name |
Optional[str] |
The name of the Kubernetes secret containing environment variable values (e.g. with credentials for the artifact store) to use with the deployment service. |
None |
labels |
Optional[Dict[str, str]] |
A dictionary of labels to apply to the Seldon Deployment. |
None |
annotations |
Optional[Dict[str, str]] |
A dictionary of annotations to apply to the Seldon Deployment. |
None |
Returns:
| Type | Description |
|---|---|
SeldonDeployment |
A minimal SeldonDeployment object built from the provided parameters. |
Source code in zenml/integrations/seldon/seldon_client.py
@classmethod
def build(
cls,
name: Optional[str] = None,
model_uri: Optional[str] = None,
model_name: Optional[str] = None,
implementation: Optional[str] = None,
secret_name: Optional[str] = None,
labels: Optional[Dict[str, str]] = None,
annotations: Optional[Dict[str, str]] = None,
) -> "SeldonDeployment":
"""Build a basic Seldon Deployment object.
Args:
name: The name of the Seldon Deployment. If not explicitly passed,
a unique name is autogenerated.
model_uri: The URI of the model.
model_name: The name of the model.
implementation: The implementation of the model.
secret_name: The name of the Kubernetes secret containing
environment variable values (e.g. with credentials for the
artifact store) to use with the deployment service.
labels: A dictionary of labels to apply to the Seldon Deployment.
annotations: A dictionary of annotations to apply to the Seldon
Deployment.
Returns:
A minimal SeldonDeployment object built from the provided
parameters.
"""
if not name:
name = f"zenml-{time.time()}"
if labels is None:
labels = {}
if annotations is None:
annotations = {}
return SeldonDeployment(
metadata=SeldonDeploymentMetadata(
name=name, labels=labels, annotations=annotations
),
spec=SeldonDeploymentSpec(
name=name,
predictors=[
SeldonDeploymentPredictor(
name=model_name or "",
graph=SeldonDeploymentPredictiveUnit(
name="default",
type=SeldonDeploymentPredictiveUnitType.MODEL,
modelUri=model_uri or "",
implementation=implementation or "",
envSecretRefName=secret_name,
),
)
],
),
)
get_error(self)
Get a message describing the error, if in an error state.
Returns:
| Type | Description |
|---|---|
Optional[str] |
A message describing the error, if in an error state, otherwise None. |
Source code in zenml/integrations/seldon/seldon_client.py
def get_error(self) -> Optional[str]:
"""Get a message describing the error, if in an error state.
Returns:
A message describing the error, if in an error state, otherwise
None.
"""
if self.status and self.is_failed():
return self.status.description
return None
get_pending_message(self)
Get a message describing the pending conditions of the Seldon Deployment.
Returns:
| Type | Description |
|---|---|
Optional[str] |
A message describing the pending condition of the Seldon Deployment, or None, if no conditions are pending. |
Source code in zenml/integrations/seldon/seldon_client.py
def get_pending_message(self) -> Optional[str]:
"""Get a message describing the pending conditions of the Seldon
Deployment.
Returns:
A message describing the pending condition of the Seldon
Deployment, or None, if no conditions are pending.
"""
if not self.status or not self.status.conditions:
return None
ready_condition_message = [
c.message
for c in self.status.conditions
if c.type == "Ready" and not c.status
]
if not ready_condition_message:
return None
return ready_condition_message[0]
is_available(self)
Checks if the Seldon Deployment is in an available state.
Returns:
| Type | Description |
|---|---|
bool |
True if the Seldon Deployment is available, False otherwise. |
Source code in zenml/integrations/seldon/seldon_client.py
def is_available(self) -> bool:
"""Checks if the Seldon Deployment is in an available state.
Returns:
True if the Seldon Deployment is available, False otherwise.
"""
return self.state == SeldonDeploymentStatusState.AVAILABLE
is_failed(self)
Checks if the Seldon Deployment is in a failed state.
Returns:
| Type | Description |
|---|---|
bool |
True if the Seldon Deployment is failed, False otherwise. |
Source code in zenml/integrations/seldon/seldon_client.py
def is_failed(self) -> bool:
"""Checks if the Seldon Deployment is in a failed state.
Returns:
True if the Seldon Deployment is failed, False otherwise.
"""
return self.state == SeldonDeploymentStatusState.FAILED
is_managed_by_zenml(self)
Checks if this Seldon Deployment is managed by ZenML.
The convention used to differentiate between SeldonDeployment instances
that are managed by ZenML and those that are not is to set the app
label value to zenml.
Returns:
| Type | Description |
|---|---|
bool |
True if the Seldon Deployment is managed by ZenML, False otherwise. |
Source code in zenml/integrations/seldon/seldon_client.py
def is_managed_by_zenml(self) -> bool:
"""Checks if this Seldon Deployment is managed by ZenML.
The convention used to differentiate between SeldonDeployment instances
that are managed by ZenML and those that are not is to set the `app`
label value to `zenml`.
Returns:
True if the Seldon Deployment is managed by ZenML, False
otherwise.
"""
return self.metadata.labels.get("app") == "zenml"
is_pending(self)
Checks if the Seldon Deployment is in a pending state.
Returns:
| Type | Description |
|---|---|
bool |
True if the Seldon Deployment is pending, False otherwise. |
Source code in zenml/integrations/seldon/seldon_client.py
def is_pending(self) -> bool:
"""Checks if the Seldon Deployment is in a pending state.
Returns:
True if the Seldon Deployment is pending, False otherwise.
"""
return self.state == SeldonDeploymentStatusState.CREATING
mark_as_managed_by_zenml(self)
Marks this Seldon Deployment as managed by ZenML.
The convention used to differentiate between SeldonDeployment instances
that are managed by ZenML and those that are not is to set the app
label value to zenml.
Source code in zenml/integrations/seldon/seldon_client.py
def mark_as_managed_by_zenml(self) -> None:
"""Marks this Seldon Deployment as managed by ZenML.
The convention used to differentiate between SeldonDeployment instances
that are managed by ZenML and those that are not is to set the `app`
label value to `zenml`.
"""
self.metadata.labels["app"] = "zenml"
SeldonDeploymentExistsError (SeldonClientError)
Raised when a SeldonDeployment resource cannot be created because a resource with the same name already exists.
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentExistsError(SeldonClientError):
"""Raised when a SeldonDeployment resource cannot be created because a
resource with the same name already exists."""
SeldonDeploymentMetadata (BaseModel)
pydantic-model
Metadata for a Seldon Deployment.
Attributes:
| Name | Type | Description |
|---|---|---|
name |
str |
the name of the Seldon Deployment. |
labels |
Dict[str, str] |
Kubernetes labels for the Seldon Deployment. |
annotations |
Dict[str, str] |
Kubernetes annotations for the Seldon Deployment. |
creationTimestamp |
Optional[str] |
the creation timestamp of the Seldon Deployment. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentMetadata(BaseModel):
"""Metadata for a Seldon Deployment.
Attributes:
name: the name of the Seldon Deployment.
labels: Kubernetes labels for the Seldon Deployment.
annotations: Kubernetes annotations for the Seldon Deployment.
creationTimestamp: the creation timestamp of the Seldon Deployment.
"""
name: str
labels: Dict[str, str] = Field(default_factory=dict)
annotations: Dict[str, str] = Field(default_factory=dict)
creationTimestamp: Optional[str]
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
Config
Pydantic configuration class.
Source code in zenml/integrations/seldon/seldon_client.py
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
SeldonDeploymentNotFoundError (SeldonClientError)
Raised when a particular SeldonDeployment resource is not found or is not managed by ZenML.
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentNotFoundError(SeldonClientError):
"""Raised when a particular SeldonDeployment resource is not found or is
not managed by ZenML."""
SeldonDeploymentPredictiveUnit (BaseModel)
pydantic-model
Seldon Deployment predictive unit.
Attributes:
| Name | Type | Description |
|---|---|---|
name |
str |
the name of the predictive unit. |
type |
Optional[zenml.integrations.seldon.seldon_client.SeldonDeploymentPredictiveUnitType] |
predictive unit type. |
implementation |
Optional[str] |
the Seldon Core implementation used to serve the model. |
modelUri |
Optional[str] |
URI of the model (or models) to serve. |
serviceAccountName |
Optional[str] |
the name of the service account to associate with the predictive unit container. |
envSecretRefName |
Optional[str] |
the name of a Kubernetes secret that contains environment variables (e.g. credentials) to be configured for the predictive unit container. |
children |
List[zenml.integrations.seldon.seldon_client.SeldonDeploymentPredictiveUnit] |
a list of child predictive units that together make up the model serving graph. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentPredictiveUnit(BaseModel):
"""Seldon Deployment predictive unit.
Attributes:
name: the name of the predictive unit.
type: predictive unit type.
implementation: the Seldon Core implementation used to serve the model.
modelUri: URI of the model (or models) to serve.
serviceAccountName: the name of the service account to associate with
the predictive unit container.
envSecretRefName: the name of a Kubernetes secret that contains
environment variables (e.g. credentials) to be configured for the
predictive unit container.
children: a list of child predictive units that together make up the
model serving graph.
"""
name: str
type: Optional[
SeldonDeploymentPredictiveUnitType
] = SeldonDeploymentPredictiveUnitType.MODEL
implementation: Optional[str]
modelUri: Optional[str]
serviceAccountName: Optional[str]
envSecretRefName: Optional[str]
children: List["SeldonDeploymentPredictiveUnit"] = Field(
default_factory=list
)
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
Config
Pydantic configuration class.
Source code in zenml/integrations/seldon/seldon_client.py
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
SeldonDeploymentPredictiveUnitType (StrEnum)
Predictive unit types for a Seldon Deployment.
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentPredictiveUnitType(StrEnum):
"""Predictive unit types for a Seldon Deployment."""
UNKNOWN_TYPE = "UNKNOWN_TYPE"
ROUTER = "ROUTER"
COMBINER = "COMBINER"
MODEL = "MODEL"
TRANSFORMER = "TRANSFORMER"
OUTPUT_TRANSFORMER = "OUTPUT_TRANSFORMER"
SeldonDeploymentPredictor (BaseModel)
pydantic-model
Seldon Deployment predictor.
Attributes:
| Name | Type | Description |
|---|---|---|
name |
str |
the name of the predictor. |
replicas |
int |
the number of pod replicas for the predictor. |
graph |
SeldonDeploymentPredictiveUnit |
the serving graph composed of one or more predictive units. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentPredictor(BaseModel):
"""Seldon Deployment predictor.
Attributes:
name: the name of the predictor.
replicas: the number of pod replicas for the predictor.
graph: the serving graph composed of one or more predictive units.
"""
name: str
replicas: int = 1
graph: SeldonDeploymentPredictiveUnit = Field(
default_factory=SeldonDeploymentPredictiveUnit
)
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
Config
Pydantic configuration class.
Source code in zenml/integrations/seldon/seldon_client.py
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
SeldonDeploymentSpec (BaseModel)
pydantic-model
Spec for a Seldon Deployment.
Attributes:
| Name | Type | Description |
|---|---|---|
name |
str |
the name of the Seldon Deployment. |
protocol |
Optional[str] |
the API protocol used for the Seldon Deployment. |
predictors |
List[zenml.integrations.seldon.seldon_client.SeldonDeploymentPredictor] |
a list of predictors that make up the serving graph. |
replicas |
int |
the default number of pod replicas used for the predictors. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentSpec(BaseModel):
"""Spec for a Seldon Deployment.
Attributes:
name: the name of the Seldon Deployment.
protocol: the API protocol used for the Seldon Deployment.
predictors: a list of predictors that make up the serving graph.
replicas: the default number of pod replicas used for the predictors.
"""
name: str
protocol: Optional[str]
predictors: List[SeldonDeploymentPredictor]
replicas: int = 1
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
Config
Pydantic configuration class.
Source code in zenml/integrations/seldon/seldon_client.py
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
SeldonDeploymentStatus (BaseModel)
pydantic-model
The status of a Seldon Deployment.
Attributes:
| Name | Type | Description |
|---|---|---|
state |
SeldonDeploymentStatusState |
the current state of the Seldon Deployment. |
description |
Optional[str] |
a human-readable description of the current state. |
replicas |
Optional[int] |
the current number of running pod replicas |
address |
Optional[zenml.integrations.seldon.seldon_client.SeldonDeploymentStatusAddress] |
the address where the Seldon Deployment API can be accessed. |
conditions |
List[zenml.integrations.seldon.seldon_client.SeldonDeploymentStatusCondition] |
the list of Kubernetes conditions for the Seldon Deployment. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentStatus(BaseModel):
"""The status of a Seldon Deployment.
Attributes:
state: the current state of the Seldon Deployment.
description: a human-readable description of the current state.
replicas: the current number of running pod replicas
address: the address where the Seldon Deployment API can be accessed.
conditions: the list of Kubernetes conditions for the Seldon Deployment.
"""
state: SeldonDeploymentStatusState = SeldonDeploymentStatusState.UNKNOWN
description: Optional[str]
replicas: Optional[int]
address: Optional[SeldonDeploymentStatusAddress]
conditions: List[SeldonDeploymentStatusCondition]
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
Config
Pydantic configuration class.
Source code in zenml/integrations/seldon/seldon_client.py
class Config:
"""Pydantic configuration class."""
# validate attribute assignments
validate_assignment = True
# Ignore extra attributes from the CRD that are not reflected here
extra = "ignore"
SeldonDeploymentStatusAddress (BaseModel)
pydantic-model
The status address for a Seldon Deployment.
Attributes:
| Name | Type | Description |
|---|---|---|
url |
str |
the URL where the Seldon Deployment API can be accessed internally. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentStatusAddress(BaseModel):
"""The status address for a Seldon Deployment.
Attributes:
url: the URL where the Seldon Deployment API can be accessed internally.
"""
url: str
SeldonDeploymentStatusCondition (BaseModel)
pydantic-model
The Kubernetes status condition entry for a Seldon Deployment.
Attributes:
| Name | Type | Description |
|---|---|---|
type |
str |
Type of runtime condition. |
status |
bool |
Status of the condition. |
reason |
Optional[str] |
Brief CamelCase string containing reason for the condition's last transition. |
message |
Optional[str] |
Human-readable message indicating details about last transition. |
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentStatusCondition(BaseModel):
"""The Kubernetes status condition entry for a Seldon Deployment.
Attributes:
type: Type of runtime condition.
status: Status of the condition.
reason: Brief CamelCase string containing reason for the condition's
last transition.
message: Human-readable message indicating details about last
transition.
"""
type: str
status: bool
reason: Optional[str]
message: Optional[str]
SeldonDeploymentStatusState (StrEnum)
Possible state values for a Seldon Deployment.
Source code in zenml/integrations/seldon/seldon_client.py
class SeldonDeploymentStatusState(StrEnum):
"""Possible state values for a Seldon Deployment."""
UNKNOWN = "Unknown"
AVAILABLE = "Available"
CREATING = "Creating"
FAILED = "Failed"
services
special
seldon_deployment
SeldonDeploymentConfig (ServiceConfig)
pydantic-model
Seldon Core deployment service configuration.
Attributes:
| Name | Type | Description |
|---|---|---|
model_uri |
str |
URI of the model (or models) to serve. |
model_name |
str |
the name of the model. Multiple versions of the same model should use the same model name. |
implementation |
str |
the Seldon Core implementation used to serve the model. |
replicas |
int |
number of replicas to use for the prediction service. |
secret_name |
Optional[str] |
the name of a Kubernetes secret containing additional configuration parameters for the Seldon Core deployment (e.g. credentials to access the Artifact Store). |
model_metadata |
Dict[str, Any] |
optional model metadata information (see https://docs.seldon.io/projects/seldon-core/en/latest/reference/apis/metadata.html). |
extra_args |
Dict[str, Any] |
additional arguments to pass to the Seldon Core deployment resource configuration. |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
class SeldonDeploymentConfig(ServiceConfig):
"""Seldon Core deployment service configuration.
Attributes:
model_uri: URI of the model (or models) to serve.
model_name: the name of the model. Multiple versions of the same model
should use the same model name.
implementation: the Seldon Core implementation used to serve the model.
replicas: number of replicas to use for the prediction service.
secret_name: the name of a Kubernetes secret containing additional
configuration parameters for the Seldon Core deployment (e.g.
credentials to access the Artifact Store).
model_metadata: optional model metadata information (see
https://docs.seldon.io/projects/seldon-core/en/latest/reference/apis/metadata.html).
extra_args: additional arguments to pass to the Seldon Core deployment
resource configuration.
"""
model_uri: str = ""
model_name: str = "default"
# TODO [ENG-775]: have an enum of all supported Seldon Core implementations
implementation: str
replicas: int = 1
secret_name: Optional[str]
model_metadata: Dict[str, Any] = Field(default_factory=dict)
extra_args: Dict[str, Any] = Field(default_factory=dict)
def get_seldon_deployment_labels(self) -> Dict[str, str]:
"""Generate the labels for the Seldon Core deployment from the
service configuration.
These labels are attached to the Seldon Core deployment resource
and may be used as label selectors in lookup operations.
Returns:
The labels for the Seldon Core deployment.
"""
labels = {}
if self.pipeline_name:
labels["zenml.pipeline_name"] = self.pipeline_name
if self.pipeline_run_id:
labels["zenml.pipeline_run_id"] = self.pipeline_run_id
if self.pipeline_step_name:
labels["zenml.pipeline_step_name"] = self.pipeline_step_name
if self.model_name:
labels["zenml.model_name"] = self.model_name
if self.model_uri:
labels["zenml.model_uri"] = self.model_uri
if self.implementation:
labels["zenml.model_type"] = self.implementation
SeldonClient.sanitize_labels(labels)
return labels
def get_seldon_deployment_annotations(self) -> Dict[str, str]:
"""Generate the annotations for the Seldon Core deployment from the
service configuration.
The annotations are used to store additional information about the
Seldon Core service that is associated with the deployment that is
not available in the labels. One annotation particularly important
is the serialized Service configuration itself, which is used to
recreate the service configuration from a remote Seldon deployment.
Returns:
The annotations for the Seldon Core deployment.
"""
annotations = {
"zenml.service_config": self.json(),
"zenml.version": __version__,
}
return annotations
@classmethod
def create_from_deployment(
cls, deployment: SeldonDeployment
) -> "SeldonDeploymentConfig":
"""Recreate the configuration of a Seldon Core Service from a deployed instance.
Args:
deployment: the Seldon Core deployment resource.
Returns:
The Seldon Core service configuration corresponding to the given
Seldon Core deployment resource.
Raises:
ValueError: if the given deployment resource does not contain
the expected annotations or it contains an invalid or
incompatible Seldon Core service configuration.
"""
config_data = deployment.metadata.annotations.get(
"zenml.service_config"
)
if not config_data:
raise ValueError(
f"The given deployment resource does not contain a "
f"'zenml.service_config' annotation: {deployment}"
)
try:
service_config = cls.parse_raw(config_data)
except ValidationError as e:
raise ValueError(
f"The loaded Seldon Core deployment resource contains an "
f"invalid or incompatible Seldon Core service configuration: "
f"{config_data}"
) from e
return service_config
create_from_deployment(deployment)
classmethod
Recreate the configuration of a Seldon Core Service from a deployed instance.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
deployment |
SeldonDeployment |
the Seldon Core deployment resource. |
required |
Returns:
| Type | Description |
|---|---|
SeldonDeploymentConfig |
The Seldon Core service configuration corresponding to the given Seldon Core deployment resource. |
Exceptions:
| Type | Description |
|---|---|
ValueError |
if the given deployment resource does not contain the expected annotations or it contains an invalid or incompatible Seldon Core service configuration. |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
@classmethod
def create_from_deployment(
cls, deployment: SeldonDeployment
) -> "SeldonDeploymentConfig":
"""Recreate the configuration of a Seldon Core Service from a deployed instance.
Args:
deployment: the Seldon Core deployment resource.
Returns:
The Seldon Core service configuration corresponding to the given
Seldon Core deployment resource.
Raises:
ValueError: if the given deployment resource does not contain
the expected annotations or it contains an invalid or
incompatible Seldon Core service configuration.
"""
config_data = deployment.metadata.annotations.get(
"zenml.service_config"
)
if not config_data:
raise ValueError(
f"The given deployment resource does not contain a "
f"'zenml.service_config' annotation: {deployment}"
)
try:
service_config = cls.parse_raw(config_data)
except ValidationError as e:
raise ValueError(
f"The loaded Seldon Core deployment resource contains an "
f"invalid or incompatible Seldon Core service configuration: "
f"{config_data}"
) from e
return service_config
get_seldon_deployment_annotations(self)
Generate the annotations for the Seldon Core deployment from the service configuration.
The annotations are used to store additional information about the Seldon Core service that is associated with the deployment that is not available in the labels. One annotation particularly important is the serialized Service configuration itself, which is used to recreate the service configuration from a remote Seldon deployment.
Returns:
| Type | Description |
|---|---|
Dict[str, str] |
The annotations for the Seldon Core deployment. |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
def get_seldon_deployment_annotations(self) -> Dict[str, str]:
"""Generate the annotations for the Seldon Core deployment from the
service configuration.
The annotations are used to store additional information about the
Seldon Core service that is associated with the deployment that is
not available in the labels. One annotation particularly important
is the serialized Service configuration itself, which is used to
recreate the service configuration from a remote Seldon deployment.
Returns:
The annotations for the Seldon Core deployment.
"""
annotations = {
"zenml.service_config": self.json(),
"zenml.version": __version__,
}
return annotations
get_seldon_deployment_labels(self)
Generate the labels for the Seldon Core deployment from the service configuration.
These labels are attached to the Seldon Core deployment resource and may be used as label selectors in lookup operations.
Returns:
| Type | Description |
|---|---|
Dict[str, str] |
The labels for the Seldon Core deployment. |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
def get_seldon_deployment_labels(self) -> Dict[str, str]:
"""Generate the labels for the Seldon Core deployment from the
service configuration.
These labels are attached to the Seldon Core deployment resource
and may be used as label selectors in lookup operations.
Returns:
The labels for the Seldon Core deployment.
"""
labels = {}
if self.pipeline_name:
labels["zenml.pipeline_name"] = self.pipeline_name
if self.pipeline_run_id:
labels["zenml.pipeline_run_id"] = self.pipeline_run_id
if self.pipeline_step_name:
labels["zenml.pipeline_step_name"] = self.pipeline_step_name
if self.model_name:
labels["zenml.model_name"] = self.model_name
if self.model_uri:
labels["zenml.model_uri"] = self.model_uri
if self.implementation:
labels["zenml.model_type"] = self.implementation
SeldonClient.sanitize_labels(labels)
return labels
SeldonDeploymentService (BaseService)
pydantic-model
A service that represents a Seldon Core deployment server.
Attributes:
| Name | Type | Description |
|---|---|---|
config |
SeldonDeploymentConfig |
service configuration. |
status |
SeldonDeploymentServiceStatus |
service status. |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
class SeldonDeploymentService(BaseService):
"""A service that represents a Seldon Core deployment server.
Attributes:
config: service configuration.
status: service status.
"""
SERVICE_TYPE = ServiceType(
name="seldon-deployment",
type="model-serving",
flavor="seldon",
description="Seldon Core prediction service",
)
config: SeldonDeploymentConfig = Field(
default_factory=SeldonDeploymentConfig
)
status: SeldonDeploymentServiceStatus = Field(
default_factory=SeldonDeploymentServiceStatus
)
def _get_client(self) -> SeldonClient:
"""Get the Seldon Core client from the active Seldon Core model deployer.
Returns:
The Seldon Core client.
"""
from zenml.integrations.seldon.model_deployers.seldon_model_deployer import (
SeldonModelDeployer,
)
model_deployer = SeldonModelDeployer.get_active_model_deployer()
return model_deployer.seldon_client
def check_status(self) -> Tuple[ServiceState, str]:
"""Check the the current operational state of the Seldon Core
deployment.
Returns:
The operational state of the Seldon Core deployment and a message
providing additional information about that state (e.g. a
description of the error, if one is encountered).
"""
client = self._get_client()
name = self.seldon_deployment_name
try:
deployment = client.get_deployment(name=name)
except SeldonDeploymentNotFoundError:
return (ServiceState.INACTIVE, "")
if deployment.is_available():
return (
ServiceState.ACTIVE,
f"Seldon Core deployment '{name}' is available",
)
if deployment.is_failed():
return (
ServiceState.ERROR,
f"Seldon Core deployment '{name}' failed: "
f"{deployment.get_error()}",
)
pending_message = deployment.get_pending_message() or ""
return (
ServiceState.PENDING_STARTUP,
"Seldon Core deployment is being created: " + pending_message,
)
@property
def seldon_deployment_name(self) -> str:
"""Get the name of the Seldon Core deployment that uniquely
corresponds to this service instance
Returns:
The name of the Seldon Core deployment.
"""
return f"zenml-{str(self.uuid)}"
def _get_seldon_deployment_labels(self) -> Dict[str, str]:
"""Generate the labels for the Seldon Core deployment from the
service configuration.
Returns:
The labels for the Seldon Core deployment.
"""
labels = self.config.get_seldon_deployment_labels()
labels["zenml.service_uuid"] = str(self.uuid)
SeldonClient.sanitize_labels(labels)
return labels
@classmethod
def create_from_deployment(
cls, deployment: SeldonDeployment
) -> "SeldonDeploymentService":
"""Recreate a Seldon Core service from a Seldon Core
deployment resource and update their operational status.
Args:
deployment: the Seldon Core deployment resource.
Returns:
The Seldon Core service corresponding to the given
Seldon Core deployment resource.
"""
config = SeldonDeploymentConfig.create_from_deployment(deployment)
uuid = deployment.metadata.labels.get("zenml.service_uuid")
if not uuid:
raise ValueError(
f"The given deployment resource does not contain a valid "
f"'zenml.service_uuid' label: {deployment}"
)
service = cls(uuid=UUID(uuid), config=config)
service.update_status()
return service
def provision(self) -> None:
"""Provision or update the remote Seldon Core deployment instance to
match the current configuration.
"""
client = self._get_client()
name = self.seldon_deployment_name
deployment = SeldonDeployment.build(
name=name,
model_uri=self.config.model_uri,
model_name=self.config.model_name,
implementation=self.config.implementation,
secret_name=self.config.secret_name,
labels=self._get_seldon_deployment_labels(),
annotations=self.config.get_seldon_deployment_annotations(),
)
deployment.spec.replicas = self.config.replicas
deployment.spec.predictors[0].replicas = self.config.replicas
# check if the Seldon deployment already exists
try:
client.get_deployment(name=name)
# update the existing deployment
client.update_deployment(deployment)
except SeldonDeploymentNotFoundError:
# create the deployment
client.create_deployment(deployment=deployment)
def deprovision(self, force: bool = False) -> None:
"""Deprovision the remote Seldon Core deployment instance.
Args:
force: if True, the remote deployment instance will be
forcefully deprovisioned.
"""
client = self._get_client()
name = self.seldon_deployment_name
try:
client.delete_deployment(name=name, force=force)
except SeldonDeploymentNotFoundError:
pass
@property
def prediction_url(self) -> Optional[str]:
"""The prediction URI exposed by the prediction service.
Returns:
The prediction URI exposed by the prediction service, or None if
the service is not yet ready.
"""
from zenml.integrations.seldon.model_deployers.seldon_model_deployer import (
SeldonModelDeployer,
)
if not self.is_running:
return None
namespace = self._get_client().namespace
model_deployer = SeldonModelDeployer.get_active_model_deployer()
return os.path.join(
model_deployer.base_url,
"seldon",
namespace,
self.seldon_deployment_name,
"api/v0.1/predictions",
)
def predict(self, request: "NDArray[Any]") -> "NDArray[Any]":
"""Make a prediction using the service.
Args:
request: a numpy array representing the request
Returns:
A numpy array representing the prediction returned by the service.
"""
if not self.is_running:
raise Exception(
"Seldon prediction service is not running. "
"Please start the service before making predictions."
)
if self.prediction_url is None:
raise ValueError("`self.prediction_url` is not set, cannot post.")
response = requests.post(
self.prediction_url,
json={"data": {"ndarray": request.tolist()}},
)
response.raise_for_status()
return np.array(response.json()["data"]["ndarray"])
prediction_url: Optional[str]
property
readonly
The prediction URI exposed by the prediction service.
Returns:
| Type | Description |
|---|---|
Optional[str] |
The prediction URI exposed by the prediction service, or None if the service is not yet ready. |
seldon_deployment_name: str
property
readonly
Get the name of the Seldon Core deployment that uniquely corresponds to this service instance
Returns:
| Type | Description |
|---|---|
str |
The name of the Seldon Core deployment. |
check_status(self)
Check the the current operational state of the Seldon Core deployment.
Returns:
| Type | Description |
|---|---|
Tuple[zenml.services.service_status.ServiceState, str] |
The operational state of the Seldon Core deployment and a message providing additional information about that state (e.g. a description of the error, if one is encountered). |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
def check_status(self) -> Tuple[ServiceState, str]:
"""Check the the current operational state of the Seldon Core
deployment.
Returns:
The operational state of the Seldon Core deployment and a message
providing additional information about that state (e.g. a
description of the error, if one is encountered).
"""
client = self._get_client()
name = self.seldon_deployment_name
try:
deployment = client.get_deployment(name=name)
except SeldonDeploymentNotFoundError:
return (ServiceState.INACTIVE, "")
if deployment.is_available():
return (
ServiceState.ACTIVE,
f"Seldon Core deployment '{name}' is available",
)
if deployment.is_failed():
return (
ServiceState.ERROR,
f"Seldon Core deployment '{name}' failed: "
f"{deployment.get_error()}",
)
pending_message = deployment.get_pending_message() or ""
return (
ServiceState.PENDING_STARTUP,
"Seldon Core deployment is being created: " + pending_message,
)
create_from_deployment(deployment)
classmethod
Recreate a Seldon Core service from a Seldon Core deployment resource and update their operational status.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
deployment |
SeldonDeployment |
the Seldon Core deployment resource. |
required |
Returns:
| Type | Description |
|---|---|
SeldonDeploymentService |
The Seldon Core service corresponding to the given Seldon Core deployment resource. |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
@classmethod
def create_from_deployment(
cls, deployment: SeldonDeployment
) -> "SeldonDeploymentService":
"""Recreate a Seldon Core service from a Seldon Core
deployment resource and update their operational status.
Args:
deployment: the Seldon Core deployment resource.
Returns:
The Seldon Core service corresponding to the given
Seldon Core deployment resource.
"""
config = SeldonDeploymentConfig.create_from_deployment(deployment)
uuid = deployment.metadata.labels.get("zenml.service_uuid")
if not uuid:
raise ValueError(
f"The given deployment resource does not contain a valid "
f"'zenml.service_uuid' label: {deployment}"
)
service = cls(uuid=UUID(uuid), config=config)
service.update_status()
return service
deprovision(self, force=False)
Deprovision the remote Seldon Core deployment instance.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
force |
bool |
if True, the remote deployment instance will be forcefully deprovisioned. |
False |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
def deprovision(self, force: bool = False) -> None:
"""Deprovision the remote Seldon Core deployment instance.
Args:
force: if True, the remote deployment instance will be
forcefully deprovisioned.
"""
client = self._get_client()
name = self.seldon_deployment_name
try:
client.delete_deployment(name=name, force=force)
except SeldonDeploymentNotFoundError:
pass
predict(self, request)
Make a prediction using the service.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
request |
NDArray[Any] |
a numpy array representing the request |
required |
Returns:
| Type | Description |
|---|---|
NDArray[Any] |
A numpy array representing the prediction returned by the service. |
Source code in zenml/integrations/seldon/services/seldon_deployment.py
def predict(self, request: "NDArray[Any]") -> "NDArray[Any]":
"""Make a prediction using the service.
Args:
request: a numpy array representing the request
Returns:
A numpy array representing the prediction returned by the service.
"""
if not self.is_running:
raise Exception(
"Seldon prediction service is not running. "
"Please start the service before making predictions."
)
if self.prediction_url is None:
raise ValueError("`self.prediction_url` is not set, cannot post.")
response = requests.post(
self.prediction_url,
json={"data": {"ndarray": request.tolist()}},
)
response.raise_for_status()
return np.array(response.json()["data"]["ndarray"])
provision(self)
Provision or update the remote Seldon Core deployment instance to match the current configuration.
Source code in zenml/integrations/seldon/services/seldon_deployment.py
def provision(self) -> None:
"""Provision or update the remote Seldon Core deployment instance to
match the current configuration.
"""
client = self._get_client()
name = self.seldon_deployment_name
deployment = SeldonDeployment.build(
name=name,
model_uri=self.config.model_uri,
model_name=self.config.model_name,
implementation=self.config.implementation,
secret_name=self.config.secret_name,
labels=self._get_seldon_deployment_labels(),
annotations=self.config.get_seldon_deployment_annotations(),
)
deployment.spec.replicas = self.config.replicas
deployment.spec.predictors[0].replicas = self.config.replicas
# check if the Seldon deployment already exists
try:
client.get_deployment(name=name)
# update the existing deployment
client.update_deployment(deployment)
except SeldonDeploymentNotFoundError:
# create the deployment
client.create_deployment(deployment=deployment)
SeldonDeploymentServiceStatus (ServiceStatus)
pydantic-model
Seldon Core deployment service status.
Source code in zenml/integrations/seldon/services/seldon_deployment.py
class SeldonDeploymentServiceStatus(ServiceStatus):
"""Seldon Core deployment service status."""
steps
special
seldon_deployer
SeldonDeployerStepConfig (BaseStepConfig)
pydantic-model
Seldon model deployer step configuration
Attributes:
| Name | Type | Description |
|---|---|---|
service_config |
SeldonDeploymentConfig |
Seldon Core deployment service configuration. |
secrets |
a list of ZenML secrets containing additional configuration parameters for the Seldon Core deployment (e.g. credentials to access the Artifact Store where the models are stored). If supplied, the information fetched from these secrets is passed to the Seldon Core deployment server as a list of environment variables. |
Source code in zenml/integrations/seldon/steps/seldon_deployer.py
class SeldonDeployerStepConfig(BaseStepConfig):
"""Seldon model deployer step configuration
Attributes:
service_config: Seldon Core deployment service configuration.
secrets: a list of ZenML secrets containing additional configuration
parameters for the Seldon Core deployment (e.g. credentials to
access the Artifact Store where the models are stored). If supplied,
the information fetched from these secrets is passed to the Seldon
Core deployment server as a list of environment variables.
"""
service_config: SeldonDeploymentConfig
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT
seldon_model_deployer_step (BaseStep)
CONFIG_CLASS (BaseStepConfig)
pydantic-model
Seldon model deployer step configuration
Attributes:
| Name | Type | Description |
|---|---|---|
service_config |
SeldonDeploymentConfig |
Seldon Core deployment service configuration. |
secrets |
a list of ZenML secrets containing additional configuration parameters for the Seldon Core deployment (e.g. credentials to access the Artifact Store where the models are stored). If supplied, the information fetched from these secrets is passed to the Seldon Core deployment server as a list of environment variables. |
Source code in zenml/integrations/seldon/steps/seldon_deployer.py
class SeldonDeployerStepConfig(BaseStepConfig):
"""Seldon model deployer step configuration
Attributes:
service_config: Seldon Core deployment service configuration.
secrets: a list of ZenML secrets containing additional configuration
parameters for the Seldon Core deployment (e.g. credentials to
access the Artifact Store where the models are stored). If supplied,
the information fetched from these secrets is passed to the Seldon
Core deployment server as a list of environment variables.
"""
service_config: SeldonDeploymentConfig
timeout: int = DEFAULT_SELDON_DEPLOYMENT_START_STOP_TIMEOUT
entrypoint(deploy_decision, config, context, model)
staticmethod
Seldon Core model deployer pipeline step
This step can be used in a pipeline to implement continuous deployment for a ML model with Seldon Core.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
deploy_decision |
bool |
whether to deploy the model or not |
required |
config |
SeldonDeployerStepConfig |
configuration for the deployer step |
required |
model |
ModelArtifact |
the model artifact to deploy |
required |
Returns:
| Type | Description |
|---|---|
SeldonDeploymentService |
Seldon Core deployment service |
Source code in zenml/integrations/seldon/steps/seldon_deployer.py
@step(enable_cache=True)
def seldon_model_deployer_step(
deploy_decision: bool,
config: SeldonDeployerStepConfig,
context: StepContext,
model: ModelArtifact,
) -> SeldonDeploymentService:
"""Seldon Core model deployer pipeline step
This step can be used in a pipeline to implement continuous
deployment for a ML model with Seldon Core.
Args:
deploy_decision: whether to deploy the model or not
config: configuration for the deployer step
model: the model artifact to deploy
Returns:
Seldon Core deployment service
"""
model_deployer = SeldonModelDeployer.get_active_model_deployer()
# get pipeline name, step name and run id
step_env = cast(StepEnvironment, Environment()[STEP_ENVIRONMENT_NAME])
pipeline_name = step_env.pipeline_name
pipeline_run_id = step_env.pipeline_run_id
step_name = step_env.step_name
# update the step configuration with the real pipeline runtime information
config.service_config.pipeline_name = pipeline_name
config.service_config.pipeline_run_id = pipeline_run_id
config.service_config.pipeline_step_name = step_name
def prepare_service_config(model_uri: str) -> SeldonDeploymentConfig:
"""Prepare the model files for model serving and create and return a
Seldon service configuration for the model.
This function ensures that the model files are in the correct format
and file structure required by the Seldon Core server implementation
used for model serving.
Args:
model_uri: the URI of the model artifact being served
Returns:
The URL to the model ready for serving.
"""
served_model_uri = os.path.join(
context.get_output_artifact_uri(), "seldon"
)
fileio.makedirs(served_model_uri)
# TODO [ENG-773]: determine how to formalize how models are organized into
# folders and sub-folders depending on the model type/format and the
# Seldon Core protocol used to serve the model.
# TODO [ENG-791]: auto-detect built-in Seldon server implementation
# from the model artifact type
# TODO [ENG-792]: validate the model artifact type against the
# supported built-in Seldon server implementations
if config.service_config.implementation == "TENSORFLOW_SERVER":
# the TensorFlow server expects model artifacts to be
# stored in numbered subdirectories, each representing a model
# version
io_utils.copy_dir(model_uri, os.path.join(served_model_uri, "1"))
elif config.service_config.implementation == "SKLEARN_SERVER":
# the sklearn server expects model artifacts to be
# stored in a file called model.joblib
model_uri = os.path.join(model.uri, "model")
if not fileio.exists(model.uri):
raise RuntimeError(
f"Expected sklearn model artifact was not found at "
f"{model_uri}"
)
fileio.copy(
model_uri, os.path.join(served_model_uri, "model.joblib")
)
else:
# default treatment for all other server implementations is to
# simply reuse the model from the artifact store path where it
# is originally stored
served_model_uri = model_uri
service_config = config.service_config.copy()
service_config.model_uri = served_model_uri
return service_config
if not deploy_decision:
logger.info(
f"Skipping model deployment because the model quality does not "
f"meet the criteria. Loading last service deployed by step "
f"'{step_name}' and pipeline '{pipeline_name}' for model "
f"'{config.service_config.model_name}'..."
)
# fetch existing services with same pipeline name, step name and
# model name
existing_services = model_deployer.find_model_server(
pipeline_name=pipeline_name,
pipeline_step_name=step_name,
model_name=config.service_config.model_name,
)
if existing_services:
service = cast(SeldonDeploymentService, existing_services[0])
else:
# if a previous service was not found, but we also don't need to
# create a new one in this iteration because the deployment
# decision didn't come through, we simply return a
# placeholder service
service_config = prepare_service_config(model.uri)
service = SeldonDeploymentService(config=service_config)
return service
service_config = prepare_service_config(model.uri)
# invoke the Seldon Core model deployer to create a new service
# or update an existing one that was previously deployed for the same
# model
service = cast(
SeldonDeploymentService,
model_deployer.deploy_model(
service_config, replace=True, timeout=config.timeout
),
)
logger.info(
f"Seldon deployment service started and reachable at:\n"
f" {service.prediction_url}\n"
)
return service
sklearn
special
SklearnIntegration (Integration)
Definition of sklearn integration for ZenML.
Source code in zenml/integrations/sklearn/__init__.py
class SklearnIntegration(Integration):
"""Definition of sklearn integration for ZenML."""
NAME = SKLEARN
REQUIREMENTS = ["scikit-learn"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.sklearn import materializers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/sklearn/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.sklearn import materializers # noqa
helpers
special
digits
get_digits()
Returns the digits dataset in the form of a tuple of numpy arrays.
Source code in zenml/integrations/sklearn/helpers/digits.py
def get_digits() -> Tuple[
"NDArray[np.float64]",
"NDArray[np.float64]",
"NDArray[np.int64]",
"NDArray[np.int64]",
]:
"""Returns the digits dataset in the form of a tuple of numpy
arrays."""
digits = load_digits()
# flatten the images
n_samples = len(digits.images)
data = digits.images.reshape((n_samples, -1))
# Split data into 50% train and 50% test subsets
X_train, X_test, y_train, y_test = train_test_split(
data, digits.target, test_size=0.5, shuffle=False
)
return X_train, X_test, y_train, y_test
get_digits_model()
Creates a support vector classifier for digits dataset.
Source code in zenml/integrations/sklearn/helpers/digits.py
def get_digits_model() -> ClassifierMixin:
"""Creates a support vector classifier for digits dataset."""
return SVC(gamma=0.001)
materializers
special
sklearn_materializer
SklearnMaterializer (BaseMaterializer)
Materializer to read data to and from sklearn.
Source code in zenml/integrations/sklearn/materializers/sklearn_materializer.py
class SklearnMaterializer(BaseMaterializer):
"""Materializer to read data to and from sklearn."""
ASSOCIATED_TYPES = (
BaseEstimator,
ClassifierMixin,
ClusterMixin,
BiclusterMixin,
OutlierMixin,
RegressorMixin,
MetaEstimatorMixin,
MultiOutputMixin,
DensityMixin,
TransformerMixin,
)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(
self, data_type: Type[Any]
) -> Union[
BaseEstimator,
ClassifierMixin,
ClusterMixin,
BiclusterMixin,
OutlierMixin,
RegressorMixin,
MetaEstimatorMixin,
MultiOutputMixin,
DensityMixin,
TransformerMixin,
]:
"""Reads a base sklearn model from a pickle file."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
with fileio.open(filepath, "rb") as fid:
clf = pickle.load(fid)
return clf
def handle_return(
self,
clf: Union[
BaseEstimator,
ClassifierMixin,
ClusterMixin,
BiclusterMixin,
OutlierMixin,
RegressorMixin,
MetaEstimatorMixin,
MultiOutputMixin,
DensityMixin,
TransformerMixin,
],
) -> None:
"""Creates a pickle for a sklearn model.
Args:
clf: A sklearn model.
"""
super().handle_return(clf)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
with fileio.open(filepath, "wb") as fid:
pickle.dump(clf, fid)
handle_input(self, data_type)
Reads a base sklearn model from a pickle file.
Source code in zenml/integrations/sklearn/materializers/sklearn_materializer.py
def handle_input(
self, data_type: Type[Any]
) -> Union[
BaseEstimator,
ClassifierMixin,
ClusterMixin,
BiclusterMixin,
OutlierMixin,
RegressorMixin,
MetaEstimatorMixin,
MultiOutputMixin,
DensityMixin,
TransformerMixin,
]:
"""Reads a base sklearn model from a pickle file."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
with fileio.open(filepath, "rb") as fid:
clf = pickle.load(fid)
return clf
handle_return(self, clf)
Creates a pickle for a sklearn model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
clf |
Union[sklearn.base.BaseEstimator, sklearn.base.ClassifierMixin, sklearn.base.ClusterMixin, sklearn.base.BiclusterMixin, sklearn.base.OutlierMixin, sklearn.base.RegressorMixin, sklearn.base.MetaEstimatorMixin, sklearn.base.MultiOutputMixin, sklearn.base.DensityMixin, sklearn.base.TransformerMixin] |
A sklearn model. |
required |
Source code in zenml/integrations/sklearn/materializers/sklearn_materializer.py
def handle_return(
self,
clf: Union[
BaseEstimator,
ClassifierMixin,
ClusterMixin,
BiclusterMixin,
OutlierMixin,
RegressorMixin,
MetaEstimatorMixin,
MultiOutputMixin,
DensityMixin,
TransformerMixin,
],
) -> None:
"""Creates a pickle for a sklearn model.
Args:
clf: A sklearn model.
"""
super().handle_return(clf)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
with fileio.open(filepath, "wb") as fid:
pickle.dump(clf, fid)
steps
special
sklearn_evaluator
SklearnEvaluator (BaseEvaluatorStep)
A simple step implementation which utilizes sklearn to evaluate the performance of a given model on a given test dataset
Source code in zenml/integrations/sklearn/steps/sklearn_evaluator.py
class SklearnEvaluator(BaseEvaluatorStep):
"""A simple step implementation which utilizes sklearn to evaluate the
performance of a given model on a given test dataset"""
def entrypoint( # type: ignore[override]
self,
dataset: pd.DataFrame,
model: tf.keras.Model,
config: SklearnEvaluatorConfig,
) -> dict: # type: ignore[type-arg]
"""Method which is responsible for the computation of the evaluation
Args:
dataset: a pandas Dataframe which represents the test dataset
model: a trained tensorflow Keras model
config: the configuration for the step
Returns:
a dictionary which has the evaluation report
"""
labels = dataset.pop(config.label_class_column)
predictions = model.predict(dataset)
predicted_classes = [1 if v > 0.5 else 0 for v in predictions]
report = classification_report(
labels, predicted_classes, output_dict=True
)
return report # type: ignore[no-any-return]
CONFIG_CLASS (BaseEvaluatorConfig)
pydantic-model
Config class for the sklearn evaluator
Source code in zenml/integrations/sklearn/steps/sklearn_evaluator.py
class SklearnEvaluatorConfig(BaseEvaluatorConfig):
"""Config class for the sklearn evaluator"""
label_class_column: str
entrypoint(self, dataset, model, config)
Method which is responsible for the computation of the evaluation
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
dataset |
DataFrame |
a pandas Dataframe which represents the test dataset |
required |
model |
Model |
a trained tensorflow Keras model |
required |
config |
SklearnEvaluatorConfig |
the configuration for the step |
required |
Returns:
| Type | Description |
|---|---|
dict |
a dictionary which has the evaluation report |
Source code in zenml/integrations/sklearn/steps/sklearn_evaluator.py
def entrypoint( # type: ignore[override]
self,
dataset: pd.DataFrame,
model: tf.keras.Model,
config: SklearnEvaluatorConfig,
) -> dict: # type: ignore[type-arg]
"""Method which is responsible for the computation of the evaluation
Args:
dataset: a pandas Dataframe which represents the test dataset
model: a trained tensorflow Keras model
config: the configuration for the step
Returns:
a dictionary which has the evaluation report
"""
labels = dataset.pop(config.label_class_column)
predictions = model.predict(dataset)
predicted_classes = [1 if v > 0.5 else 0 for v in predictions]
report = classification_report(
labels, predicted_classes, output_dict=True
)
return report # type: ignore[no-any-return]
SklearnEvaluatorConfig (BaseEvaluatorConfig)
pydantic-model
Config class for the sklearn evaluator
Source code in zenml/integrations/sklearn/steps/sklearn_evaluator.py
class SklearnEvaluatorConfig(BaseEvaluatorConfig):
"""Config class for the sklearn evaluator"""
label_class_column: str
sklearn_splitter
SklearnSplitter (BaseSplitStep)
A simple step implementation which utilizes sklearn to split a given dataset into train, test and validation splits
Source code in zenml/integrations/sklearn/steps/sklearn_splitter.py
class SklearnSplitter(BaseSplitStep):
"""A simple step implementation which utilizes sklearn to split a given
dataset into train, test and validation splits"""
def entrypoint( # type: ignore[override]
self,
dataset: pd.DataFrame,
config: SklearnSplitterConfig,
) -> Output( # type:ignore[valid-type]
train=pd.DataFrame, test=pd.DataFrame, validation=pd.DataFrame
):
"""Method which is responsible for the splitting logic
Args:
dataset: a pandas Dataframe which entire dataset
config: the configuration for the step
Returns:
three dataframes representing the splits
"""
if (
any(
[
split not in config.ratios
for split in ["train", "test", "validation"]
]
)
or len(config.ratios) != 3
):
raise KeyError(
f"Make sure that you only use 'train', 'test' and "
f"'validation' as keys in the ratios dict. Current keys: "
f"{config.ratios.keys()}"
)
if sum(config.ratios.values()) != 1:
raise ValueError(
f"Make sure that the ratios sum up to 1. Current "
f"ratios: {config.ratios}"
)
train_dataset, test_dataset = train_test_split(
dataset, test_size=config.ratios["test"]
)
train_dataset, val_dataset = train_test_split(
train_dataset,
test_size=(
config.ratios["validation"]
/ (config.ratios["validation"] + config.ratios["train"])
),
)
return train_dataset, test_dataset, val_dataset
CONFIG_CLASS (BaseSplitStepConfig)
pydantic-model
Config class for the sklearn splitter
Source code in zenml/integrations/sklearn/steps/sklearn_splitter.py
class SklearnSplitterConfig(BaseSplitStepConfig):
"""Config class for the sklearn splitter"""
ratios: Dict[str, float]
entrypoint(self, dataset, config)
Method which is responsible for the splitting logic
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
dataset |
DataFrame |
a pandas Dataframe which entire dataset |
required |
config |
SklearnSplitterConfig |
the configuration for the step |
required |
Returns:
| Type | Description |
|---|---|
<zenml.steps.step_output.Output object at 0x7f1e876b5b50> |
three dataframes representing the splits |
Source code in zenml/integrations/sklearn/steps/sklearn_splitter.py
def entrypoint( # type: ignore[override]
self,
dataset: pd.DataFrame,
config: SklearnSplitterConfig,
) -> Output( # type:ignore[valid-type]
train=pd.DataFrame, test=pd.DataFrame, validation=pd.DataFrame
):
"""Method which is responsible for the splitting logic
Args:
dataset: a pandas Dataframe which entire dataset
config: the configuration for the step
Returns:
three dataframes representing the splits
"""
if (
any(
[
split not in config.ratios
for split in ["train", "test", "validation"]
]
)
or len(config.ratios) != 3
):
raise KeyError(
f"Make sure that you only use 'train', 'test' and "
f"'validation' as keys in the ratios dict. Current keys: "
f"{config.ratios.keys()}"
)
if sum(config.ratios.values()) != 1:
raise ValueError(
f"Make sure that the ratios sum up to 1. Current "
f"ratios: {config.ratios}"
)
train_dataset, test_dataset = train_test_split(
dataset, test_size=config.ratios["test"]
)
train_dataset, val_dataset = train_test_split(
train_dataset,
test_size=(
config.ratios["validation"]
/ (config.ratios["validation"] + config.ratios["train"])
),
)
return train_dataset, test_dataset, val_dataset
SklearnSplitterConfig (BaseSplitStepConfig)
pydantic-model
Config class for the sklearn splitter
Source code in zenml/integrations/sklearn/steps/sklearn_splitter.py
class SklearnSplitterConfig(BaseSplitStepConfig):
"""Config class for the sklearn splitter"""
ratios: Dict[str, float]
sklearn_standard_scaler
SklearnStandardScaler (BasePreprocessorStep)
Simple step implementation which utilizes the StandardScaler from sklearn to transform the numeric columns of a pd.DataFrame
Source code in zenml/integrations/sklearn/steps/sklearn_standard_scaler.py
class SklearnStandardScaler(BasePreprocessorStep):
"""Simple step implementation which utilizes the StandardScaler from sklearn
to transform the numeric columns of a pd.DataFrame"""
def entrypoint( # type: ignore[override]
self,
train_dataset: pd.DataFrame,
test_dataset: pd.DataFrame,
validation_dataset: pd.DataFrame,
statistics: pd.DataFrame,
schema: pd.DataFrame,
config: SklearnStandardScalerConfig,
) -> Output( # type:ignore[valid-type]
train_transformed=pd.DataFrame,
test_transformed=pd.DataFrame,
validation_transformed=pd.DataFrame,
):
"""Main entrypoint function for the StandardScaler
Args:
train_dataset: pd.DataFrame, the training dataset
test_dataset: pd.DataFrame, the test dataset
validation_dataset: pd.DataFrame, the validation dataset
statistics: pd.DataFrame, the statistics over the train dataset
schema: pd.DataFrame, the detected schema of the dataset
config: the configuration for the step
Returns:
the transformed train, test and validation datasets as
pd.DataFrames
"""
schema_dict = {k: v[0] for k, v in schema.to_dict().items()}
# Exclude columns
feature_set = set(train_dataset.columns) - set(config.exclude_columns)
for feature, feature_type in schema_dict.items():
if feature_type != "int64" and feature_type != "float64":
feature_set.remove(feature)
logger.warning(
f"{feature} column is a not numeric, thus it is excluded "
f"from the standard scaling."
)
transform_feature_set = feature_set - set(config.ignore_columns)
# Transform the datasets
scaler = StandardScaler()
scaler.mean_ = statistics["mean"][transform_feature_set]
scaler.scale_ = statistics["std"][transform_feature_set]
train_dataset[list(transform_feature_set)] = scaler.transform(
train_dataset[transform_feature_set]
)
test_dataset[list(transform_feature_set)] = scaler.transform(
test_dataset[transform_feature_set]
)
validation_dataset[list(transform_feature_set)] = scaler.transform(
validation_dataset[transform_feature_set]
)
return train_dataset, test_dataset, validation_dataset
CONFIG_CLASS (BasePreprocessorConfig)
pydantic-model
Config class for the sklearn standard scaler
ignore_columns: a list of column names which should not be scaled exclude_columns: a list of column names to be excluded from the dataset
Source code in zenml/integrations/sklearn/steps/sklearn_standard_scaler.py
class SklearnStandardScalerConfig(BasePreprocessorConfig):
"""Config class for the sklearn standard scaler
ignore_columns: a list of column names which should not be scaled
exclude_columns: a list of column names to be excluded from the dataset
"""
ignore_columns: List[str] = []
exclude_columns: List[str] = []
entrypoint(self, train_dataset, test_dataset, validation_dataset, statistics, schema, config)
Main entrypoint function for the StandardScaler
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
train_dataset |
DataFrame |
pd.DataFrame, the training dataset |
required |
test_dataset |
DataFrame |
pd.DataFrame, the test dataset |
required |
validation_dataset |
DataFrame |
pd.DataFrame, the validation dataset |
required |
statistics |
DataFrame |
pd.DataFrame, the statistics over the train dataset |
required |
schema |
DataFrame |
pd.DataFrame, the detected schema of the dataset |
required |
config |
SklearnStandardScalerConfig |
the configuration for the step |
required |
Returns:
| Type | Description |
|---|---|
<zenml.steps.step_output.Output object at 0x7f1e876ca820> |
the transformed train, test and validation datasets as pd.DataFrames |
Source code in zenml/integrations/sklearn/steps/sklearn_standard_scaler.py
def entrypoint( # type: ignore[override]
self,
train_dataset: pd.DataFrame,
test_dataset: pd.DataFrame,
validation_dataset: pd.DataFrame,
statistics: pd.DataFrame,
schema: pd.DataFrame,
config: SklearnStandardScalerConfig,
) -> Output( # type:ignore[valid-type]
train_transformed=pd.DataFrame,
test_transformed=pd.DataFrame,
validation_transformed=pd.DataFrame,
):
"""Main entrypoint function for the StandardScaler
Args:
train_dataset: pd.DataFrame, the training dataset
test_dataset: pd.DataFrame, the test dataset
validation_dataset: pd.DataFrame, the validation dataset
statistics: pd.DataFrame, the statistics over the train dataset
schema: pd.DataFrame, the detected schema of the dataset
config: the configuration for the step
Returns:
the transformed train, test and validation datasets as
pd.DataFrames
"""
schema_dict = {k: v[0] for k, v in schema.to_dict().items()}
# Exclude columns
feature_set = set(train_dataset.columns) - set(config.exclude_columns)
for feature, feature_type in schema_dict.items():
if feature_type != "int64" and feature_type != "float64":
feature_set.remove(feature)
logger.warning(
f"{feature} column is a not numeric, thus it is excluded "
f"from the standard scaling."
)
transform_feature_set = feature_set - set(config.ignore_columns)
# Transform the datasets
scaler = StandardScaler()
scaler.mean_ = statistics["mean"][transform_feature_set]
scaler.scale_ = statistics["std"][transform_feature_set]
train_dataset[list(transform_feature_set)] = scaler.transform(
train_dataset[transform_feature_set]
)
test_dataset[list(transform_feature_set)] = scaler.transform(
test_dataset[transform_feature_set]
)
validation_dataset[list(transform_feature_set)] = scaler.transform(
validation_dataset[transform_feature_set]
)
return train_dataset, test_dataset, validation_dataset
SklearnStandardScalerConfig (BasePreprocessorConfig)
pydantic-model
Config class for the sklearn standard scaler
ignore_columns: a list of column names which should not be scaled exclude_columns: a list of column names to be excluded from the dataset
Source code in zenml/integrations/sklearn/steps/sklearn_standard_scaler.py
class SklearnStandardScalerConfig(BasePreprocessorConfig):
"""Config class for the sklearn standard scaler
ignore_columns: a list of column names which should not be scaled
exclude_columns: a list of column names to be excluded from the dataset
"""
ignore_columns: List[str] = []
exclude_columns: List[str] = []
tensorflow
special
TensorflowIntegration (Integration)
Definition of Tensorflow integration for ZenML.
Source code in zenml/integrations/tensorflow/__init__.py
class TensorflowIntegration(Integration):
"""Definition of Tensorflow integration for ZenML."""
NAME = TENSORFLOW
REQUIREMENTS = ["tensorflow==2.8.0", "tensorflow_io==0.24.0"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
# need to import this explicitly to load the Tensoflow file IO support
# for S3 and other file systems
import tensorflow_io # type: ignore [import]
from zenml.integrations.tensorflow import materializers # noqa
from zenml.integrations.tensorflow import services # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/tensorflow/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
# need to import this explicitly to load the Tensoflow file IO support
# for S3 and other file systems
import tensorflow_io # type: ignore [import]
from zenml.integrations.tensorflow import materializers # noqa
from zenml.integrations.tensorflow import services # noqa
materializers
special
keras_materializer
KerasMaterializer (BaseMaterializer)
Materializer to read/write Keras models.
Source code in zenml/integrations/tensorflow/materializers/keras_materializer.py
class KerasMaterializer(BaseMaterializer):
"""Materializer to read/write Keras models."""
ASSOCIATED_TYPES = (keras.Model,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> keras.Model:
"""Reads and returns a Keras model.
Returns:
A tf.keras.Model model.
"""
super().handle_input(data_type)
return keras.models.load_model(self.artifact.uri)
def handle_return(self, model: keras.Model) -> None:
"""Writes a keras model.
Args:
model: A tf.keras.Model model.
"""
super().handle_return(model)
model.save(self.artifact.uri)
handle_input(self, data_type)
Reads and returns a Keras model.
Returns:
| Type | Description |
|---|---|
Model |
A tf.keras.Model model. |
Source code in zenml/integrations/tensorflow/materializers/keras_materializer.py
def handle_input(self, data_type: Type[Any]) -> keras.Model:
"""Reads and returns a Keras model.
Returns:
A tf.keras.Model model.
"""
super().handle_input(data_type)
return keras.models.load_model(self.artifact.uri)
handle_return(self, model)
Writes a keras model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
model |
Model |
A tf.keras.Model model. |
required |
Source code in zenml/integrations/tensorflow/materializers/keras_materializer.py
def handle_return(self, model: keras.Model) -> None:
"""Writes a keras model.
Args:
model: A tf.keras.Model model.
"""
super().handle_return(model)
model.save(self.artifact.uri)
tf_dataset_materializer
TensorflowDatasetMaterializer (BaseMaterializer)
Materializer to read data to and from tf.data.Dataset.
Source code in zenml/integrations/tensorflow/materializers/tf_dataset_materializer.py
class TensorflowDatasetMaterializer(BaseMaterializer):
"""Materializer to read data to and from tf.data.Dataset."""
ASSOCIATED_TYPES = (tf.data.Dataset,)
ASSOCIATED_ARTIFACT_TYPES = (DataArtifact,)
def handle_input(self, data_type: Type[Any]) -> Any:
"""Reads data into tf.data.Dataset"""
super().handle_input(data_type)
path = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
return tf.data.experimental.load(path)
def handle_return(self, dataset: tf.data.Dataset) -> None:
"""Persists a tf.data.Dataset object."""
super().handle_return(dataset)
path = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
tf.data.experimental.save(
dataset, path, compression=None, shard_func=None
)
handle_input(self, data_type)
Reads data into tf.data.Dataset
Source code in zenml/integrations/tensorflow/materializers/tf_dataset_materializer.py
def handle_input(self, data_type: Type[Any]) -> Any:
"""Reads data into tf.data.Dataset"""
super().handle_input(data_type)
path = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
return tf.data.experimental.load(path)
handle_return(self, dataset)
Persists a tf.data.Dataset object.
Source code in zenml/integrations/tensorflow/materializers/tf_dataset_materializer.py
def handle_return(self, dataset: tf.data.Dataset) -> None:
"""Persists a tf.data.Dataset object."""
super().handle_return(dataset)
path = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
tf.data.experimental.save(
dataset, path, compression=None, shard_func=None
)
services
special
tensorboard_service
TensorboardService (LocalDaemonService)
pydantic-model
Tensorboard service that can be used to start a local Tensorboard server for one or more models.
Attributes:
| Name | Type | Description |
|---|---|---|
SERVICE_TYPE |
ClassVar[zenml.services.service_type.ServiceType] |
a service type descriptor with information describing the Tensorboard service class |
config |
TensorboardServiceConfig |
service configuration |
endpoint |
LocalDaemonServiceEndpoint |
optional service endpoint |
Source code in zenml/integrations/tensorflow/services/tensorboard_service.py
class TensorboardService(LocalDaemonService):
"""Tensorboard service that can be used to start a local Tensorboard server
for one or more models.
Attributes:
SERVICE_TYPE: a service type descriptor with information describing
the Tensorboard service class
config: service configuration
endpoint: optional service endpoint
"""
SERVICE_TYPE = ServiceType(
name="tensorboard",
type="visualization",
flavor="tensorboard",
description="Tensorboard visualization service",
)
config: TensorboardServiceConfig
endpoint: LocalDaemonServiceEndpoint
def __init__(
self,
config: Union[TensorboardServiceConfig, Dict[str, Any]],
**attrs: Any,
) -> None:
# ensure that the endpoint is created before the service is initialized
# TODO [ENG-697]: implement a service factory or builder for Tensorboard
# deployment services
if (
isinstance(config, TensorboardServiceConfig)
and "endpoint" not in attrs
):
endpoint = LocalDaemonServiceEndpoint(
config=LocalDaemonServiceEndpointConfig(
protocol=ServiceEndpointProtocol.HTTP,
),
monitor=HTTPEndpointHealthMonitor(
config=HTTPEndpointHealthMonitorConfig(
healthcheck_uri_path="",
use_head_request=True,
)
),
)
attrs["endpoint"] = endpoint
super().__init__(config=config, **attrs)
def run(self) -> None:
logger.info(
"Starting Tensorboard service as blocking "
"process... press CTRL+C once to stop it."
)
self.endpoint.prepare_for_start()
try:
tensorboard = program.TensorBoard(
plugins=default.get_plugins(),
subcommands=[uploader_subcommand.UploaderSubcommand()],
)
tensorboard.configure(
logdir=self.config.logdir,
port=self.endpoint.status.port,
host="localhost",
max_reload_threads=self.config.max_reload_threads,
reload_interval=self.config.reload_interval,
)
tensorboard.main()
except KeyboardInterrupt:
logger.info(
"Tensorboard service stopped. Resuming normal execution."
)
run(self)
Run the service daemon process associated with this service.
Subclasses must implement this method to provide the service daemon
functionality. This method will be executed in the context of the
running daemon, not in the context of the process that calls the
start method.
Source code in zenml/integrations/tensorflow/services/tensorboard_service.py
def run(self) -> None:
logger.info(
"Starting Tensorboard service as blocking "
"process... press CTRL+C once to stop it."
)
self.endpoint.prepare_for_start()
try:
tensorboard = program.TensorBoard(
plugins=default.get_plugins(),
subcommands=[uploader_subcommand.UploaderSubcommand()],
)
tensorboard.configure(
logdir=self.config.logdir,
port=self.endpoint.status.port,
host="localhost",
max_reload_threads=self.config.max_reload_threads,
reload_interval=self.config.reload_interval,
)
tensorboard.main()
except KeyboardInterrupt:
logger.info(
"Tensorboard service stopped. Resuming normal execution."
)
TensorboardServiceConfig (LocalDaemonServiceConfig)
pydantic-model
Tensorboard service configuration.
Attributes:
| Name | Type | Description |
|---|---|---|
logdir |
str |
location of Tensorboard log files. |
max_reload_threads |
int |
the max number of threads that TensorBoard can use to reload runs. Each thread reloads one run at a time. |
reload_interval |
int |
how often the backend should load more data, in seconds. Set to 0 to load just once at startup. |
Source code in zenml/integrations/tensorflow/services/tensorboard_service.py
class TensorboardServiceConfig(LocalDaemonServiceConfig):
"""Tensorboard service configuration.
Attributes:
logdir: location of Tensorboard log files.
max_reload_threads: the max number of threads that TensorBoard can use
to reload runs. Each thread reloads one run at a time.
reload_interval: how often the backend should load more data, in
seconds. Set to 0 to load just once at startup.
"""
logdir: str
max_reload_threads: int = 1
reload_interval: int = 5
steps
special
tensorflow_trainer
TensorflowBinaryClassifier (BaseTrainerStep)
Simple step implementation which creates a simple tensorflow feedforward neural network and trains it on a given pd.DataFrame dataset
Source code in zenml/integrations/tensorflow/steps/tensorflow_trainer.py
class TensorflowBinaryClassifier(BaseTrainerStep):
"""Simple step implementation which creates a simple tensorflow feedforward
neural network and trains it on a given pd.DataFrame dataset
"""
def entrypoint( # type: ignore[override]
self,
train_dataset: pd.DataFrame,
validation_dataset: pd.DataFrame,
config: TensorflowBinaryClassifierConfig,
) -> tf.keras.Model:
"""Main entrypoint for the tensorflow trainer
Args:
train_dataset: pd.DataFrame, the training dataset
validation_dataset: pd.DataFrame, the validation dataset
config: the configuration of the step
Returns:
the trained tf.keras.Model
"""
model = tf.keras.Sequential()
model.add(tf.keras.layers.InputLayer(input_shape=config.input_shape))
model.add(tf.keras.layers.Flatten())
last_layer = config.layers.pop()
for i, layer in enumerate(config.layers):
model.add(tf.keras.layers.Dense(layer, activation="relu"))
model.add(tf.keras.layers.Dense(last_layer, activation="sigmoid"))
model.compile(
optimizer=tf.keras.optimizers.Adam(config.learning_rate),
loss=tf.keras.losses.BinaryCrossentropy(),
metrics=config.metrics,
)
train_target = train_dataset.pop(config.target_column)
validation_target = validation_dataset.pop(config.target_column)
model.fit(
x=train_dataset,
y=train_target,
validation_data=(validation_dataset, validation_target),
batch_size=config.batch_size,
epochs=config.epochs,
)
model.summary()
return model
CONFIG_CLASS (BaseTrainerConfig)
pydantic-model
Config class for the tensorflow trainer
target_column: the name of the label column layers: the number of units in the fully connected layers input_shape: the shape of the input learning_rate: the learning rate metrics: the list of metrics to be computed epochs: the number of epochs batch_size: the size of the batch
Source code in zenml/integrations/tensorflow/steps/tensorflow_trainer.py
class TensorflowBinaryClassifierConfig(BaseTrainerConfig):
"""Config class for the tensorflow trainer
target_column: the name of the label column
layers: the number of units in the fully connected layers
input_shape: the shape of the input
learning_rate: the learning rate
metrics: the list of metrics to be computed
epochs: the number of epochs
batch_size: the size of the batch
"""
target_column: str
layers: List[int] = [256, 64, 1]
input_shape: Tuple[int] = (8,)
learning_rate: float = 0.001
metrics: List[str] = ["accuracy"]
epochs: int = 50
batch_size: int = 8
entrypoint(self, train_dataset, validation_dataset, config)
Main entrypoint for the tensorflow trainer
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
train_dataset |
DataFrame |
pd.DataFrame, the training dataset |
required |
validation_dataset |
DataFrame |
pd.DataFrame, the validation dataset |
required |
config |
TensorflowBinaryClassifierConfig |
the configuration of the step |
required |
Returns:
| Type | Description |
|---|---|
Model |
the trained tf.keras.Model |
Source code in zenml/integrations/tensorflow/steps/tensorflow_trainer.py
def entrypoint( # type: ignore[override]
self,
train_dataset: pd.DataFrame,
validation_dataset: pd.DataFrame,
config: TensorflowBinaryClassifierConfig,
) -> tf.keras.Model:
"""Main entrypoint for the tensorflow trainer
Args:
train_dataset: pd.DataFrame, the training dataset
validation_dataset: pd.DataFrame, the validation dataset
config: the configuration of the step
Returns:
the trained tf.keras.Model
"""
model = tf.keras.Sequential()
model.add(tf.keras.layers.InputLayer(input_shape=config.input_shape))
model.add(tf.keras.layers.Flatten())
last_layer = config.layers.pop()
for i, layer in enumerate(config.layers):
model.add(tf.keras.layers.Dense(layer, activation="relu"))
model.add(tf.keras.layers.Dense(last_layer, activation="sigmoid"))
model.compile(
optimizer=tf.keras.optimizers.Adam(config.learning_rate),
loss=tf.keras.losses.BinaryCrossentropy(),
metrics=config.metrics,
)
train_target = train_dataset.pop(config.target_column)
validation_target = validation_dataset.pop(config.target_column)
model.fit(
x=train_dataset,
y=train_target,
validation_data=(validation_dataset, validation_target),
batch_size=config.batch_size,
epochs=config.epochs,
)
model.summary()
return model
TensorflowBinaryClassifierConfig (BaseTrainerConfig)
pydantic-model
Config class for the tensorflow trainer
target_column: the name of the label column layers: the number of units in the fully connected layers input_shape: the shape of the input learning_rate: the learning rate metrics: the list of metrics to be computed epochs: the number of epochs batch_size: the size of the batch
Source code in zenml/integrations/tensorflow/steps/tensorflow_trainer.py
class TensorflowBinaryClassifierConfig(BaseTrainerConfig):
"""Config class for the tensorflow trainer
target_column: the name of the label column
layers: the number of units in the fully connected layers
input_shape: the shape of the input
learning_rate: the learning rate
metrics: the list of metrics to be computed
epochs: the number of epochs
batch_size: the size of the batch
"""
target_column: str
layers: List[int] = [256, 64, 1]
input_shape: Tuple[int] = (8,)
learning_rate: float = 0.001
metrics: List[str] = ["accuracy"]
epochs: int = 50
batch_size: int = 8
visualizers
special
tensorboard_visualizer
TensorboardVisualizer (BaseStepVisualizer)
The implementation of a Whylogs Visualizer.
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
class TensorboardVisualizer(BaseStepVisualizer):
"""The implementation of a Whylogs Visualizer."""
@classmethod
def find_running_tensorboard_server(
cls, logdir: str
) -> Optional[TensorBoardInfo]:
"""Find a local Tensorboard server instance running for the supplied
logdir location and return its TCP port.
Returns:
The TensorBoardInfo describing the running Tensorboard server or
None if no server is running for the supplied logdir location.
"""
for server in get_all():
if (
server.logdir == logdir
and server.pid
and psutil.pid_exists(server.pid)
):
return server
return None
def visualize(
self,
object: StepView,
*args: Any,
height: int = 800,
**kwargs: Any,
) -> None:
"""Start a Tensorboard server to visualize all models logged as
artifacts by the indicated step. The server will monitor and display
all the models logged by past and future step runs.
Args:
object: StepView fetched from run.get_step().
height: Height of the generated visualization.
"""
for _, artifact_view in object.outputs.items():
# filter out anything but model artifacts
if artifact_view.type == ModelArtifact.TYPE_NAME:
logdir = os.path.dirname(artifact_view.uri)
# first check if a Tensorboard server is already running for
# the same logdir location and use that one
running_server = self.find_running_tensorboard_server(logdir)
if running_server:
self.visualize_tensorboard(running_server.port, height)
return
if sys.platform == "win32":
# Daemon service functionality is currently not supported on Windows
print(
"You can run:\n"
f"[italic green] tensorboard --logdir {logdir}"
"[/italic green]\n"
"...to visualize the Tensorboard logs for your trained model."
)
else:
# start a new Tensorboard server
service = TensorboardService(
TensorboardServiceConfig(
logdir=logdir,
)
)
service.start(timeout=20)
if service.endpoint.status.port:
self.visualize_tensorboard(
service.endpoint.status.port, height
)
return
def visualize_tensorboard(
self,
port: int,
height: int,
) -> None:
"""Generate a visualization of a Tensorboard.
Args:
port: the TCP port where the Tensorboard server is listening for
requests.
height: Height of the generated visualization.
logdir: The logdir location for the Tensorboard server.
"""
if Environment.in_notebook():
notebook.display(port, height=height)
return
print(
"You can visit:\n"
f"[italic green] http://localhost:{port}/[/italic green]\n"
"...to visualize the Tensorboard logs for your trained model."
)
def stop(
self,
object: StepView,
) -> None:
"""Stop the Tensorboard server previously started for a pipeline step.
Args:
object: StepView fetched from run.get_step().
"""
for _, artifact_view in object.outputs.items():
# filter out anything but model artifacts
if artifact_view.type == ModelArtifact.TYPE_NAME:
logdir = os.path.dirname(artifact_view.uri)
# first check if a Tensorboard server is already running for
# the same logdir location and use that one
running_server = self.find_running_tensorboard_server(logdir)
if not running_server:
return
logger.debug(
"Stopping tensorboard server with PID '%d' ...",
running_server.pid,
)
try:
p = psutil.Process(running_server.pid)
except psutil.Error:
logger.error(
"Could not find process for PID '%d' ...",
running_server.pid,
)
continue
p.kill()
return
find_running_tensorboard_server(logdir)
classmethod
Find a local Tensorboard server instance running for the supplied logdir location and return its TCP port.
Returns:
| Type | Description |
|---|---|
Optional[tensorboard.manager.TensorBoardInfo] |
The TensorBoardInfo describing the running Tensorboard server or None if no server is running for the supplied logdir location. |
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
@classmethod
def find_running_tensorboard_server(
cls, logdir: str
) -> Optional[TensorBoardInfo]:
"""Find a local Tensorboard server instance running for the supplied
logdir location and return its TCP port.
Returns:
The TensorBoardInfo describing the running Tensorboard server or
None if no server is running for the supplied logdir location.
"""
for server in get_all():
if (
server.logdir == logdir
and server.pid
and psutil.pid_exists(server.pid)
):
return server
return None
stop(self, object)
Stop the Tensorboard server previously started for a pipeline step.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
object |
StepView |
StepView fetched from run.get_step(). |
required |
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
def stop(
self,
object: StepView,
) -> None:
"""Stop the Tensorboard server previously started for a pipeline step.
Args:
object: StepView fetched from run.get_step().
"""
for _, artifact_view in object.outputs.items():
# filter out anything but model artifacts
if artifact_view.type == ModelArtifact.TYPE_NAME:
logdir = os.path.dirname(artifact_view.uri)
# first check if a Tensorboard server is already running for
# the same logdir location and use that one
running_server = self.find_running_tensorboard_server(logdir)
if not running_server:
return
logger.debug(
"Stopping tensorboard server with PID '%d' ...",
running_server.pid,
)
try:
p = psutil.Process(running_server.pid)
except psutil.Error:
logger.error(
"Could not find process for PID '%d' ...",
running_server.pid,
)
continue
p.kill()
return
visualize(self, object, *args, *, height=800, **kwargs)
Start a Tensorboard server to visualize all models logged as artifacts by the indicated step. The server will monitor and display all the models logged by past and future step runs.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
object |
StepView |
StepView fetched from run.get_step(). |
required |
height |
int |
Height of the generated visualization. |
800 |
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
def visualize(
self,
object: StepView,
*args: Any,
height: int = 800,
**kwargs: Any,
) -> None:
"""Start a Tensorboard server to visualize all models logged as
artifacts by the indicated step. The server will monitor and display
all the models logged by past and future step runs.
Args:
object: StepView fetched from run.get_step().
height: Height of the generated visualization.
"""
for _, artifact_view in object.outputs.items():
# filter out anything but model artifacts
if artifact_view.type == ModelArtifact.TYPE_NAME:
logdir = os.path.dirname(artifact_view.uri)
# first check if a Tensorboard server is already running for
# the same logdir location and use that one
running_server = self.find_running_tensorboard_server(logdir)
if running_server:
self.visualize_tensorboard(running_server.port, height)
return
if sys.platform == "win32":
# Daemon service functionality is currently not supported on Windows
print(
"You can run:\n"
f"[italic green] tensorboard --logdir {logdir}"
"[/italic green]\n"
"...to visualize the Tensorboard logs for your trained model."
)
else:
# start a new Tensorboard server
service = TensorboardService(
TensorboardServiceConfig(
logdir=logdir,
)
)
service.start(timeout=20)
if service.endpoint.status.port:
self.visualize_tensorboard(
service.endpoint.status.port, height
)
return
visualize_tensorboard(self, port, height)
Generate a visualization of a Tensorboard.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
port |
int |
the TCP port where the Tensorboard server is listening for requests. |
required |
height |
int |
Height of the generated visualization. |
required |
logdir |
The logdir location for the Tensorboard server. |
required |
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
def visualize_tensorboard(
self,
port: int,
height: int,
) -> None:
"""Generate a visualization of a Tensorboard.
Args:
port: the TCP port where the Tensorboard server is listening for
requests.
height: Height of the generated visualization.
logdir: The logdir location for the Tensorboard server.
"""
if Environment.in_notebook():
notebook.display(port, height=height)
return
print(
"You can visit:\n"
f"[italic green] http://localhost:{port}/[/italic green]\n"
"...to visualize the Tensorboard logs for your trained model."
)
get_step(pipeline_name, step_name)
Get the StepView for the specified pipeline and step name.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline_name |
str |
The name of the pipeline. |
required |
step_name |
str |
The name of the step. |
required |
Returns:
| Type | Description |
|---|---|
StepView |
The StepView for the specified pipeline and step name. |
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
def get_step(pipeline_name: str, step_name: str) -> StepView:
"""Get the StepView for the specified pipeline and step name.
Args:
pipeline_name: The name of the pipeline.
step_name: The name of the step.
Returns:
The StepView for the specified pipeline and step name.
"""
repo = Repository()
pipeline = repo.get_pipeline(pipeline_name)
if pipeline is None:
raise RuntimeError(f"No pipeline with name `{pipeline_name}` was found")
last_run = pipeline.runs[-1]
step = last_run.get_step(name=step_name)
if step is None:
raise RuntimeError(
f"No pipeline step with name `{step_name}` was found in "
f"pipeline `{pipeline_name}`"
)
return cast(StepView, step)
stop_tensorboard_server(pipeline_name, step_name)
Stop the Tensorboard server previously started for a pipeline step.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline_name |
str |
the name of the pipeline |
required |
step_name |
str |
pipeline step name |
required |
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
def stop_tensorboard_server(pipeline_name: str, step_name: str) -> None:
"""Stop the Tensorboard server previously started for a pipeline step.
Args:
pipeline_name: the name of the pipeline
step_name: pipeline step name
"""
step = get_step(pipeline_name, step_name)
TensorboardVisualizer().stop(step)
visualize_tensorboard(pipeline_name, step_name)
Start a Tensorboard server to visualize all models logged as output by the named pipeline step. The server will monitor and display all the models logged by past and future step runs.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline_name |
str |
the name of the pipeline |
required |
step_name |
str |
pipeline step name |
required |
Source code in zenml/integrations/tensorflow/visualizers/tensorboard_visualizer.py
def visualize_tensorboard(pipeline_name: str, step_name: str) -> None:
"""Start a Tensorboard server to visualize all models logged as output by
the named pipeline step. The server will monitor and display all the models
logged by past and future step runs.
Args:
pipeline_name: the name of the pipeline
step_name: pipeline step name
"""
step = get_step(pipeline_name, step_name)
TensorboardVisualizer().visualize(step)
utils
get_integration_for_module(module_name)
Gets the integration class for a module inside an integration.
If the module given by module_name is not part of a ZenML integration,
this method will return None. If it is part of a ZenML integration,
it will return the integration class found inside the integration
init file.
Source code in zenml/integrations/utils.py
def get_integration_for_module(module_name: str) -> Optional[Type[Integration]]:
"""Gets the integration class for a module inside an integration.
If the module given by `module_name` is not part of a ZenML integration,
this method will return `None`. If it is part of a ZenML integration,
it will return the integration class found inside the integration
__init__ file.
"""
integration_prefix = "zenml.integrations."
if not module_name.startswith(integration_prefix):
return None
integration_module_name = ".".join(module_name.split(".", 3)[:3])
try:
integration_module = sys.modules[integration_module_name]
except KeyError:
integration_module = importlib.import_module(integration_module_name)
for name, member in inspect.getmembers(integration_module):
if (
member is not Integration
and isinstance(member, IntegrationMeta)
and issubclass(member, Integration)
):
return cast(Type[Integration], member)
return None
get_requirements_for_module(module_name)
Gets requirements for a module inside an integration.
If the module given by module_name is not part of a ZenML integration,
this method will return an empty list. If it is part of a ZenML integration,
it will return the list of requirements specified inside the integration
class found inside the integration init file.
Source code in zenml/integrations/utils.py
def get_requirements_for_module(module_name: str) -> List[str]:
"""Gets requirements for a module inside an integration.
If the module given by `module_name` is not part of a ZenML integration,
this method will return an empty list. If it is part of a ZenML integration,
it will return the list of requirements specified inside the integration
class found inside the integration __init__ file.
"""
integration = get_integration_for_module(module_name)
return integration.REQUIREMENTS if integration else []
vertex
special
The Vertex integration submodule provides a way to run ZenML pipelines in a Vertex AI environment.
VertexIntegration (Integration)
Definition of Vertex AI integration for ZenML.
Source code in zenml/integrations/vertex/__init__.py
class VertexIntegration(Integration):
"""Definition of Vertex AI integration for ZenML."""
NAME = VERTEX
REQUIREMENTS = ["google-cloud-aiplatform>=1.11.0"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.vertex import step_operators # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/vertex/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.vertex import step_operators # noqa
orchestrator
special
vertex_ai_orchestrator
VertexOrchestrator (KubeflowOrchestrator)
pydantic-model
Orchestrator responsible for running pipelines on Vertex AI.
Source code in zenml/integrations/vertex/orchestrator/vertex_ai_orchestrator.py
class VertexOrchestrator(KubeflowOrchestrator):
"""Orchestrator responsible for running pipelines on Vertex AI."""
FLAVOR: ClassVar[str] = "vertex"
def run_pipeline(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> Any:
"""Runs a pipeline on Vertex AI using the Kubeflow orchestrator."""
raise NotImplementedError("Vertex AI orchestration is coming soon!")
# super().run_pipeline(pipeline, stack, runtime_configuration)
# aiplatform.init(
# project=GOOGLE_CLOUD_PROJECT, location=GOOGLE_CLOUD_REGION
# )
# job = pipeline_jobs.PipelineJob(
# template_path=PIPELINE_DEFINITION_FILE, display_name=PIPELINE_NAME
# )
# job.submit()
run_pipeline(self, pipeline, stack, runtime_configuration)
Runs a pipeline on Vertex AI using the Kubeflow orchestrator.
Source code in zenml/integrations/vertex/orchestrator/vertex_ai_orchestrator.py
def run_pipeline(
self,
pipeline: "BasePipeline",
stack: "Stack",
runtime_configuration: "RuntimeConfiguration",
) -> Any:
"""Runs a pipeline on Vertex AI using the Kubeflow orchestrator."""
raise NotImplementedError("Vertex AI orchestration is coming soon!")
# super().run_pipeline(pipeline, stack, runtime_configuration)
# aiplatform.init(
# project=GOOGLE_CLOUD_PROJECT, location=GOOGLE_CLOUD_REGION
# )
# job = pipeline_jobs.PipelineJob(
# template_path=PIPELINE_DEFINITION_FILE, display_name=PIPELINE_NAME
# )
# job.submit()
step_operators
special
vertex_step_operator
Code heavily inspired by TFX Implementation: https://github.com/tensorflow/tfx/blob/master/tfx/extensions/ google_cloud_ai_platform/training_clients.py
VertexStepOperator (BaseStepOperator)
pydantic-model
Step operator to run a step on Vertex AI.
This class defines code that can set up a Vertex AI environment and run the ZenML entrypoint command in it.
Attributes:
| Name | Type | Description |
|---|---|---|
region |
str |
Region name, e.g., |
project |
Optional[str] |
[Optional] GCP project name. If left None, inferred from the environment. |
accelerator_type |
Optional[str] |
[Optional] Accelerator type from list: https://cloud.google.com/vertex-ai/docs/reference/rest/v1/MachineSpec#AcceleratorType |
accelerator_count |
int |
[Optional] Defines number of accelerators to be used for the job. |
machine_type |
str |
[Optional] Machine type specified here: https://cloud.google.com/vertex-ai/docs/training/configure-compute#machine-types |
base_image |
Optional[str] |
[Optional] Base image for building the custom job container. |
encryption_spec_key_name |
Optional[str] |
[Optional]: Encryption spec key name. |
service_account_path |
Optional[str] |
[Optional]: Path to service account file specifying credentials of the GCP user. If not provided, falls back to Default Credentials. |
Source code in zenml/integrations/vertex/step_operators/vertex_step_operator.py
class VertexStepOperator(BaseStepOperator):
"""Step operator to run a step on Vertex AI.
This class defines code that can set up a Vertex AI environment and run the
ZenML entrypoint command in it.
Attributes:
region: Region name, e.g., `europe-west1`.
project: [Optional] GCP project name. If left None, inferred from the
environment.
accelerator_type: [Optional] Accelerator type from list: https://cloud.google.com/vertex-ai/docs/reference/rest/v1/MachineSpec#AcceleratorType
accelerator_count: [Optional] Defines number of accelerators to be
used for the job.
machine_type: [Optional] Machine type specified here: https://cloud.google.com/vertex-ai/docs/training/configure-compute#machine-types
base_image: [Optional] Base image for building the custom job container.
encryption_spec_key_name: [Optional]: Encryption spec key name.
service_account_path: [Optional]: Path to service account file
specifying credentials of the GCP user. If not provided, falls back
to Default Credentials.
"""
region: str
project: Optional[str] = None
accelerator_type: Optional[str] = None
accelerator_count: int = 0
machine_type: str = "n1-standard-4"
base_image: Optional[str] = None
# customer managed encryption key resource name
# will be applied to all Vertex AI resources if set
encryption_spec_key_name: Optional[str] = None
# path to google service account
# environment default credentials used if not set
service_account_path: Optional[str] = None
# Class configuration
FLAVOR: ClassVar[str] = "vertex"
@property
def validator(self) -> Optional[StackValidator]:
"""Validates that the stack contains a container registry."""
def _ensure_local_orchestrator(stack: Stack) -> Tuple[bool, str]:
# For now this only works on local orchestrator and GCP artifact
# store
return (
(
stack.orchestrator.FLAVOR == "local"
and stack.artifact_store.FLAVOR == "gcp"
),
"Only local orchestrator and GCP artifact store are currently "
"supported",
)
return StackValidator(
required_components={StackComponentType.CONTAINER_REGISTRY},
custom_validation_function=_ensure_local_orchestrator,
)
@property_validator("accelerator_type")
def validate_accelerator_enum(cls, accelerator_type: Optional[str]) -> None:
accepted_vals = list(
aiplatform.gapic.AcceleratorType.__members__.keys()
)
if accelerator_type and accelerator_type.upper() not in accepted_vals:
raise RuntimeError(
f"Accelerator must be one of the following: {accepted_vals}"
)
def _get_authentication(
self,
) -> Tuple[Optional[auth_credentials.Credentials], Optional[str]]:
if self.service_account_path:
credentials, project_id = load_credentials_from_file(
self.service_account_path
)
else:
credentials, project_id = default()
return credentials, project_id
def _build_and_push_docker_image(
self,
pipeline_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> str:
repo = Repository()
container_registry = repo.active_stack.container_registry
if not container_registry:
raise RuntimeError("Missing container registry")
registry_uri = container_registry.uri.rstrip("/")
image_name = f"{registry_uri}/zenml-vertex:{pipeline_name}"
docker_utils.build_docker_image(
build_context_path=get_source_root_path(),
image_name=image_name,
entrypoint=" ".join(entrypoint_command),
requirements=set(requirements),
base_image=self.base_image,
)
docker_utils.push_docker_image(image_name)
return docker_utils.get_image_digest(image_name) or image_name
def launch(
self,
pipeline_name: str,
run_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> None:
"""Launches a step on Vertex AI.
Args:
pipeline_name: Name of the pipeline which the step to be executed
is part of.
run_name: Name of the pipeline run which the step to be executed
is part of.
entrypoint_command: Command that executes the step.
requirements: List of pip requirements that must be installed
inside the step operator environment.
"""
job_labels = {"source": f"zenml-{__version__.replace('.', '_')}"}
# Step 1: Authenticate with Google
credentials, project_id = self._get_authentication()
if self.project:
if self.project != project_id:
logger.warning(
f"Authenticated with project {project_id}, but this "
f"operator is configured to use project {self.project}."
)
else:
self.project = project_id
# Step 2: Build and push image
image_name = self._build_and_push_docker_image(
pipeline_name=pipeline_name,
requirements=requirements,
entrypoint_command=entrypoint_command,
)
# Step 3: Launch the job
# The AI Platform services require regional API endpoints.
client_options = {"api_endpoint": self.region + VERTEX_ENDPOINT_SUFFIX}
# Initialize client that will be used to create and send requests.
# This client only needs to be created once, and can be reused for multiple requests.
client = aiplatform.gapic.JobServiceClient(
credentials=credentials, client_options=client_options
)
custom_job = {
"display_name": run_name,
"job_spec": {
"worker_pool_specs": [
{
"machine_spec": {
"machine_type": self.machine_type,
"accelerator_type": self.accelerator_type,
"accelerator_count": self.accelerator_count
if self.accelerator_type
else 0,
},
"replica_count": 1,
"container_spec": {
"image_uri": image_name,
"command": [],
"args": [],
},
}
]
},
"labels": job_labels,
"encryption_spec": {"kmsKeyName": self.encryption_spec_key_name}
if self.encryption_spec_key_name
else {},
}
logger.debug("Vertex AI Job=%s", custom_job)
parent = f"projects/{self.project}/locations/{self.region}"
logger.info(
"Submitting custom job='%s', path='%s' to Vertex AI Training.",
custom_job["display_name"],
parent,
)
response = client.create_custom_job(
parent=parent, custom_job=custom_job
)
logger.debug("Vertex AI response:", response)
# Step 4: Monitor the job
# Monitors the long-running operation by polling the job state
# periodically, and retries the polling when a transient connectivity
# issue is encountered.
#
# Long-running operation monitoring:
# The possible states of "get job" response can be found at
# https://cloud.google.com/ai-platform/training/docs/reference/rest/v1/projects.jobs#State
# where SUCCEEDED/FAILED/CANCELED are considered to be final states.
# The following logic will keep polling the state of the job until
# the job enters a final state.
#
# During the polling, if a connection error was encountered, the GET
# request will be retried by recreating the Python API client to
# refresh the lifecycle of the connection being used. See
# https://github.com/googleapis/google-api-python-client/issues/218
# for a detailed description of the problem. If the error persists for
# _CONNECTION_ERROR_RETRY_LIMIT consecutive attempts, the function
# will raise ConnectionError.
retry_count = 0
job_id = response.name
while response.state not in VERTEX_JOB_STATES_COMPLETED:
time.sleep(POLLING_INTERVAL_IN_SECONDS)
try:
response = client.get_custom_job(name=job_id)
retry_count = 0
# Handle transient connection error.
except ConnectionError as err:
if retry_count < CONNECTION_ERROR_RETRY_LIMIT:
retry_count += 1
logger.warning(
"ConnectionError (%s) encountered when polling job: "
"%s. Trying to recreate the API client.",
err,
job_id,
)
# Recreate the Python API client.
client = aiplatform.gapic.JobServiceClient(
client_options=client_options
)
else:
logger.error(
"Request failed after %s retries.",
CONNECTION_ERROR_RETRY_LIMIT,
)
raise
if response.state in VERTEX_JOB_STATES_FAILED:
err_msg = (
"Job '{}' did not succeed. Detailed response {}.".format(
job_id, response
)
)
logger.error(err_msg)
raise RuntimeError(err_msg)
# Cloud training complete
logger.info("Job '%s' successful.", job_id)
validator: Optional[zenml.stack.stack_validator.StackValidator]
property
readonly
Validates that the stack contains a container registry.
launch(self, pipeline_name, run_name, requirements, entrypoint_command)
Launches a step on Vertex AI.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
pipeline_name |
str |
Name of the pipeline which the step to be executed is part of. |
required |
run_name |
str |
Name of the pipeline run which the step to be executed is part of. |
required |
entrypoint_command |
List[str] |
Command that executes the step. |
required |
requirements |
List[str] |
List of pip requirements that must be installed inside the step operator environment. |
required |
Source code in zenml/integrations/vertex/step_operators/vertex_step_operator.py
def launch(
self,
pipeline_name: str,
run_name: str,
requirements: List[str],
entrypoint_command: List[str],
) -> None:
"""Launches a step on Vertex AI.
Args:
pipeline_name: Name of the pipeline which the step to be executed
is part of.
run_name: Name of the pipeline run which the step to be executed
is part of.
entrypoint_command: Command that executes the step.
requirements: List of pip requirements that must be installed
inside the step operator environment.
"""
job_labels = {"source": f"zenml-{__version__.replace('.', '_')}"}
# Step 1: Authenticate with Google
credentials, project_id = self._get_authentication()
if self.project:
if self.project != project_id:
logger.warning(
f"Authenticated with project {project_id}, but this "
f"operator is configured to use project {self.project}."
)
else:
self.project = project_id
# Step 2: Build and push image
image_name = self._build_and_push_docker_image(
pipeline_name=pipeline_name,
requirements=requirements,
entrypoint_command=entrypoint_command,
)
# Step 3: Launch the job
# The AI Platform services require regional API endpoints.
client_options = {"api_endpoint": self.region + VERTEX_ENDPOINT_SUFFIX}
# Initialize client that will be used to create and send requests.
# This client only needs to be created once, and can be reused for multiple requests.
client = aiplatform.gapic.JobServiceClient(
credentials=credentials, client_options=client_options
)
custom_job = {
"display_name": run_name,
"job_spec": {
"worker_pool_specs": [
{
"machine_spec": {
"machine_type": self.machine_type,
"accelerator_type": self.accelerator_type,
"accelerator_count": self.accelerator_count
if self.accelerator_type
else 0,
},
"replica_count": 1,
"container_spec": {
"image_uri": image_name,
"command": [],
"args": [],
},
}
]
},
"labels": job_labels,
"encryption_spec": {"kmsKeyName": self.encryption_spec_key_name}
if self.encryption_spec_key_name
else {},
}
logger.debug("Vertex AI Job=%s", custom_job)
parent = f"projects/{self.project}/locations/{self.region}"
logger.info(
"Submitting custom job='%s', path='%s' to Vertex AI Training.",
custom_job["display_name"],
parent,
)
response = client.create_custom_job(
parent=parent, custom_job=custom_job
)
logger.debug("Vertex AI response:", response)
# Step 4: Monitor the job
# Monitors the long-running operation by polling the job state
# periodically, and retries the polling when a transient connectivity
# issue is encountered.
#
# Long-running operation monitoring:
# The possible states of "get job" response can be found at
# https://cloud.google.com/ai-platform/training/docs/reference/rest/v1/projects.jobs#State
# where SUCCEEDED/FAILED/CANCELED are considered to be final states.
# The following logic will keep polling the state of the job until
# the job enters a final state.
#
# During the polling, if a connection error was encountered, the GET
# request will be retried by recreating the Python API client to
# refresh the lifecycle of the connection being used. See
# https://github.com/googleapis/google-api-python-client/issues/218
# for a detailed description of the problem. If the error persists for
# _CONNECTION_ERROR_RETRY_LIMIT consecutive attempts, the function
# will raise ConnectionError.
retry_count = 0
job_id = response.name
while response.state not in VERTEX_JOB_STATES_COMPLETED:
time.sleep(POLLING_INTERVAL_IN_SECONDS)
try:
response = client.get_custom_job(name=job_id)
retry_count = 0
# Handle transient connection error.
except ConnectionError as err:
if retry_count < CONNECTION_ERROR_RETRY_LIMIT:
retry_count += 1
logger.warning(
"ConnectionError (%s) encountered when polling job: "
"%s. Trying to recreate the API client.",
err,
job_id,
)
# Recreate the Python API client.
client = aiplatform.gapic.JobServiceClient(
client_options=client_options
)
else:
logger.error(
"Request failed after %s retries.",
CONNECTION_ERROR_RETRY_LIMIT,
)
raise
if response.state in VERTEX_JOB_STATES_FAILED:
err_msg = (
"Job '{}' did not succeed. Detailed response {}.".format(
job_id, response
)
)
logger.error(err_msg)
raise RuntimeError(err_msg)
# Cloud training complete
logger.info("Job '%s' successful.", job_id)
wandb
special
The wandb integrations currently enables you to use wandb tracking as a convenient way to visualize your experiment runs within the wandb ui
WandbIntegration (Integration)
Definition of Plotly integration for ZenML.
Source code in zenml/integrations/wandb/__init__.py
class WandbIntegration(Integration):
"""Definition of Plotly integration for ZenML."""
NAME = WANDB
REQUIREMENTS = ["wandb>=0.12.12", "Pillow>=9.1.0"]
@staticmethod
def activate() -> None:
"""Activate the Wandb integration."""
from zenml.integrations.wandb import experiment_trackers # noqa
activate()
staticmethod
Activate the Wandb integration.
Source code in zenml/integrations/wandb/__init__.py
@staticmethod
def activate() -> None:
"""Activate the Wandb integration."""
from zenml.integrations.wandb import experiment_trackers # noqa
experiment_trackers
special
wandb_experiment_tracker
WandbExperimentTracker (BaseExperimentTracker)
pydantic-model
Stores wandb configuration options.
ZenML should take care of configuring wandb for you, but should you still need access to the configuration inside your step you can do it using a step context:
from zenml.steps import StepContext
@enable_wandb
@step
def my_step(context: StepContext, ...)
context.stack.experiment_tracker # get the tracking_uri etc. from here
Attributes:
| Name | Type | Description |
|---|---|---|
entity |
Optional[str] |
Name of an existing wandb entity. |
project_name |
Optional[str] |
Name of an existing wandb project to log to. |
api_key |
str |
API key to should be authorized to log to the configured wandb entity and project. |
Source code in zenml/integrations/wandb/experiment_trackers/wandb_experiment_tracker.py
class WandbExperimentTracker(BaseExperimentTracker):
"""Stores wandb configuration options.
ZenML should take care of configuring wandb for you, but should you still
need access to the configuration inside your step you can do it using a
step context:
```python
from zenml.steps import StepContext
@enable_wandb
@step
def my_step(context: StepContext, ...)
context.stack.experiment_tracker # get the tracking_uri etc. from here
```
Attributes:
entity: Name of an existing wandb entity.
project_name: Name of an existing wandb project to log to.
api_key: API key to should be authorized to log to the configured wandb
entity and project.
"""
api_key: str
entity: Optional[str] = None
project_name: Optional[str] = None
# Class Configuration
FLAVOR: ClassVar[str] = WANDB
def prepare_step_run(self) -> None:
"""Sets the wandb api key."""
os.environ[WANDB_API_KEY] = self.api_key
@contextmanager
def activate_wandb_run(
self,
run_name: str,
tags: Tuple[str, ...] = (),
settings: Optional[wandb.Settings] = None,
) -> Iterator[None]:
"""Activates a wandb run for the duration of this context manager.
Anything logged to wandb that is run while this context manager is
active will automatically log to the same wandb run configured by the
run name passed as an argument to this function.
Args:
run_name: Name of the wandb run to create.
tags: Tags to attach to the wandb run.
settings: Additional settings for the wandb run.
"""
try:
logger.info(
f"Initializing wandb with project name: {self.project_name}, "
f"run_name: {run_name}, entity: {self.entity}."
)
wandb.init(
project=self.project_name,
name=run_name,
entity=self.entity,
settings=settings,
tags=tags,
)
yield
finally:
wandb.finish()
activate_wandb_run(self, run_name, tags=(), settings=None)
Activates a wandb run for the duration of this context manager.
Anything logged to wandb that is run while this context manager is active will automatically log to the same wandb run configured by the run name passed as an argument to this function.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
run_name |
str |
Name of the wandb run to create. |
required |
tags |
Tuple[str, ...] |
Tags to attach to the wandb run. |
() |
settings |
Optional[wandb.sdk.wandb_settings.Settings] |
Additional settings for the wandb run. |
None |
Source code in zenml/integrations/wandb/experiment_trackers/wandb_experiment_tracker.py
@contextmanager
def activate_wandb_run(
self,
run_name: str,
tags: Tuple[str, ...] = (),
settings: Optional[wandb.Settings] = None,
) -> Iterator[None]:
"""Activates a wandb run for the duration of this context manager.
Anything logged to wandb that is run while this context manager is
active will automatically log to the same wandb run configured by the
run name passed as an argument to this function.
Args:
run_name: Name of the wandb run to create.
tags: Tags to attach to the wandb run.
settings: Additional settings for the wandb run.
"""
try:
logger.info(
f"Initializing wandb with project name: {self.project_name}, "
f"run_name: {run_name}, entity: {self.entity}."
)
wandb.init(
project=self.project_name,
name=run_name,
entity=self.entity,
settings=settings,
tags=tags,
)
yield
finally:
wandb.finish()
prepare_step_run(self)
Sets the wandb api key.
Source code in zenml/integrations/wandb/experiment_trackers/wandb_experiment_tracker.py
def prepare_step_run(self) -> None:
"""Sets the wandb api key."""
os.environ[WANDB_API_KEY] = self.api_key
wandb_step_decorator
enable_wandb(_step=None, *, settings=None)
Decorator to enable wandb for a step function.
Apply this decorator to a ZenML pipeline step to enable wandb experiment
tracking. The wandb tracking configuration (project name, experiment name,
entity) will be automatically configured before the step code is executed,
so the step can simply use the wandb module to log metrics and artifacts,
like so:
@enable_wandb
@step
def tf_evaluator(
x_test: np.ndarray,
y_test: np.ndarray,
model: tf.keras.Model,
) -> float:
_, test_acc = model.evaluate(x_test, y_test, verbose=2)
wandb.log_metric("val_accuracy", test_acc)
return test_acc
All wandb artifacts and metrics logged from all the steps in a pipeline
run are by default grouped under a single experiment named after the
pipeline. To log wandb artifacts and metrics from a step in a separate
wandb experiment, pass a custom experiment_name argument value to the
decorator.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
_step |
Optional[~S] |
The decorated step class. |
None |
project_name |
Name of project. |
required | |
experiment_name |
optional wandb experiment name to use for the step. If not provided, the name of the pipeline in the context of which the step is executed will be used as experiment name. |
required |
Returns:
| Type | Description |
|---|---|
Union[~S, Callable[[~S], ~S]] |
The inner decorator which enhances the input step class with wandb tracking functionality |
Source code in zenml/integrations/wandb/wandb_step_decorator.py
def enable_wandb(
_step: Optional[S] = None, *, settings: Optional[wandb.Settings] = None
) -> Union[S, Callable[[S], S]]:
"""Decorator to enable wandb for a step function.
Apply this decorator to a ZenML pipeline step to enable wandb experiment
tracking. The wandb tracking configuration (project name, experiment name,
entity) will be automatically configured before the step code is executed,
so the step can simply use the `wandb` module to log metrics and artifacts,
like so:
```python
@enable_wandb
@step
def tf_evaluator(
x_test: np.ndarray,
y_test: np.ndarray,
model: tf.keras.Model,
) -> float:
_, test_acc = model.evaluate(x_test, y_test, verbose=2)
wandb.log_metric("val_accuracy", test_acc)
return test_acc
```
All wandb artifacts and metrics logged from all the steps in a pipeline
run are by default grouped under a single experiment named after the
pipeline. To log wandb artifacts and metrics from a step in a separate
wandb experiment, pass a custom `experiment_name` argument value to the
decorator.
Args:
_step: The decorated step class.
project_name: Name of project.
experiment_name: optional wandb experiment name to use for the step.
If not provided, the name of the pipeline in the context of which
the step is executed will be used as experiment name.
Returns:
The inner decorator which enhances the input step class with wandb
tracking functionality
"""
def inner_decorator(_step: S) -> S:
"""Inner decorator for step enable_wandb."""
logger.debug(
"Applying 'enable_wandb' decorator to step %s", _step.__name__
)
if not issubclass(_step, BaseStep):
raise RuntimeError(
"The `enable_wandb` decorator can only be applied to a ZenML "
"`step` decorated function or a BaseStep subclass."
)
source_fn = getattr(_step, STEP_INNER_FUNC_NAME)
return cast(
S,
type( # noqa
_step.__name__,
(_step,),
{
STEP_INNER_FUNC_NAME: staticmethod(
wandb_step_entrypoint(
settings=settings,
)(source_fn)
),
"__module__": _step.__module__,
},
),
)
if _step is None:
return inner_decorator
else:
return inner_decorator(_step)
wandb_step_entrypoint(settings=None)
Decorator for a step entrypoint to enable wandb.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
project_name |
Name of wandb project. |
required | |
experiment_name |
optional wandb experiment name to use for the step. If not provided, the name of the pipeline in the context of which the step is executed will be used as experiment name. |
required |
Returns:
| Type | Description |
|---|---|
Callable[[~F], ~F] |
the input function enhanced with wandb profiling functionality |
Source code in zenml/integrations/wandb/wandb_step_decorator.py
def wandb_step_entrypoint(
settings: Optional[wandb.Settings] = None,
) -> Callable[[F], F]:
"""Decorator for a step entrypoint to enable wandb.
Args:
project_name: Name of wandb project.
experiment_name: optional wandb experiment name to use for the step.
If not provided, the name of the pipeline in the context of which
the step is executed will be used as experiment name.
Returns:
the input function enhanced with wandb profiling functionality
"""
def inner_decorator(func: F) -> F:
"""Inner decorator for step entrypoint."""
logger.debug(
"Applying 'wandb_step_entrypoint' decorator to step entrypoint %s",
func.__name__,
)
@functools.wraps(func)
def wrapper(*args: Any, **kwargs: Any) -> Any: # noqa
"""Wrapper function for decorator"""
logger.debug(
"Setting up wandb backend before running step entrypoint %s",
func.__name__,
)
step_env = Environment().step_environment
run_name = f"{step_env.pipeline_run_id}_{step_env.step_name}"
tags = (step_env.pipeline_name, step_env.pipeline_run_id)
experiment_tracker = Repository( # type: ignore[call-arg]
skip_repository_check=True
).active_stack.experiment_tracker
if not isinstance(experiment_tracker, WandbExperimentTracker):
raise ValueError(
"The active stack needs to have a wandb experiment tracker "
"component registered to be able to track experiments "
"using wandb. You can create a new stack with a wandb "
"experiment tracker component or update your existing "
"stack to add this component, e.g.:\n\n"
" 'zenml experiment-tracker register wandb_tracker "
"--type=wandb --entity=<WANDB_ENTITY> --project_name="
"<WANDB_PROJECT_NAME> --api_key=<WANDB_API_KEY>'\n"
" 'zenml stack register stack-name -e wandb_tracker ...'\n"
)
with experiment_tracker.activate_wandb_run(
run_name=run_name,
tags=tags,
settings=settings,
):
return func(*args, **kwargs)
return cast(F, wrapper)
return inner_decorator
whylogs
special
WhylogsIntegration (Integration)
Definition of whylogs integration for ZenML.
Source code in zenml/integrations/whylogs/__init__.py
class WhylogsIntegration(Integration):
"""Definition of [whylogs](https://github.com/whylabs/whylogs)
integration for ZenML."""
NAME = WHYLOGS
REQUIREMENTS = ["whylogs>=0.6.22", "pybars3>=0.9.7"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.whylogs import materializers # noqa
from zenml.integrations.whylogs import visualizers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/whylogs/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.whylogs import materializers # noqa
from zenml.integrations.whylogs import visualizers # noqa
materializers
special
whylogs_materializer
WhylogsMaterializer (BaseMaterializer)
Materializer to read/write whylogs dataset profiles.
Source code in zenml/integrations/whylogs/materializers/whylogs_materializer.py
class WhylogsMaterializer(BaseMaterializer):
"""Materializer to read/write whylogs dataset profiles."""
ASSOCIATED_TYPES = (DatasetProfile,)
ASSOCIATED_ARTIFACT_TYPES = (StatisticsArtifact,)
def handle_input(self, data_type: Type[Any]) -> DatasetProfile:
"""Reads and returns a whylogs DatasetProfile.
Returns:
A loaded whylogs DatasetProfile.
"""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, PROFILE_FILENAME)
with fileio.open(filepath, "rb") as f:
protobuf = DatasetProfile.parse_delimited(f.read())[0]
return protobuf
def handle_return(self, profile: DatasetProfile) -> None:
"""Writes a whylogs DatasetProfile.
Args:
profile: A DatasetProfile object from whylogs.
"""
super().handle_return(profile)
filepath = os.path.join(self.artifact.uri, PROFILE_FILENAME)
protobuf = profile.serialize_delimited()
with fileio.open(filepath, "wb") as f:
f.write(protobuf)
# TODO [ENG-439]: uploading profiles to whylabs should be enabled and
# configurable at step level or pipeline level instead of being
# globally enabled.
if os.environ.get("WHYLABS_DEFAULT_ORG_ID"):
upload_profile(profile)
handle_input(self, data_type)
Reads and returns a whylogs DatasetProfile.
Returns:
| Type | Description |
|---|---|
DatasetProfile |
A loaded whylogs DatasetProfile. |
Source code in zenml/integrations/whylogs/materializers/whylogs_materializer.py
def handle_input(self, data_type: Type[Any]) -> DatasetProfile:
"""Reads and returns a whylogs DatasetProfile.
Returns:
A loaded whylogs DatasetProfile.
"""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, PROFILE_FILENAME)
with fileio.open(filepath, "rb") as f:
protobuf = DatasetProfile.parse_delimited(f.read())[0]
return protobuf
handle_return(self, profile)
Writes a whylogs DatasetProfile.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
profile |
DatasetProfile |
A DatasetProfile object from whylogs. |
required |
Source code in zenml/integrations/whylogs/materializers/whylogs_materializer.py
def handle_return(self, profile: DatasetProfile) -> None:
"""Writes a whylogs DatasetProfile.
Args:
profile: A DatasetProfile object from whylogs.
"""
super().handle_return(profile)
filepath = os.path.join(self.artifact.uri, PROFILE_FILENAME)
protobuf = profile.serialize_delimited()
with fileio.open(filepath, "wb") as f:
f.write(protobuf)
# TODO [ENG-439]: uploading profiles to whylabs should be enabled and
# configurable at step level or pipeline level instead of being
# globally enabled.
if os.environ.get("WHYLABS_DEFAULT_ORG_ID"):
upload_profile(profile)
steps
special
whylogs_profiler
WhylogsProfilerConfig (BaseAnalyzerConfig)
pydantic-model
Config class for the WhylogsProfiler step.
Attributes:
| Name | Type | Description |
|---|---|---|
dataset_name |
Optional[str] |
the name of the dataset (Optional). If not specified, the pipeline step name is used |
dataset_timestamp |
Optional[datetime.datetime] |
timestamp to associate with the generated dataset profile (Optional). The current time is used if not supplied. |
tags |
Optional[Dict[str, str]] |
custom metadata tags associated with the whylogs profile |
Also see WhylogsContext.log_dataframe.
Source code in zenml/integrations/whylogs/steps/whylogs_profiler.py
class WhylogsProfilerConfig(BaseAnalyzerConfig):
"""Config class for the WhylogsProfiler step.
Attributes:
dataset_name: the name of the dataset (Optional). If not specified,
the pipeline step name is used
dataset_timestamp: timestamp to associate with the generated
dataset profile (Optional). The current time is used if not
supplied.
tags: custom metadata tags associated with the whylogs profile
Also see `WhylogsContext.log_dataframe`.
"""
dataset_name: Optional[str] = None
dataset_timestamp: Optional[datetime.datetime]
tags: Optional[Dict[str, str]] = None
WhylogsProfilerStep (BaseAnalyzerStep)
Simple step implementation which generates a whylogs data profile from a a given pd.DataFrame
Source code in zenml/integrations/whylogs/steps/whylogs_profiler.py
class WhylogsProfilerStep(BaseAnalyzerStep):
"""Simple step implementation which generates a whylogs data profile from a
a given pd.DataFrame"""
def entrypoint( # type: ignore[override]
self,
dataset: pd.DataFrame,
config: WhylogsProfilerConfig,
context: StepContext,
) -> DatasetProfile:
"""Main entrypoint function for the whylogs profiler
Args:
dataset: pd.DataFrame, the given dataset
config: the configuration of the step
context: the context of the step
Returns:
whylogs profile with statistics generated for the input dataset
"""
whylogs_context = WhylogsContext(context)
profile = whylogs_context.profile_dataframe(
dataset, dataset_name=config.dataset_name, tags=config.tags
)
return profile
CONFIG_CLASS (BaseAnalyzerConfig)
pydantic-model
Config class for the WhylogsProfiler step.
Attributes:
| Name | Type | Description |
|---|---|---|
dataset_name |
Optional[str] |
the name of the dataset (Optional). If not specified, the pipeline step name is used |
dataset_timestamp |
Optional[datetime.datetime] |
timestamp to associate with the generated dataset profile (Optional). The current time is used if not supplied. |
tags |
Optional[Dict[str, str]] |
custom metadata tags associated with the whylogs profile |
Also see WhylogsContext.log_dataframe.
Source code in zenml/integrations/whylogs/steps/whylogs_profiler.py
class WhylogsProfilerConfig(BaseAnalyzerConfig):
"""Config class for the WhylogsProfiler step.
Attributes:
dataset_name: the name of the dataset (Optional). If not specified,
the pipeline step name is used
dataset_timestamp: timestamp to associate with the generated
dataset profile (Optional). The current time is used if not
supplied.
tags: custom metadata tags associated with the whylogs profile
Also see `WhylogsContext.log_dataframe`.
"""
dataset_name: Optional[str] = None
dataset_timestamp: Optional[datetime.datetime]
tags: Optional[Dict[str, str]] = None
entrypoint(self, dataset, config, context)
Main entrypoint function for the whylogs profiler
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
dataset |
DataFrame |
pd.DataFrame, the given dataset |
required |
config |
WhylogsProfilerConfig |
the configuration of the step |
required |
context |
StepContext |
the context of the step |
required |
Returns:
| Type | Description |
|---|---|
DatasetProfile |
whylogs profile with statistics generated for the input dataset |
Source code in zenml/integrations/whylogs/steps/whylogs_profiler.py
def entrypoint( # type: ignore[override]
self,
dataset: pd.DataFrame,
config: WhylogsProfilerConfig,
context: StepContext,
) -> DatasetProfile:
"""Main entrypoint function for the whylogs profiler
Args:
dataset: pd.DataFrame, the given dataset
config: the configuration of the step
context: the context of the step
Returns:
whylogs profile with statistics generated for the input dataset
"""
whylogs_context = WhylogsContext(context)
profile = whylogs_context.profile_dataframe(
dataset, dataset_name=config.dataset_name, tags=config.tags
)
return profile
whylogs_profiler_step(step_name, enable_cache=None, dataset_name=None, dataset_timestamp=None, tags=None)
Shortcut function to create a new instance of the WhylogsProfilerStep step.
The returned WhylogsProfilerStep can be used in a pipeline to generate a whylogs DatasetProfile from a given pd.DataFrame and save it as an artifact.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
step_name |
str |
The name of the step |
required |
enable_cache |
Optional[bool] |
Specify whether caching is enabled for this step. If no value is passed, caching is enabled by default |
None |
dataset_name |
Optional[str] |
the dataset name to be used for the whylogs profile (Optional). If not specified, the step name is used |
None |
dataset_timestamp |
Optional[datetime.datetime] |
timestamp to associate with the generated dataset profile (Optional). The current time is used if not supplied. |
None |
tags |
Optional[Dict[str, str]] |
custom metadata tags associated with the whylogs profile |
None |
Returns:
| Type | Description |
|---|---|
WhylogsProfilerStep |
a WhylogsProfilerStep step instance |
Source code in zenml/integrations/whylogs/steps/whylogs_profiler.py
def whylogs_profiler_step(
step_name: str,
enable_cache: Optional[bool] = None,
dataset_name: Optional[str] = None,
dataset_timestamp: Optional[datetime.datetime] = None,
tags: Optional[Dict[str, str]] = None,
) -> WhylogsProfilerStep:
"""Shortcut function to create a new instance of the WhylogsProfilerStep step.
The returned WhylogsProfilerStep can be used in a pipeline to generate a
whylogs DatasetProfile from a given pd.DataFrame and save it as an artifact.
Args:
step_name: The name of the step
enable_cache: Specify whether caching is enabled for this step. If no
value is passed, caching is enabled by default
dataset_name: the dataset name to be used for the whylogs profile
(Optional). If not specified, the step name is used
dataset_timestamp: timestamp to associate with the generated
dataset profile (Optional). The current time is used if not
supplied.
tags: custom metadata tags associated with the whylogs profile
Returns:
a WhylogsProfilerStep step instance
"""
# enable cache explicitly to compensate for the fact that this step
# takes in a context object
if enable_cache is None:
enable_cache = True
step_type = type(
step_name,
(WhylogsProfilerStep,),
{
INSTANCE_CONFIGURATION: {
PARAM_ENABLE_CACHE: enable_cache,
PARAM_CREATED_BY_FUNCTIONAL_API: True,
},
},
)
return cast(
WhylogsProfilerStep,
step_type(
WhylogsProfilerConfig(
dataset_name=dataset_name,
dataset_timestamp=dataset_timestamp,
tags=tags,
)
),
)
visualizers
special
whylogs_visualizer
WhylogsPlots (StrEnum)
All supported whylogs plot types.
Source code in zenml/integrations/whylogs/visualizers/whylogs_visualizer.py
class WhylogsPlots(StrEnum):
"""All supported whylogs plot types."""
DISTRIBUTION = "plot_distribution"
MISSING_VALUES = "plot_missing_values"
UNIQUENESS = "plot_uniqueness"
DATA_TYPES = "plot_data_types"
STRING_LENGTH = "plot_string_length"
TOKEN_LENGTH = "plot_token_length"
CHAR_POS = "plot_char_pos"
STRING = "plot_string"
WhylogsVisualizer (BaseStepVisualizer)
The implementation of a Whylogs Visualizer.
Source code in zenml/integrations/whylogs/visualizers/whylogs_visualizer.py
class WhylogsVisualizer(BaseStepVisualizer):
"""The implementation of a Whylogs Visualizer."""
def visualize(
self,
object: StepView,
*args: Any,
plots: Optional[List[WhylogsPlots]] = None,
**kwargs: Any,
) -> None:
"""Visualize all whylogs dataset profiles present as outputs in the
step view
Args:
object: StepView fetched from run.get_step().
plots: optional list of whylogs plots to visualize. Defaults to
using all available plot types if not set
"""
whylogs_artifact_datatype = (
f"{DatasetProfile.__module__}.{DatasetProfile.__name__}"
)
for artifact_name, artifact_view in object.outputs.items():
# filter out anything but whylog dataset profile artifacts
if artifact_view.data_type == whylogs_artifact_datatype:
profile = artifact_view.read()
# whylogs doesn't currently support visualizing multiple
# non-related profiles side-by-side, so we open them in
# separate viewers for now
self.visualize_profile(artifact_name, profile, plots)
@staticmethod
def _get_plot_method(
visualizer: ProfileVisualizer, plot: WhylogsPlots
) -> Any:
"""Get the Whylogs ProfileVisualizer plot method corresponding to a
WhylogsPlots enum value.
Args:
visualizer: a ProfileVisualizer instance
plot: a WhylogsPlots enum value
Raises:
ValueError: if the supplied WhylogsPlots enum value does not
correspond to a valid ProfileVisualizer plot method
Returns:
The ProfileVisualizer plot method corresponding to the input
WhylogsPlots enum value
"""
plot_method = getattr(visualizer, plot, None)
if plot_method is None:
nl = "\n"
raise ValueError(
f"Invalid whylogs plot type: {plot} \n\n"
f"Valid and supported options are: {nl}- "
f'{f"{nl}- ".join(WhylogsPlots.names())}'
)
return plot_method
def visualize_profile(
self,
name: str,
profile: DatasetProfile,
plots: Optional[List[WhylogsPlots]] = None,
) -> None:
"""Generate a visualization of a whylogs dataset profile.
Args:
name: name identifying the profile if multiple profiles are
displayed at the same time
profile: whylogs DatasetProfile to visualize
plots: optional list of whylogs plots to visualize. Defaults to
using all available plot types if not set
"""
if Environment.in_notebook():
from IPython.core.display import display
if not plots:
# default to using all plots if none are supplied
plots = list(WhylogsPlots)
for column in sorted(profile.columns):
for plot in plots:
visualizer = ProfileVisualizer()
visualizer.set_profiles([profile])
plot_method = self._get_plot_method(visualizer, plot)
display(plot_method(column))
else:
logger.warning(
"The magic functions are only usable in a Jupyter notebook."
)
with tempfile.NamedTemporaryFile(
delete=False, suffix=f"-{name}.html"
) as f:
logger.info("Opening %s in a new browser.." % f.name)
profile_viewer([profile], output_path=f.name)
visualize(self, object, *args, *, plots=None, **kwargs)
Visualize all whylogs dataset profiles present as outputs in the step view
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
object |
StepView |
StepView fetched from run.get_step(). |
required |
plots |
Optional[List[zenml.integrations.whylogs.visualizers.whylogs_visualizer.WhylogsPlots]] |
optional list of whylogs plots to visualize. Defaults to using all available plot types if not set |
None |
Source code in zenml/integrations/whylogs/visualizers/whylogs_visualizer.py
def visualize(
self,
object: StepView,
*args: Any,
plots: Optional[List[WhylogsPlots]] = None,
**kwargs: Any,
) -> None:
"""Visualize all whylogs dataset profiles present as outputs in the
step view
Args:
object: StepView fetched from run.get_step().
plots: optional list of whylogs plots to visualize. Defaults to
using all available plot types if not set
"""
whylogs_artifact_datatype = (
f"{DatasetProfile.__module__}.{DatasetProfile.__name__}"
)
for artifact_name, artifact_view in object.outputs.items():
# filter out anything but whylog dataset profile artifacts
if artifact_view.data_type == whylogs_artifact_datatype:
profile = artifact_view.read()
# whylogs doesn't currently support visualizing multiple
# non-related profiles side-by-side, so we open them in
# separate viewers for now
self.visualize_profile(artifact_name, profile, plots)
visualize_profile(self, name, profile, plots=None)
Generate a visualization of a whylogs dataset profile.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name |
str |
name identifying the profile if multiple profiles are displayed at the same time |
required |
profile |
DatasetProfile |
whylogs DatasetProfile to visualize |
required |
plots |
Optional[List[zenml.integrations.whylogs.visualizers.whylogs_visualizer.WhylogsPlots]] |
optional list of whylogs plots to visualize. Defaults to using all available plot types if not set |
None |
Source code in zenml/integrations/whylogs/visualizers/whylogs_visualizer.py
def visualize_profile(
self,
name: str,
profile: DatasetProfile,
plots: Optional[List[WhylogsPlots]] = None,
) -> None:
"""Generate a visualization of a whylogs dataset profile.
Args:
name: name identifying the profile if multiple profiles are
displayed at the same time
profile: whylogs DatasetProfile to visualize
plots: optional list of whylogs plots to visualize. Defaults to
using all available plot types if not set
"""
if Environment.in_notebook():
from IPython.core.display import display
if not plots:
# default to using all plots if none are supplied
plots = list(WhylogsPlots)
for column in sorted(profile.columns):
for plot in plots:
visualizer = ProfileVisualizer()
visualizer.set_profiles([profile])
plot_method = self._get_plot_method(visualizer, plot)
display(plot_method(column))
else:
logger.warning(
"The magic functions are only usable in a Jupyter notebook."
)
with tempfile.NamedTemporaryFile(
delete=False, suffix=f"-{name}.html"
) as f:
logger.info("Opening %s in a new browser.." % f.name)
profile_viewer([profile], output_path=f.name)
whylogs_context
WhylogsContext
This is a step context extension that can be used to facilitate whylogs data logging and profiling inside a step function.
It acts as a wrapper built around the whylogs API that transparently incorporates ZenML specific information into the generated whylogs dataset profiles that can be used to associate whylogs profiles with the corresponding ZenML step run that produces them.
It also simplifies the whylogs profile generation process by abstracting away some of the whylogs specific details, such as whylogs session and logger initialization and management.
Source code in zenml/integrations/whylogs/whylogs_context.py
class WhylogsContext:
"""This is a step context extension that can be used to facilitate whylogs
data logging and profiling inside a step function.
It acts as a wrapper built around the whylogs API that transparently
incorporates ZenML specific information into the generated whylogs dataset
profiles that can be used to associate whylogs profiles with the
corresponding ZenML step run that produces them.
It also simplifies the whylogs profile generation process by abstracting
away some of the whylogs specific details, such as whylogs session and
logger initialization and management.
"""
_session: Session = None
def __init__(
self,
step_context: StepContext,
project: Optional[str] = None,
pipeline: Optional[str] = None,
tags: Optional[Dict[str, str]] = None,
) -> None:
"""Create a ZenML whylogs context based on a generic step context.
Args:
step_context: a StepContext instance that provides information
about the currently running step, such as the step name
project: optional project name to use for the whylogs session
pipeline: optional pipeline name to use for the whylogs session
tags: optional list of tags to apply to all whylogs profiles
generated through this context
"""
self._step_context = step_context
self._project = project
self._pipeline = pipeline
self._tags = tags
def get_whylogs_session(
self,
) -> Session:
"""Get the whylogs session associated with the current step.
Returns:
The whylogs Session instance associated with the current step
"""
if self._session is not None:
return self._session
self._session = Session(
project=self._project or self._step_context.step_name,
pipeline=self._pipeline or self._step_context.step_name,
# keeping the writers list empty, serialization is done in the
# materializer
writers=[],
)
return self._session
def profile_dataframe(
self,
df: pd.DataFrame,
dataset_name: Optional[str] = None,
dataset_timestamp: Optional[datetime.datetime] = None,
tags: Optional[Dict[str, str]] = None,
) -> DatasetProfile:
"""Generate whylogs statistics for a Pandas dataframe.
Args:
df: a Pandas dataframe to profile.
dataset_name: the name of the dataset (Optional). If not specified,
the pipeline step name is used
dataset_timestamp: timestamp to associate with the generated
dataset profile (Optional). The current time is used if not
supplied.
tags: custom metadata tags associated with the whylogs profile
Returns:
A whylogs DatasetProfile with the statistics generated from the
input dataset.
"""
session = self.get_whylogs_session()
# TODO [ENG-437]: use a default whylogs dataset_name that is unique across
# multiple pipelines
dataset_name = dataset_name or self._step_context.step_name
final_tags = self._tags.copy() if self._tags else dict()
# TODO [ENG-438]: add more zenml specific tags to the whylogs profile, such
# as the pipeline name and run ID
final_tags["zenml.step"] = self._step_context.step_name
# the datasetId tag is used to identify dataset profiles in whylabs.
# dataset profiles with the same datasetID are considered to belong
# to the same dataset/model.
final_tags.setdefault("datasetId", dataset_name)
if tags:
final_tags.update(tags)
logger = session.logger(
dataset_name, dataset_timestamp=dataset_timestamp, tags=final_tags
)
logger.log_dataframe(df)
return logger.close()
__init__(self, step_context, project=None, pipeline=None, tags=None)
special
Create a ZenML whylogs context based on a generic step context.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
step_context |
StepContext |
a StepContext instance that provides information about the currently running step, such as the step name |
required |
project |
Optional[str] |
optional project name to use for the whylogs session |
None |
pipeline |
Optional[str] |
optional pipeline name to use for the whylogs session |
None |
tags |
Optional[Dict[str, str]] |
optional list of tags to apply to all whylogs profiles generated through this context |
None |
Source code in zenml/integrations/whylogs/whylogs_context.py
def __init__(
self,
step_context: StepContext,
project: Optional[str] = None,
pipeline: Optional[str] = None,
tags: Optional[Dict[str, str]] = None,
) -> None:
"""Create a ZenML whylogs context based on a generic step context.
Args:
step_context: a StepContext instance that provides information
about the currently running step, such as the step name
project: optional project name to use for the whylogs session
pipeline: optional pipeline name to use for the whylogs session
tags: optional list of tags to apply to all whylogs profiles
generated through this context
"""
self._step_context = step_context
self._project = project
self._pipeline = pipeline
self._tags = tags
get_whylogs_session(self)
Get the whylogs session associated with the current step.
Returns:
| Type | Description |
|---|---|
Session |
The whylogs Session instance associated with the current step |
Source code in zenml/integrations/whylogs/whylogs_context.py
def get_whylogs_session(
self,
) -> Session:
"""Get the whylogs session associated with the current step.
Returns:
The whylogs Session instance associated with the current step
"""
if self._session is not None:
return self._session
self._session = Session(
project=self._project or self._step_context.step_name,
pipeline=self._pipeline or self._step_context.step_name,
# keeping the writers list empty, serialization is done in the
# materializer
writers=[],
)
return self._session
profile_dataframe(self, df, dataset_name=None, dataset_timestamp=None, tags=None)
Generate whylogs statistics for a Pandas dataframe.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
df |
DataFrame |
a Pandas dataframe to profile. |
required |
dataset_name |
Optional[str] |
the name of the dataset (Optional). If not specified, the pipeline step name is used |
None |
dataset_timestamp |
Optional[datetime.datetime] |
timestamp to associate with the generated dataset profile (Optional). The current time is used if not supplied. |
None |
tags |
Optional[Dict[str, str]] |
custom metadata tags associated with the whylogs profile |
None |
Returns:
| Type | Description |
|---|---|
DatasetProfile |
A whylogs DatasetProfile with the statistics generated from the input dataset. |
Source code in zenml/integrations/whylogs/whylogs_context.py
def profile_dataframe(
self,
df: pd.DataFrame,
dataset_name: Optional[str] = None,
dataset_timestamp: Optional[datetime.datetime] = None,
tags: Optional[Dict[str, str]] = None,
) -> DatasetProfile:
"""Generate whylogs statistics for a Pandas dataframe.
Args:
df: a Pandas dataframe to profile.
dataset_name: the name of the dataset (Optional). If not specified,
the pipeline step name is used
dataset_timestamp: timestamp to associate with the generated
dataset profile (Optional). The current time is used if not
supplied.
tags: custom metadata tags associated with the whylogs profile
Returns:
A whylogs DatasetProfile with the statistics generated from the
input dataset.
"""
session = self.get_whylogs_session()
# TODO [ENG-437]: use a default whylogs dataset_name that is unique across
# multiple pipelines
dataset_name = dataset_name or self._step_context.step_name
final_tags = self._tags.copy() if self._tags else dict()
# TODO [ENG-438]: add more zenml specific tags to the whylogs profile, such
# as the pipeline name and run ID
final_tags["zenml.step"] = self._step_context.step_name
# the datasetId tag is used to identify dataset profiles in whylabs.
# dataset profiles with the same datasetID are considered to belong
# to the same dataset/model.
final_tags.setdefault("datasetId", dataset_name)
if tags:
final_tags.update(tags)
logger = session.logger(
dataset_name, dataset_timestamp=dataset_timestamp, tags=final_tags
)
logger.log_dataframe(df)
return logger.close()
whylogs_step_decorator
enable_whylogs(_step=None, *, project=None, pipeline=None, tags=None)
Decorator to enable whylogs profiling for a step function.
Apply this decorator to a ZenML pipeline step to enable whylogs profiling.
The decorated function will be given access to a StepContext whylogs
field that facilitates access to the whylogs dataset profiling API,
like so:
@enable_whylogs
@step(enable_cache=True)
def data_loader(
context: StepContext,
) -> Output(data=pd.DataFrame, profile=DatasetProfile,):
...
data = pd.DataFrame(...)
profile = context.whylogs.profile_dataframe(data, dataset_name="input_data")
...
return data, profile
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
_step |
Optional[~S] |
The decorated step class. |
None |
project |
Optional[str] |
optional project name to use for the whylogs session |
None |
pipeline |
Optional[str] |
optional pipeline name to use for the whylogs session |
None |
tags |
Optional[Dict[str, str]] |
optional list of tags to apply to all profiles generated by this step |
None |
Returns:
| Type | Description |
|---|---|
Union[~S, Callable[[~S], ~S]] |
the inner decorator which enhaces the input step class with whylogs profiling functionality |
Source code in zenml/integrations/whylogs/whylogs_step_decorator.py
def enable_whylogs(
_step: Optional[S] = None,
*,
project: Optional[str] = None,
pipeline: Optional[str] = None,
tags: Optional[Dict[str, str]] = None,
) -> Union[S, Callable[[S], S]]:
"""Decorator to enable whylogs profiling for a step function.
Apply this decorator to a ZenML pipeline step to enable whylogs profiling.
The decorated function will be given access to a StepContext `whylogs`
field that facilitates access to the whylogs dataset profiling API,
like so:
```python
@enable_whylogs
@step(enable_cache=True)
def data_loader(
context: StepContext,
) -> Output(data=pd.DataFrame, profile=DatasetProfile,):
...
data = pd.DataFrame(...)
profile = context.whylogs.profile_dataframe(data, dataset_name="input_data")
...
return data, profile
```
Args:
_step: The decorated step class.
project: optional project name to use for the whylogs session
pipeline: optional pipeline name to use for the whylogs session
tags: optional list of tags to apply to all profiles generated by this
step
Returns:
the inner decorator which enhaces the input step class with whylogs
profiling functionality
"""
def inner_decorator(_step: S) -> S:
source_fn = _step.entrypoint
return cast(
S,
type( # noqa
_step.__name__,
(_step,),
{
STEP_INNER_FUNC_NAME: staticmethod(
whylogs_entrypoint(project, pipeline, tags)(source_fn)
),
"__module__": _step.__module__,
},
),
)
if _step is None:
return inner_decorator
else:
return inner_decorator(_step)
whylogs_entrypoint(project=None, pipeline=None, tags=None)
Decorator for a step entrypoint to enable whylogs.
Apply this decorator to a ZenML pipeline step to enable whylogs profiling.
The decorated function will be given access to a StepContext whylogs
field that facilitates access to the whylogs dataset profiling API,
like so:
.. highlight:: python .. code-block:: python
@step(enable_cache=True)
@whylogs_entrypoint()
def data_loader(
context: StepContext,
) -> Output(data=pd.DataFrame, profile=DatasetProfile,):
...
data = pd.DataFrame(...)
profile = context.whylogs.profile_dataframe(data, dataset_name="input_data")
...
return data, profile
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
project |
Optional[str] |
optional project name to use for the whylogs session |
None |
pipeline |
Optional[str] |
optional pipeline name to use for the whylogs session |
None |
tags |
Optional[Dict[str, str]] |
optional list of tags to apply to all profiles generated by this step |
None |
Returns:
| Type | Description |
|---|---|
Callable[[~F], ~F] |
the input function enhanced with whylogs profiling functionality |
Source code in zenml/integrations/whylogs/whylogs_step_decorator.py
def whylogs_entrypoint(
project: Optional[str] = None,
pipeline: Optional[str] = None,
tags: Optional[Dict[str, str]] = None,
) -> Callable[[F], F]:
"""Decorator for a step entrypoint to enable whylogs.
Apply this decorator to a ZenML pipeline step to enable whylogs profiling.
The decorated function will be given access to a StepContext `whylogs`
field that facilitates access to the whylogs dataset profiling API,
like so:
.. highlight:: python
.. code-block:: python
@step(enable_cache=True)
@whylogs_entrypoint()
def data_loader(
context: StepContext,
) -> Output(data=pd.DataFrame, profile=DatasetProfile,):
...
data = pd.DataFrame(...)
profile = context.whylogs.profile_dataframe(data, dataset_name="input_data")
...
return data, profile
Args:
project: optional project name to use for the whylogs session
pipeline: optional pipeline name to use for the whylogs session
tags: optional list of tags to apply to all profiles generated by this
step
Returns:
the input function enhanced with whylogs profiling functionality
"""
def inner_decorator(func: F) -> F:
@functools.wraps(func)
def wrapper(*args: Any, **kwargs: Any) -> Any: # noqa
for arg in args + tuple(kwargs.values()):
if isinstance(arg, StepContext):
arg.__dict__["whylogs"] = WhylogsContext(
arg, project, pipeline, tags
)
break
return func(*args, **kwargs)
return cast(F, wrapper)
return inner_decorator
xgboost
special
XgboostIntegration (Integration)
Definition of xgboost integration for ZenML.
Source code in zenml/integrations/xgboost/__init__.py
class XgboostIntegration(Integration):
"""Definition of xgboost integration for ZenML."""
NAME = XGBOOST
REQUIREMENTS = ["xgboost>=1.0.0"]
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.xgboost import materializers # noqa
activate()
classmethod
Activates the integration.
Source code in zenml/integrations/xgboost/__init__.py
@classmethod
def activate(cls) -> None:
"""Activates the integration."""
from zenml.integrations.xgboost import materializers # noqa
materializers
special
xgboost_booster_materializer
XgboostBoosterMaterializer (BaseMaterializer)
Materializer to read data to and from xgboost.Booster.
Source code in zenml/integrations/xgboost/materializers/xgboost_booster_materializer.py
class XgboostBoosterMaterializer(BaseMaterializer):
"""Materializer to read data to and from xgboost.Booster."""
ASSOCIATED_TYPES = (xgb.Booster,)
ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)
def handle_input(self, data_type: Type[Any]) -> xgb.Booster:
"""Reads a xgboost Booster model from a serialized JSON file."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
booster = xgb.Booster()
booster.load_model(temp_file)
# Cleanup and return
fileio.rmtree(temp_dir)
return booster
def handle_return(self, booster: xgb.Booster) -> None:
"""Creates a JSON serialization for a xgboost Booster model.
Args:
booster: A xgboost Booster model.
"""
super().handle_return(booster)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
with tempfile.NamedTemporaryFile(
mode="w", suffix=".json", delete=True
) as f:
booster.save_model(f.name)
# Copy it into artifact store
fileio.copy(f.name, filepath)
handle_input(self, data_type)
Reads a xgboost Booster model from a serialized JSON file.
Source code in zenml/integrations/xgboost/materializers/xgboost_booster_materializer.py
def handle_input(self, data_type: Type[Any]) -> xgb.Booster:
"""Reads a xgboost Booster model from a serialized JSON file."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
booster = xgb.Booster()
booster.load_model(temp_file)
# Cleanup and return
fileio.rmtree(temp_dir)
return booster
handle_return(self, booster)
Creates a JSON serialization for a xgboost Booster model.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
booster |
Booster |
A xgboost Booster model. |
required |
Source code in zenml/integrations/xgboost/materializers/xgboost_booster_materializer.py
def handle_return(self, booster: xgb.Booster) -> None:
"""Creates a JSON serialization for a xgboost Booster model.
Args:
booster: A xgboost Booster model.
"""
super().handle_return(booster)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
with tempfile.NamedTemporaryFile(
mode="w", suffix=".json", delete=True
) as f:
booster.save_model(f.name)
# Copy it into artifact store
fileio.copy(f.name, filepath)
xgboost_dmatrix_materializer
XgboostDMatrixMaterializer (BaseMaterializer)
Materializer to read data to and from xgboost.DMatrix
Source code in zenml/integrations/xgboost/materializers/xgboost_dmatrix_materializer.py
class XgboostDMatrixMaterializer(BaseMaterializer):
"""Materializer to read data to and from xgboost.DMatrix"""
ASSOCIATED_TYPES = (xgb.DMatrix,)
ASSOCIATED_ARTIFACT_TYPES = (DataArtifact,)
def handle_input(self, data_type: Type[Any]) -> xgb.DMatrix:
"""Reads a xgboost.DMatrix binary file and loads it."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
matrix = xgb.DMatrix(temp_file)
# Cleanup and return
fileio.rmtree(temp_dir)
return matrix
def handle_return(self, matrix: xgb.DMatrix) -> None:
"""Creates a binary serialization for a xgboost.DMatrix object.
Args:
matrix: A xgboost.DMatrix object.
"""
super().handle_return(matrix)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
with tempfile.NamedTemporaryFile(mode="wb", delete=True) as f:
matrix.save_binary(f.name)
# Copy it into artifact store
fileio.copy(f.name, filepath)
handle_input(self, data_type)
Reads a xgboost.DMatrix binary file and loads it.
Source code in zenml/integrations/xgboost/materializers/xgboost_dmatrix_materializer.py
def handle_input(self, data_type: Type[Any]) -> xgb.DMatrix:
"""Reads a xgboost.DMatrix binary file and loads it."""
super().handle_input(data_type)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Create a temporary folder
temp_dir = tempfile.mkdtemp(prefix="zenml-temp-")
temp_file = os.path.join(str(temp_dir), DEFAULT_FILENAME)
# Copy from artifact store to temporary file
fileio.copy(filepath, temp_file)
matrix = xgb.DMatrix(temp_file)
# Cleanup and return
fileio.rmtree(temp_dir)
return matrix
handle_return(self, matrix)
Creates a binary serialization for a xgboost.DMatrix object.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
matrix |
DMatrix |
A xgboost.DMatrix object. |
required |
Source code in zenml/integrations/xgboost/materializers/xgboost_dmatrix_materializer.py
def handle_return(self, matrix: xgb.DMatrix) -> None:
"""Creates a binary serialization for a xgboost.DMatrix object.
Args:
matrix: A xgboost.DMatrix object.
"""
super().handle_return(matrix)
filepath = os.path.join(self.artifact.uri, DEFAULT_FILENAME)
# Make a temporary phantom artifact
with tempfile.NamedTemporaryFile(mode="wb", delete=True) as f:
matrix.save_binary(f.name)
# Copy it into artifact store
fileio.copy(f.name, filepath)