hc99's picture
Add files using upload-large-folder tool
476455e verified
# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License"). You
# may not use this file except in compliance with the License. A copy of
# the License is located at
#
# http://aws.amazon.com/apache2.0/
#
# or in the "license" file accompanying this file. This file is
# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
# ANY KIND, either express or implied. See the License for the specific
# language governing permissions and limitations under the License.
"""The Pipeline entity for workflow."""
from __future__ import absolute_import
import json
import logging
from copy import deepcopy
from typing import Any, Dict, List, Set, Sequence, Union, Optional
import attr
import botocore
from botocore.exceptions import ClientError
from sagemaker import s3
from sagemaker._studio import _append_project_tags
from sagemaker.session import Session
from sagemaker.workflow.callback_step import CallbackOutput, CallbackStep
from sagemaker.workflow.lambda_step import LambdaOutput, LambdaStep
from sagemaker.workflow.entities import (
Entity,
Expression,
RequestType,
)
from sagemaker.workflow.execution_variables import ExecutionVariables
from sagemaker.workflow.parameters import Parameter
from sagemaker.workflow.pipeline_experiment_config import PipelineExperimentConfig
from sagemaker.workflow.parallelism_config import ParallelismConfiguration
from sagemaker.workflow.properties import Properties
from sagemaker.workflow.steps import Step, StepTypeEnum
from sagemaker.workflow.step_collections import StepCollection
from sagemaker.workflow.condition_step import ConditionStep
from sagemaker.workflow.utilities import list_to_request
logger = logging.getLogger(__name__)
_DEFAULT_EXPERIMENT_CFG = PipelineExperimentConfig(
ExecutionVariables.PIPELINE_NAME, ExecutionVariables.PIPELINE_EXECUTION_ID
)
class Pipeline(Entity):
"""Pipeline for workflow."""
def __init__(
self,
name: str = "",
parameters: Optional[Sequence[Parameter]] = None,
pipeline_experiment_config: Optional[PipelineExperimentConfig] = _DEFAULT_EXPERIMENT_CFG,
steps: Optional[Sequence[Union[Step, StepCollection]]] = None,
sagemaker_session: Optional[Session] = None,
):
"""Initialize a Pipeline
Args:
name (str): The name of the pipeline.
parameters (Sequence[Parameter]): The list of the parameters.
pipeline_experiment_config (Optional[PipelineExperimentConfig]): If set,
the workflow will attempt to create an experiment and trial before
executing the steps. Creation will be skipped if an experiment or a trial with
the same name already exists. By default, pipeline name is used as
experiment name and execution id is used as the trial name.
If set to None, no experiment or trial will be created automatically.
steps (Sequence[Union[Step, StepCollection]]): The list of the non-conditional steps
associated with the pipeline. Any steps that are within the
`if_steps` or `else_steps` of a `ConditionStep` cannot be listed in the steps of a
pipeline. Of particular note, the workflow service rejects any pipeline definitions
that specify a step in the list of steps of a pipeline and that step in the
`if_steps` or `else_steps` of any `ConditionStep`.
sagemaker_session (sagemaker.session.Session): Session object that manages interactions
with Amazon SageMaker APIs and any other AWS services needed. If not specified, the
pipeline creates one using the default AWS configuration chain.
"""
self.name = name
self.parameters = parameters if parameters else []
self.pipeline_experiment_config = pipeline_experiment_config
self.steps = steps if steps else []
self.sagemaker_session = sagemaker_session if sagemaker_session else Session()
self._version = "2020-12-01"
self._metadata = dict()
self._step_map = dict()
_generate_step_map(self.steps, self._step_map)
def to_request(self) -> RequestType:
"""Gets the request structure for workflow service calls."""
return {
"Version": self._version,
"Metadata": self._metadata,
"Parameters": list_to_request(self.parameters),
"PipelineExperimentConfig": self.pipeline_experiment_config.to_request()
if self.pipeline_experiment_config is not None
else None,
"Steps": list_to_request(self.steps),
}
def create(
self,
role_arn: str,
description: str = None,
tags: List[Dict[str, str]] = None,
parallelism_config: ParallelismConfiguration = None,
) -> Dict[str, Any]:
"""Creates a Pipeline in the Pipelines service.
Args:
role_arn (str): The role arn that is assumed by the pipeline to create step artifacts.
description (str): A description of the pipeline.
tags (List[Dict[str, str]]): A list of {"Key": "string", "Value": "string"} dicts as
tags.
parallelism_config (Optional[ParallelismConfiguration]): Parallelism configuration
that is applied to each of the executions of the pipeline. It takes precedence
over the parallelism configuration of the parent pipeline.
Returns:
A response dict from the service.
"""
if self.sagemaker_session.local_mode:
if parallelism_config:
logger.warning("Pipeline parallelism config is not supported in the local mode.")
return self.sagemaker_session.sagemaker_client.create_pipeline(self, description)
tags = _append_project_tags(tags)
kwargs = self._create_args(role_arn, description, parallelism_config)
update_args(
kwargs,
Tags=tags,
)
return self.sagemaker_session.sagemaker_client.create_pipeline(**kwargs)
def _create_args(
self, role_arn: str, description: str, parallelism_config: ParallelismConfiguration
):
"""Constructs the keyword argument dict for a create_pipeline call.
Args:
role_arn (str): The role arn that is assumed by pipelines to create step artifacts.
description (str): A description of the pipeline.
parallelism_config (Optional[ParallelismConfiguration]): Parallelism configuration
that is applied to each of the executions of the pipeline. It takes precedence
over the parallelism configuration of the parent pipeline.
Returns:
A keyword argument dict for calling create_pipeline.
"""
pipeline_definition = self.definition()
kwargs = dict(
PipelineName=self.name,
RoleArn=role_arn,
)
# If pipeline definition is large, upload to S3 bucket and
# provide PipelineDefinitionS3Location to request instead.
if len(pipeline_definition.encode("utf-8")) < 1024 * 100:
kwargs["PipelineDefinition"] = pipeline_definition
else:
desired_s3_uri = s3.s3_path_join(
"s3://", self.sagemaker_session.default_bucket(), self.name
)
s3.S3Uploader.upload_string_as_file_body(
body=pipeline_definition,
desired_s3_uri=desired_s3_uri,
sagemaker_session=self.sagemaker_session,
)
kwargs["PipelineDefinitionS3Location"] = {
"Bucket": self.sagemaker_session.default_bucket(),
"ObjectKey": self.name,
}
update_args(
kwargs, PipelineDescription=description, ParallelismConfiguration=parallelism_config
)
return kwargs
def describe(self) -> Dict[str, Any]:
"""Describes a Pipeline in the Workflow service.
Returns:
Response dict from the service. See `boto3 client documentation
<https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/\
sagemaker.html#SageMaker.Client.describe_pipeline>`_
"""
return self.sagemaker_session.sagemaker_client.describe_pipeline(PipelineName=self.name)
def update(
self,
role_arn: str,
description: str = None,
parallelism_config: ParallelismConfiguration = None,
) -> Dict[str, Any]:
"""Updates a Pipeline in the Workflow service.
Args:
role_arn (str): The role arn that is assumed by pipelines to create step artifacts.
description (str): A description of the pipeline.
parallelism_config (Optional[ParallelismConfiguration]): Parallelism configuration
that is applied to each of the executions of the pipeline. It takes precedence
over the parallelism configuration of the parent pipeline.
Returns:
A response dict from the service.
"""
if self.sagemaker_session.local_mode:
if parallelism_config:
logger.warning("Pipeline parallelism config is not supported in the local mode.")
return self.sagemaker_session.sagemaker_client.update_pipeline(self, description)
self._step_map = dict()
_generate_step_map(self.steps, self._step_map)
kwargs = self._create_args(role_arn, description, parallelism_config)
return self.sagemaker_session.sagemaker_client.update_pipeline(**kwargs)
def upsert(
self,
role_arn: str,
description: str = None,
tags: List[Dict[str, str]] = None,
parallelism_config: ParallelismConfiguration = None,
) -> Dict[str, Any]:
"""Creates a pipeline or updates it, if it already exists.
Args:
role_arn (str): The role arn that is assumed by workflow to create step artifacts.
description (str): A description of the pipeline.
tags (List[Dict[str, str]]): A list of {"Key": "string", "Value": "string"} dicts as
tags.
parallelism_config (Optional[Config for parallel steps, Parallelism configuration that
is applied to each of. the executions
Returns:
response dict from service
"""
exists = True
try:
self.describe()
except ClientError as e:
err = e.response.get("Error", {})
if err.get("Code", None) == "ResourceNotFound":
exists = False
else:
raise e
if not exists:
response = self.create(role_arn, description, tags, parallelism_config)
else:
response = self.update(role_arn, description)
if tags is not None:
old_tags = self.sagemaker_session.sagemaker_client.list_tags(
ResourceArn=response["PipelineArn"]
)["Tags"]
tag_keys = [tag["Key"] for tag in tags]
for old_tag in old_tags:
if old_tag["Key"] not in tag_keys:
tags.append(old_tag)
self.sagemaker_session.sagemaker_client.add_tags(
ResourceArn=response["PipelineArn"], Tags=tags
)
return response
def delete(self) -> Dict[str, Any]:
"""Deletes a Pipeline in the Workflow service.
Returns:
A response dict from the service.
"""
return self.sagemaker_session.sagemaker_client.delete_pipeline(PipelineName=self.name)
def start(
self,
parameters: Dict[str, Union[str, bool, int, float]] = None,
execution_display_name: str = None,
execution_description: str = None,
parallelism_config: ParallelismConfiguration = None,
):
"""Starts a Pipeline execution in the Workflow service.
Args:
parameters (Dict[str, Union[str, bool, int, float]]): values to override
pipeline parameters.
execution_display_name (str): The display name of the pipeline execution.
execution_description (str): A description of the execution.
parallelism_config (Optional[ParallelismConfiguration]): Parallelism configuration
that is applied to each of the executions of the pipeline. It takes precedence
over the parallelism configuration of the parent pipeline.
Returns:
A `_PipelineExecution` instance, if successful.
"""
kwargs = dict(PipelineName=self.name)
update_args(
kwargs,
PipelineExecutionDescription=execution_description,
PipelineExecutionDisplayName=execution_display_name,
ParallelismConfiguration=parallelism_config,
)
if self.sagemaker_session.local_mode:
update_args(kwargs, PipelineParameters=parameters)
return self.sagemaker_session.sagemaker_client.start_pipeline_execution(**kwargs)
update_args(kwargs, PipelineParameters=format_start_parameters(parameters))
response = self.sagemaker_session.sagemaker_client.start_pipeline_execution(**kwargs)
return _PipelineExecution(
arn=response["PipelineExecutionArn"],
sagemaker_session=self.sagemaker_session,
)
def definition(self) -> str:
"""Converts a request structure to string representation for workflow service calls."""
request_dict = self.to_request()
self._interpolate_step_collection_name_in_depends_on(request_dict["Steps"])
request_dict["PipelineExperimentConfig"] = interpolate(
request_dict["PipelineExperimentConfig"], {}, {}
)
callback_output_to_step_map = _map_callback_outputs(self.steps)
lambda_output_to_step_name = _map_lambda_outputs(self.steps)
request_dict["Steps"] = interpolate(
request_dict["Steps"],
callback_output_to_step_map=callback_output_to_step_map,
lambda_output_to_step_map=lambda_output_to_step_name,
)
return json.dumps(request_dict)
def _interpolate_step_collection_name_in_depends_on(self, step_requests: list):
"""Insert step names as per `StepCollection` name in depends_on list
Args:
step_requests (list): The list of raw step request dicts without any interpolation.
"""
for step_request in step_requests:
depends_on = []
for depend_step_name in step_request.get("DependsOn", []):
if isinstance(self._step_map[depend_step_name], StepCollection):
depends_on.extend([s.name for s in self._step_map[depend_step_name].steps])
else:
depends_on.append(depend_step_name)
if depends_on:
step_request["DependsOn"] = depends_on
if step_request["Type"] == StepTypeEnum.CONDITION.value:
sub_step_requests = (
step_request["Arguments"]["IfSteps"] + step_request["Arguments"]["ElseSteps"]
)
self._interpolate_step_collection_name_in_depends_on(sub_step_requests)
def format_start_parameters(parameters: Dict[str, Any]) -> List[Dict[str, Any]]:
"""Formats start parameter overrides as a list of dicts.
This list of dicts adheres to the request schema of:
`{"Name": "MyParameterName", "Value": "MyValue"}`
Args:
parameters (Dict[str, Any]): A dict of named values where the keys are
the names of the parameters to pass values into.
"""
if parameters is None:
return None
return [{"Name": name, "Value": str(value)} for name, value in parameters.items()]
def interpolate(
request_obj: RequestType,
callback_output_to_step_map: Dict[str, str],
lambda_output_to_step_map: Dict[str, str],
) -> RequestType:
"""Replaces Parameter values in a list of nested Dict[str, Any] with their workflow expression.
Args:
request_obj (RequestType): The request dict.
callback_output_to_step_map (Dict[str, str]): A dict of output name -> step name.
lambda_output_to_step_map (Dict[str, str]): A dict of output name -> step name.
Returns:
RequestType: The request dict with Parameter values replaced by their expression.
"""
try:
request_obj_copy = deepcopy(request_obj)
return _interpolate(
request_obj_copy,
callback_output_to_step_map=callback_output_to_step_map,
lambda_output_to_step_map=lambda_output_to_step_map,
)
except TypeError as type_err:
raise TypeError("Not able to interpolate Pipeline definition: %s" % type_err)
def _interpolate(
obj: Union[RequestType, Any],
callback_output_to_step_map: Dict[str, str],
lambda_output_to_step_map: Dict[str, str],
):
"""Walks the nested request dict, replacing Parameter type values with workflow expressions.
Args:
obj (Union[RequestType, Any]): The request dict.
callback_output_to_step_map (Dict[str, str]): A dict of output name -> step name.
"""
if isinstance(obj, (Expression, Parameter, Properties)):
return obj.expr
if isinstance(obj, CallbackOutput):
step_name = callback_output_to_step_map[obj.output_name]
return obj.expr(step_name)
if isinstance(obj, LambdaOutput):
step_name = lambda_output_to_step_map[obj.output_name]
return obj.expr(step_name)
if isinstance(obj, dict):
new = obj.__class__()
for key, value in obj.items():
new[key] = interpolate(value, callback_output_to_step_map, lambda_output_to_step_map)
elif isinstance(obj, (list, set, tuple)):
new = obj.__class__(
interpolate(value, callback_output_to_step_map, lambda_output_to_step_map)
for value in obj
)
else:
return obj
return new
def _map_callback_outputs(steps: List[Step]):
"""Iterate over the provided steps, building a map of callback output parameters to step names.
Args:
step (List[Step]): The steps list.
"""
callback_output_map = {}
for step in steps:
if isinstance(step, CallbackStep):
if step.outputs:
for output in step.outputs:
callback_output_map[output.output_name] = step.name
return callback_output_map
def _map_lambda_outputs(steps: List[Step]):
"""Iterate over the provided steps, building a map of lambda output parameters to step names.
Args:
step (List[Step]): The steps list.
"""
lambda_output_map = {}
for step in steps:
if isinstance(step, LambdaStep):
if step.outputs:
for output in step.outputs:
lambda_output_map[output.output_name] = step.name
return lambda_output_map
def update_args(args: Dict[str, Any], **kwargs):
"""Updates the request arguments dict with a value, if populated.
This handles the case when the service API doesn't like NoneTypes for argument values.
Args:
request_args (Dict[str, Any]): The request arguments dict.
kwargs: key, value pairs to update the args dict with.
"""
for key, value in kwargs.items():
if value is not None:
args.update({key: value})
def _generate_step_map(
steps: Sequence[Union[Step, StepCollection]], step_map: dict
) -> Dict[str, Any]:
"""Helper method to create a mapping from Step/Step Collection name to itself."""
for step in steps:
if step.name in step_map:
raise ValueError(
"Pipeline steps cannot have duplicate names. In addition, steps added in "
"the ConditionStep cannot be added in the Pipeline steps list."
)
step_map[step.name] = step
if isinstance(step, ConditionStep):
_generate_step_map(step.if_steps + step.else_steps, step_map)
if isinstance(step, StepCollection):
_generate_step_map(step.steps, step_map)
@attr.s
class _PipelineExecution:
"""Internal class for encapsulating pipeline execution instances.
Attributes:
arn (str): The arn of the pipeline execution.
sagemaker_session (sagemaker.session.Session): Session object which manages interactions
with Amazon SageMaker APIs and any other AWS services needed. If not specified, the
pipeline creates one using the default AWS configuration chain.
"""
arn: str = attr.ib()
sagemaker_session: Session = attr.ib(factory=Session)
def stop(self):
"""Stops a pipeline execution."""
return self.sagemaker_session.sagemaker_client.stop_pipeline_execution(
PipelineExecutionArn=self.arn
)
def describe(self):
"""Describes a pipeline execution.
Returns:
Information about the pipeline execution. See
`boto3 client describe_pipeline_execution
<https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/\
sagemaker.html#SageMaker.Client.describe_pipeline_execution>`_.
"""
return self.sagemaker_session.sagemaker_client.describe_pipeline_execution(
PipelineExecutionArn=self.arn,
)
def list_steps(self):
"""Describes a pipeline execution's steps.
Returns:
Information about the steps of the pipeline execution. See
`boto3 client list_pipeline_execution_steps
<https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/\
sagemaker.html#SageMaker.Client.list_pipeline_execution_steps>`_.
"""
response = self.sagemaker_session.sagemaker_client.list_pipeline_execution_steps(
PipelineExecutionArn=self.arn
)
return response["PipelineExecutionSteps"]
def wait(self, delay=30, max_attempts=60):
"""Waits for a pipeline execution.
Args:
delay (int): The polling interval. (Defaults to 30 seconds)
max_attempts (int): The maximum number of polling attempts.
(Defaults to 60 polling attempts)
"""
waiter_id = "PipelineExecutionComplete"
# TODO: this waiter should be included in the botocore
model = botocore.waiter.WaiterModel(
{
"version": 2,
"waiters": {
waiter_id: {
"delay": delay,
"maxAttempts": max_attempts,
"operation": "DescribePipelineExecution",
"acceptors": [
{
"expected": "Succeeded",
"matcher": "path",
"state": "success",
"argument": "PipelineExecutionStatus",
},
{
"expected": "Failed",
"matcher": "path",
"state": "failure",
"argument": "PipelineExecutionStatus",
},
],
}
},
}
)
waiter = botocore.waiter.create_waiter_with_client(
waiter_id, model, self.sagemaker_session.sagemaker_client
)
waiter.wait(PipelineExecutionArn=self.arn)
class PipelineGraph:
"""Helper class representing the Pipeline Directed Acyclic Graph (DAG)
Attributes:
steps (Sequence[Union[Step, StepCollection]]): Sequence of `Step`s and/or `StepCollection`s
that represent each node in the pipeline DAG
"""
def __init__(self, steps: Sequence[Union[Step, StepCollection]]):
self.step_map = {}
_generate_step_map(steps, self.step_map)
self.adjacency_list = self._initialize_adjacency_list()
if self.is_cyclic():
raise ValueError("Cycle detected in pipeline step graph.")
@classmethod
def from_pipeline(cls, pipeline: Pipeline):
"""Create a PipelineGraph object from the Pipeline object."""
return cls(pipeline.steps)
def _initialize_adjacency_list(self) -> Dict[str, List[str]]:
"""Generate an adjacency list representing the step dependency DAG in this pipeline."""
from collections import defaultdict
dependency_list = defaultdict(set)
for step in self.step_map.values():
if isinstance(step, Step):
dependency_list[step.name].update(step._find_step_dependencies(self.step_map))
if isinstance(step, ConditionStep):
for child_step in step.if_steps + step.else_steps:
if isinstance(child_step, Step):
dependency_list[child_step.name].add(step.name)
elif isinstance(child_step, StepCollection):
child_first_step = self.step_map[child_step.name].steps[0].name
dependency_list[child_first_step].add(step.name)
adjacency_list = {}
for step in dependency_list:
for step_dependency in dependency_list[step]:
adjacency_list[step_dependency] = list(
set(adjacency_list.get(step_dependency, []) + [step])
)
for step in dependency_list:
if step not in adjacency_list:
adjacency_list[step] = []
return adjacency_list
def is_cyclic(self) -> bool:
"""Check if this pipeline graph is cyclic.
Returns true if it is cyclic, false otherwise.
"""
def is_cyclic_helper(current_step):
visited_steps.add(current_step)
recurse_steps.add(current_step)
for child_step in self.adjacency_list[current_step]:
if child_step in recurse_steps:
return True
if child_step not in visited_steps:
if is_cyclic_helper(child_step):
return True
recurse_steps.remove(current_step)
return False
visited_steps = set()
recurse_steps = set()
for step in self.adjacency_list:
if step not in visited_steps:
if is_cyclic_helper(step):
return True
return False
def get_steps_in_sub_dag(
self, current_step: Union[Step, StepCollection], sub_dag_steps: Set[str] = None
) -> Set[str]:
"""Get names of all steps (including current step) in the sub dag of current step.
Returns a set of step names in the sub dag.
"""
if sub_dag_steps is None:
sub_dag_steps = set()
if isinstance(current_step, StepCollection):
current_steps = current_step.steps
else:
current_steps = [current_step]
for step in current_steps:
if step.name not in self.adjacency_list:
raise ValueError("Step: %s does not exist in the pipeline." % step.name)
sub_dag_steps.add(step.name)
for sub_step in self.adjacency_list[step.name]:
self.get_steps_in_sub_dag(self.step_map.get(sub_step), sub_dag_steps)
return sub_dag_steps
def __iter__(self):
"""Perform topological sort traversal of the Pipeline Graph."""
def topological_sort(current_step):
visited_steps.add(current_step)
for child_step in self.adjacency_list[current_step]:
if child_step not in visited_steps:
topological_sort(child_step)
self.stack.append(current_step)
visited_steps = set()
self.stack = [] # pylint: disable=W0201
for step in self.adjacency_list:
if step not in visited_steps:
topological_sort(step)
return self
def __next__(self) -> Step:
"""Return the next Step node from the Topological sort order."""
while self.stack:
return self.step_map.get(self.stack.pop())
raise StopIteration