# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License 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.
from __future__ import annotations
import time
from functools import cached_property
from typing import TYPE_CHECKING, Sequence
from deprecated import deprecated
from airflow.exceptions import AirflowException, AirflowProviderDeprecationWarning, AirflowSkipException
from airflow.providers.amazon.aws.hooks.sagemaker import LogState, SageMakerHook
from airflow.sensors.base import BaseSensorOperator
if TYPE_CHECKING:
from airflow.utils.context import Context
[docs]class SageMakerBaseSensor(BaseSensorOperator):
"""
Contains general sensor behavior for SageMaker.
Subclasses should implement get_sagemaker_response() and state_from_response() methods.
Subclasses should also implement NON_TERMINAL_STATES and FAILED_STATE methods.
"""
def __init__(self, *, aws_conn_id: str | None = "aws_default", resource_type: str = "job", **kwargs):
super().__init__(**kwargs)
self.aws_conn_id = aws_conn_id
self.resource_type = resource_type # only used for logs, to say what kind of resource we are sensing
@deprecated(reason="use `hook` property instead.", category=AirflowProviderDeprecationWarning)
[docs] def get_hook(self) -> SageMakerHook:
"""Get SageMakerHook."""
return self.hook
@cached_property
[docs] def hook(self) -> SageMakerHook:
return SageMakerHook(aws_conn_id=self.aws_conn_id)
[docs] def poke(self, context: Context):
response = self.get_sagemaker_response()
if response["ResponseMetadata"]["HTTPStatusCode"] != 200:
self.log.info("Bad HTTP response: %s", response)
return False
state = self.state_from_response(response)
self.log.info("%s currently %s", self.resource_type, state)
if state in self.non_terminal_states():
return False
if state in self.failed_states():
failed_reason = self.get_failed_reason_from_response(response)
# TODO: remove this if block when min_airflow_version is set to higher than 2.7.1
message = f"Sagemaker {self.resource_type} failed for the following reason: {failed_reason}"
if self.soft_fail:
raise AirflowSkipException(message)
raise AirflowException(message)
return True
[docs] def non_terminal_states(self) -> set[str]:
"""Return states with should not terminate."""
raise NotImplementedError("Please implement non_terminal_states() in subclass")
[docs] def failed_states(self) -> set[str]:
"""Return states with are considered failed."""
raise NotImplementedError("Please implement failed_states() in subclass")
[docs] def get_sagemaker_response(self) -> dict:
"""Check status of a SageMaker task."""
raise NotImplementedError("Please implement get_sagemaker_response() in subclass")
[docs] def get_failed_reason_from_response(self, response: dict) -> str:
"""Extract the reason for failure from an AWS response."""
return "Unknown"
[docs] def state_from_response(self, response: dict) -> str:
"""Extract the state from an AWS response."""
raise NotImplementedError("Please implement state_from_response() in subclass")
[docs]class SageMakerEndpointSensor(SageMakerBaseSensor):
"""
Poll the endpoint state until it reaches a terminal state; raise AirflowException with the failure reason.
.. seealso::
For more information on how to use this sensor, take a look at the guide:
:ref:`howto/sensor:SageMakerEndpointSensor`
:param endpoint_name: Name of the endpoint instance to watch.
"""
[docs] template_fields: Sequence[str] = ("endpoint_name",)
[docs] template_ext: Sequence[str] = ()
def __init__(self, *, endpoint_name, **kwargs):
super().__init__(**kwargs)
self.endpoint_name = endpoint_name
[docs] def non_terminal_states(self):
return SageMakerHook.endpoint_non_terminal_states
[docs] def failed_states(self):
return SageMakerHook.failed_states
[docs] def get_sagemaker_response(self):
self.log.info("Poking Sagemaker Endpoint %s", self.endpoint_name)
return self.hook.describe_endpoint(self.endpoint_name)
[docs] def get_failed_reason_from_response(self, response):
return response["FailureReason"]
[docs] def state_from_response(self, response):
return response["EndpointStatus"]
[docs]class SageMakerTuningSensor(SageMakerBaseSensor):
"""
Poll the tuning state until it reaches a terminal state; raise AirflowException with the failure reason.
.. seealso::
For more information on how to use this sensor, take a look at the guide:
:ref:`howto/sensor:SageMakerTuningSensor`
:param job_name: Name of the tuning instance to watch.
"""
[docs] template_fields: Sequence[str] = ("job_name",)
[docs] template_ext: Sequence[str] = ()
def __init__(self, *, job_name: str, **kwargs):
super().__init__(**kwargs)
self.job_name = job_name
[docs] def non_terminal_states(self):
return SageMakerHook.non_terminal_states
[docs] def failed_states(self):
return SageMakerHook.failed_states
[docs] def get_sagemaker_response(self):
self.log.info("Poking Sagemaker Tuning Job %s", self.job_name)
return self.hook.describe_tuning_job(self.job_name)
[docs] def get_failed_reason_from_response(self, response):
return response["FailureReason"]
[docs] def state_from_response(self, response):
return response["HyperParameterTuningJobStatus"]
[docs]class SageMakerTrainingSensor(SageMakerBaseSensor):
"""
Poll the training job until it reaches a terminal state; raise AirflowException with the failure reason.
.. seealso::
For more information on how to use this sensor, take a look at the guide:
:ref:`howto/sensor:SageMakerTrainingSensor`
:param job_name: Name of the training job to watch.
:param print_log: Prints the cloudwatch log if True; Defaults to True.
"""
[docs] template_fields: Sequence[str] = ("job_name",)
[docs] template_ext: Sequence[str] = ()
def __init__(self, *, job_name, print_log=True, **kwargs):
super().__init__(**kwargs)
self.job_name = job_name
self.print_log = print_log
self.positions = {}
self.stream_names = []
self.instance_count: int | None = None
self.state: int | None = None
self.last_description = None
self.last_describe_job_call = None
self.log_resource_inited = False
[docs] def init_log_resource(self, hook: SageMakerHook) -> None:
"""Set tailing LogState for associated training job."""
description = hook.describe_training_job(self.job_name)
self.instance_count = description["ResourceConfig"]["InstanceCount"]
status = description["TrainingJobStatus"]
job_already_completed = status not in self.non_terminal_states()
self.state = LogState.COMPLETE if job_already_completed else LogState.TAILING
self.last_description = description
self.last_describe_job_call = time.monotonic()
self.log_resource_inited = True
[docs] def non_terminal_states(self):
return SageMakerHook.non_terminal_states
[docs] def failed_states(self):
return SageMakerHook.failed_states
[docs] def get_sagemaker_response(self):
if self.print_log:
if not self.log_resource_inited:
self.init_log_resource(self.hook)
(
self.state,
self.last_description,
self.last_describe_job_call,
) = self.hook.describe_training_job_with_log(
self.job_name,
self.positions,
self.stream_names,
self.instance_count,
self.state,
self.last_description,
self.last_describe_job_call,
)
else:
self.last_description = self.hook.describe_training_job(self.job_name)
status = self.state_from_response(self.last_description)
if (status not in self.non_terminal_states()) and (status not in self.failed_states()):
billable_time = (
self.last_description["TrainingEndTime"] - self.last_description["TrainingStartTime"]
) * self.last_description["ResourceConfig"]["InstanceCount"]
self.log.info("Billable seconds: %s", (int(billable_time.total_seconds()) + 1))
return self.last_description
[docs] def get_failed_reason_from_response(self, response):
return response["FailureReason"]
[docs] def state_from_response(self, response):
return response["TrainingJobStatus"]
[docs]class SageMakerPipelineSensor(SageMakerBaseSensor):
"""
Poll the pipeline until it reaches a terminal state; raise AirflowException with the failure reason.
.. seealso::
For more information on how to use this sensor, take a look at the guide:
:ref:`howto/sensor:SageMakerPipelineSensor`
:param pipeline_exec_arn: ARN of the pipeline to watch.
:param verbose: Whether to print steps details while waiting for completion.
Defaults to true, consider turning off for pipelines that have thousands of steps.
"""
[docs] template_fields: Sequence[str] = ("pipeline_exec_arn",)
def __init__(self, *, pipeline_exec_arn: str, verbose: bool = True, **kwargs):
super().__init__(resource_type="pipeline", **kwargs)
self.pipeline_exec_arn = pipeline_exec_arn
self.verbose = verbose
[docs] def non_terminal_states(self) -> set[str]:
return SageMakerHook.pipeline_non_terminal_states
[docs] def failed_states(self) -> set[str]:
return SageMakerHook.failed_states
[docs] def get_sagemaker_response(self) -> dict:
self.log.info("Poking Sagemaker Pipeline Execution %s", self.pipeline_exec_arn)
return self.hook.describe_pipeline_exec(self.pipeline_exec_arn, self.verbose)
[docs] def state_from_response(self, response: dict) -> str:
return response["PipelineExecutionStatus"]
[docs]class SageMakerAutoMLSensor(SageMakerBaseSensor):
"""
Poll the auto ML job until it reaches a terminal state; raise AirflowException with the failure reason.
.. seealso::
For more information on how to use this sensor, take a look at the guide:
:ref:`howto/sensor:SageMakerAutoMLSensor`
:param job_name: unique name of the AutoML job to watch.
"""
[docs] template_fields: Sequence[str] = ("job_name",)
def __init__(self, *, job_name: str, **kwargs):
super().__init__(resource_type="autoML job", **kwargs)
self.job_name = job_name
[docs] def non_terminal_states(self) -> set[str]:
return SageMakerHook.non_terminal_states
[docs] def failed_states(self) -> set[str]:
return SageMakerHook.failed_states
[docs] def get_sagemaker_response(self) -> dict:
self.log.info("Poking Sagemaker AutoML Execution %s", self.job_name)
return self.hook._describe_auto_ml_job(self.job_name)
[docs] def state_from_response(self, response: dict) -> str:
return response["AutoMLJobStatus"]
