#
# 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 os
import re
import tarfile
import tempfile
import time
from collections import Counter, namedtuple
from collections.abc import AsyncGenerator, Generator
from datetime import datetime
from functools import partial
from typing import Any, Callable, cast
from asgiref.sync import sync_to_async
from botocore.exceptions import ClientError
from airflow.exceptions import AirflowException
from airflow.providers.amazon.aws.hooks.base_aws import AwsBaseHook
from airflow.providers.amazon.aws.hooks.logs import AwsLogsHook
from airflow.providers.amazon.aws.hooks.s3 import S3Hook
from airflow.providers.amazon.aws.utils.tags import format_tags
from airflow.utils import timezone
[docs]class LogState:
"""
Enum-style class holding all possible states of CloudWatch log streams.
https://sagemaker.readthedocs.io/en/stable/session.html#sagemaker.session.LogState
"""
# Position is a tuple that includes the last read timestamp and the number of items that were read
# at that time. This is used to figure out which event to start with on the next read.
[docs]Position = namedtuple("Position", ["timestamp", "skip"])
[docs]def argmin(arr, f: Callable) -> int | None:
"""
Given callable ``f``, find index in ``arr`` to minimize ``f(arr[i])``.
None is returned if ``arr`` is empty.
"""
min_value = None
min_idx = None
for idx, item in enumerate(arr):
if item is not None:
if min_value is None or f(item) < min_value:
min_value = f(item)
min_idx = idx
return min_idx
[docs]def secondary_training_status_changed(current_job_description: dict, prev_job_description: dict) -> bool:
"""
Check if training job's secondary status message has changed.
:param current_job_description: Current job description, returned from DescribeTrainingJob call.
:param prev_job_description: Previous job description, returned from DescribeTrainingJob call.
:return: Whether the secondary status message of a training job changed or not.
"""
current_secondary_status_transitions = current_job_description.get("SecondaryStatusTransitions")
if not current_secondary_status_transitions:
return False
prev_job_secondary_status_transitions = (
prev_job_description.get("SecondaryStatusTransitions") if prev_job_description is not None else None
)
last_message = (
prev_job_secondary_status_transitions[-1]["StatusMessage"]
if prev_job_secondary_status_transitions
else ""
)
message = current_job_description["SecondaryStatusTransitions"][-1]["StatusMessage"]
return message != last_message
[docs]def secondary_training_status_message(
job_description: dict[str, list[Any]], prev_description: dict | None
) -> str:
"""
Format string containing start time and the secondary training job status message.
:param job_description: Returned response from DescribeTrainingJob call
:param prev_description: Previous job description from DescribeTrainingJob call
:return: Job status string to be printed.
"""
current_transitions = job_description.get("SecondaryStatusTransitions")
if not current_transitions:
return ""
prev_transitions_num = 0
if prev_description is not None:
if prev_description.get("SecondaryStatusTransitions") is not None:
prev_transitions_num = len(prev_description["SecondaryStatusTransitions"])
transitions_to_print = (
current_transitions[-1:]
if len(current_transitions) == prev_transitions_num
else current_transitions[prev_transitions_num - len(current_transitions) :]
)
status_strs = []
for transition in transitions_to_print:
message = transition["StatusMessage"]
time_utc = timezone.convert_to_utc(cast(datetime, job_description["LastModifiedTime"]))
status_strs.append(f"{time_utc:%Y-%m-%d %H:%M:%S} {transition['Status']} - {message}")
return "\n".join(status_strs)
[docs]class SageMakerHook(AwsBaseHook):
"""
Interact with Amazon SageMaker.
Provide thick wrapper around
:external+boto3:py:class:`boto3.client("sagemaker") <SageMaker.Client>`.
Additional arguments (such as ``aws_conn_id``) may be specified and
are passed down to the underlying AwsBaseHook.
.. seealso::
- :class:`airflow.providers.amazon.aws.hooks.base_aws.AwsBaseHook`
"""
[docs] non_terminal_states = {"InProgress", "Stopping"}
[docs] endpoint_non_terminal_states = {"Creating", "Updating", "SystemUpdating", "RollingBack", "Deleting"}
[docs] pipeline_non_terminal_states = {"Executing", "Stopping"}
[docs] processing_job_non_terminal_states = {"InProgress", "Stopping"}
[docs] failed_states = {"Failed"}
[docs] processing_job_failed_states = {*failed_states, "Stopped"}
[docs] training_failed_states = {*failed_states, "Stopped"}
def __init__(self, *args, **kwargs):
super().__init__(client_type="sagemaker", *args, **kwargs)
self.s3_hook = S3Hook(aws_conn_id=self.aws_conn_id)
self.logs_hook = AwsLogsHook(aws_conn_id=self.aws_conn_id)
[docs] def tar_and_s3_upload(self, path: str, key: str, bucket: str) -> None:
"""
Tar the local file or directory and upload to s3.
:param path: local file or directory
:param key: s3 key
:param bucket: s3 bucket
"""
with tempfile.TemporaryFile() as temp_file:
if os.path.isdir(path):
files = [os.path.join(path, name) for name in os.listdir(path)]
else:
files = [path]
with tarfile.open(mode="w:gz", fileobj=temp_file) as tar_file:
for f in files:
tar_file.add(f, arcname=os.path.basename(f))
temp_file.seek(0)
self.s3_hook.load_file_obj(temp_file, key, bucket, replace=True)
[docs] def check_s3_url(self, s3url: str) -> bool:
"""
Check if an S3 URL exists.
:param s3url: S3 url
"""
bucket, key = S3Hook.parse_s3_url(s3url)
if not self.s3_hook.check_for_bucket(bucket_name=bucket):
raise AirflowException(f"The input S3 Bucket {bucket} does not exist ")
if (
key
and not self.s3_hook.check_for_key(key=key, bucket_name=bucket)
and not self.s3_hook.check_for_prefix(prefix=key, bucket_name=bucket, delimiter="/")
):
# check if s3 key exists in the case user provides a single file
# or if s3 prefix exists in the case user provides multiple files in
# a prefix
raise AirflowException(
f"The input S3 Key or Prefix {s3url} does not exist in the Bucket {bucket}"
)
return True
[docs] def check_training_config(self, training_config: dict) -> None:
"""
Check if a training configuration is valid.
:param training_config: training_config
"""
if "InputDataConfig" in training_config:
for channel in training_config["InputDataConfig"]:
if "S3DataSource" in channel["DataSource"]:
self.check_s3_url(channel["DataSource"]["S3DataSource"]["S3Uri"])
[docs] def check_tuning_config(self, tuning_config: dict) -> None:
"""
Check if a tuning configuration is valid.
:param tuning_config: tuning_config
"""
for channel in tuning_config["TrainingJobDefinition"]["InputDataConfig"]:
if "S3DataSource" in channel["DataSource"]:
self.check_s3_url(channel["DataSource"]["S3DataSource"]["S3Uri"])
[docs] def multi_stream_iter(self, log_group: str, streams: list, positions=None) -> Generator:
"""
Iterate over the available events.
The events coming from a set of log streams in a single log group
interleaving the events from each stream so they're yielded in timestamp order.
:param log_group: The name of the log group.
:param streams: A list of the log stream names. The position of the stream in this list is
the stream number.
:param positions: A list of pairs of (timestamp, skip) which represents the last record
read from each stream.
:return: A tuple of (stream number, cloudwatch log event).
"""
positions = positions or {s: Position(timestamp=0, skip=0) for s in streams}
event_iters = [
self.logs_hook.get_log_events(log_group, s, positions[s].timestamp, positions[s].skip)
for s in streams
]
events: list[Any | None] = []
for event_stream in event_iters:
if event_stream:
try:
events.append(next(event_stream))
except StopIteration:
events.append(None)
else:
events.append(None)
while any(events):
i = argmin(events, lambda x: x["timestamp"] if x else 9999999999) or 0
yield i, events[i]
try:
events[i] = next(event_iters[i])
except StopIteration:
events[i] = None
[docs] def create_training_job(
self,
config: dict,
wait_for_completion: bool = True,
print_log: bool = True,
check_interval: int = 30,
max_ingestion_time: int | None = None,
):
"""
Start a model training job.
After training completes, Amazon SageMaker saves the resulting model
artifacts to an Amazon S3 location that you specify.
:param config: the config for training
:param wait_for_completion: if the program should keep running until job finishes
:param check_interval: the time interval in seconds which the operator
will check the status of any SageMaker job
:param max_ingestion_time: the maximum ingestion time in seconds. Any
SageMaker jobs that run longer than this will fail. Setting this to
None implies no timeout for any SageMaker job.
:return: A response to training job creation
"""
self.check_training_config(config)
response = self.get_conn().create_training_job(**config)
if print_log:
self.check_training_status_with_log(
config["TrainingJobName"],
self.non_terminal_states,
self.training_failed_states,
wait_for_completion,
check_interval,
max_ingestion_time,
)
elif wait_for_completion:
describe_response = self.check_status(
config["TrainingJobName"],
"TrainingJobStatus",
self.describe_training_job,
check_interval,
max_ingestion_time,
)
billable_seconds = SageMakerHook.count_billable_seconds(
training_start_time=describe_response["TrainingStartTime"],
training_end_time=describe_response["TrainingEndTime"],
instance_count=describe_response["ResourceConfig"]["InstanceCount"],
)
self.log.info("Billable seconds: %d", billable_seconds)
return response
[docs] def create_tuning_job(
self,
config: dict,
wait_for_completion: bool = True,
check_interval: int = 30,
max_ingestion_time: int | None = None,
):
"""
Start a hyperparameter tuning job.
A hyperparameter tuning job finds the best version of a model by running
many training jobs on your dataset using the algorithm you choose and
values for hyperparameters within ranges that you specify. It then
chooses the hyperparameter values that result in a model that performs
the best, as measured by an objective metric that you choose.
:param config: the config for tuning
:param wait_for_completion: if the program should keep running until job finishes
:param check_interval: the time interval in seconds which the operator
will check the status of any SageMaker job
:param max_ingestion_time: the maximum ingestion time in seconds. Any
SageMaker jobs that run longer than this will fail. Setting this to
None implies no timeout for any SageMaker job.
:return: A response to tuning job creation
"""
self.check_tuning_config(config)
response = self.get_conn().create_hyper_parameter_tuning_job(**config)
if wait_for_completion:
self.check_status(
config["HyperParameterTuningJobName"],
"HyperParameterTuningJobStatus",
self.describe_tuning_job,
check_interval,
max_ingestion_time,
)
return response
[docs] def create_processing_job(
self,
config: dict,
wait_for_completion: bool = True,
check_interval: int = 30,
max_ingestion_time: int | None = None,
):
"""
Use Amazon SageMaker Processing to analyze data and evaluate models.
With Processing, you can use a simplified, managed experience on
SageMaker to run your data processing workloads, such as feature
engineering, data validation, model evaluation, and model
interpretation.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.create_processing_job`
:param config: the config for processing job
:param wait_for_completion: if the program should keep running until job finishes
:param check_interval: the time interval in seconds which the operator
will check the status of any SageMaker job
:param max_ingestion_time: the maximum ingestion time in seconds. Any
SageMaker jobs that run longer than this will fail. Setting this to
None implies no timeout for any SageMaker job.
:return: A response to transform job creation
"""
response = self.get_conn().create_processing_job(**config)
if wait_for_completion:
self.check_status(
config["ProcessingJobName"],
"ProcessingJobStatus",
self.describe_processing_job,
check_interval,
max_ingestion_time,
)
return response
[docs] def create_model(self, config: dict):
"""
Create a model in Amazon SageMaker.
In the request, you name the model and describe a primary container. For
the primary container, you specify the Docker image that contains
inference code, artifacts (from prior training), and a custom
environment map that the inference code uses when you deploy the model
for predictions.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.create_model`
:param config: the config for model
:return: A response to model creation
"""
return self.get_conn().create_model(**config)
[docs] def create_endpoint_config(self, config: dict):
"""
Create an endpoint configuration to deploy models.
In the configuration, you identify one or more models, created using the
CreateModel API, to deploy and the resources that you want Amazon
SageMaker to provision.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.create_endpoint_config`
- :class:`airflow.providers.amazon.aws.hooks.sagemaker.SageMakerHook.create_model`
- :class:`airflow.providers.amazon.aws.hooks.sagemaker.SageMakerHook.create_endpoint`
:param config: the config for endpoint-config
:return: A response to endpoint config creation
"""
return self.get_conn().create_endpoint_config(**config)
[docs] def create_endpoint(
self,
config: dict,
wait_for_completion: bool = True,
check_interval: int = 30,
max_ingestion_time: int | None = None,
):
"""
Create an endpoint from configuration.
When you create a serverless endpoint, SageMaker provisions and manages
the compute resources for you. Then, you can make inference requests to
the endpoint and receive model predictions in response. SageMaker scales
the compute resources up and down as needed to handle your request traffic.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.create_endpoint`
- :class:`airflow.providers.amazon.aws.hooks.sagemaker.SageMakerHook.create_endpoint`
:param config: the config for endpoint
:param wait_for_completion: if the program should keep running until job finishes
:param check_interval: the time interval in seconds which the operator
will check the status of any SageMaker job
:param max_ingestion_time: the maximum ingestion time in seconds. Any
SageMaker jobs that run longer than this will fail. Setting this to
None implies no timeout for any SageMaker job.
:return: A response to endpoint creation
"""
response = self.get_conn().create_endpoint(**config)
if wait_for_completion:
self.check_status(
config["EndpointName"],
"EndpointStatus",
self.describe_endpoint,
check_interval,
max_ingestion_time,
non_terminal_states=self.endpoint_non_terminal_states,
)
return response
[docs] def update_endpoint(
self,
config: dict,
wait_for_completion: bool = True,
check_interval: int = 30,
max_ingestion_time: int | None = None,
):
"""
Deploy the config in the request and switch to using the new endpoint.
Resources provisioned for the endpoint using the previous EndpointConfig
are deleted (there is no availability loss).
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.update_endpoint`
:param config: the config for endpoint
:param wait_for_completion: if the program should keep running until job finishes
:param check_interval: the time interval in seconds which the operator
will check the status of any SageMaker job
:param max_ingestion_time: the maximum ingestion time in seconds. Any
SageMaker jobs that run longer than this will fail. Setting this to
None implies no timeout for any SageMaker job.
:return: A response to endpoint update
"""
response = self.get_conn().update_endpoint(**config)
if wait_for_completion:
self.check_status(
config["EndpointName"],
"EndpointStatus",
self.describe_endpoint,
check_interval,
max_ingestion_time,
non_terminal_states=self.endpoint_non_terminal_states,
)
return response
[docs] def describe_training_job(self, name: str):
"""
Get the training job info associated with the name.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.describe_training_job`
:param name: the name of the training job
:return: A dict contains all the training job info
"""
return self.get_conn().describe_training_job(TrainingJobName=name)
[docs] def describe_training_job_with_log(
self,
job_name: str,
positions,
stream_names: list,
instance_count: int,
state: int,
last_description: dict,
last_describe_job_call: float,
):
"""Get the associated training job info and print CloudWatch logs."""
log_group = "/aws/sagemaker/TrainingJobs"
if len(stream_names) < instance_count:
# Log streams are created whenever a container starts writing to stdout/err, so this list
# may be dynamic until we have a stream for every instance.
logs_conn = self.logs_hook.get_conn()
try:
streams = logs_conn.describe_log_streams(
logGroupName=log_group,
logStreamNamePrefix=job_name + "/",
orderBy="LogStreamName",
limit=instance_count,
)
stream_names = [s["logStreamName"] for s in streams["logStreams"]]
positions.update(
[(s, Position(timestamp=0, skip=0)) for s in stream_names if s not in positions]
)
except logs_conn.exceptions.ResourceNotFoundException:
# On the very first training job run on an account, there's no log group until
# the container starts logging, so ignore any errors thrown about that
pass
if stream_names:
for idx, event in self.multi_stream_iter(log_group, stream_names, positions):
self.log.info(event["message"])
ts, count = positions[stream_names[idx]]
if event["timestamp"] == ts:
positions[stream_names[idx]] = Position(timestamp=ts, skip=count + 1)
else:
positions[stream_names[idx]] = Position(timestamp=event["timestamp"], skip=1)
if state == LogState.COMPLETE:
return state, last_description, last_describe_job_call
if state == LogState.JOB_COMPLETE:
state = LogState.COMPLETE
elif time.monotonic() - last_describe_job_call >= 30:
description = self.describe_training_job(job_name)
last_describe_job_call = time.monotonic()
if secondary_training_status_changed(description, last_description):
self.log.info(secondary_training_status_message(description, last_description))
last_description = description
status = description["TrainingJobStatus"]
if status not in self.non_terminal_states:
state = LogState.JOB_COMPLETE
return state, last_description, last_describe_job_call
[docs] def describe_tuning_job(self, name: str) -> dict:
"""
Get the tuning job info associated with the name.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.describe_hyper_parameter_tuning_job`
:param name: the name of the tuning job
:return: A dict contains all the tuning job info
"""
return self.get_conn().describe_hyper_parameter_tuning_job(HyperParameterTuningJobName=name)
[docs] def describe_model(self, name: str) -> dict:
"""
Get the SageMaker model info associated with the name.
:param name: the name of the SageMaker model
:return: A dict contains all the model info
"""
return self.get_conn().describe_model(ModelName=name)
[docs] def describe_processing_job(self, name: str) -> dict:
"""
Get the processing job info associated with the name.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.describe_processing_job`
:param name: the name of the processing job
:return: A dict contains all the processing job info
"""
return self.get_conn().describe_processing_job(ProcessingJobName=name)
[docs] def describe_endpoint_config(self, name: str) -> dict:
"""
Get the endpoint config info associated with the name.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.describe_endpoint_config`
:param name: the name of the endpoint config
:return: A dict contains all the endpoint config info
"""
return self.get_conn().describe_endpoint_config(EndpointConfigName=name)
[docs] def describe_endpoint(self, name: str) -> dict:
"""
Get the description of an endpoint.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.describe_endpoint`
:param name: the name of the endpoint
:return: A dict contains all the endpoint info
"""
return self.get_conn().describe_endpoint(EndpointName=name)
[docs] def check_status(
self,
job_name: str,
key: str,
describe_function: Callable,
check_interval: int,
max_ingestion_time: int | None = None,
non_terminal_states: set | None = None,
) -> dict:
"""
Check status of a SageMaker resource.
:param job_name: name of the resource to check status, can be a job but
also pipeline for instance.
:param key: the key of the response dict that points to the state
:param describe_function: the function used to retrieve the status
:param args: the arguments for the function
:param check_interval: the time interval in seconds which the operator
will check the status of any SageMaker resource
:param max_ingestion_time: the maximum ingestion time in seconds. Any
SageMaker resources that run longer than this will fail. Setting
this to None implies no timeout for any SageMaker resource.
:param non_terminal_states: the set of nonterminal states
:return: response of describe call after resource is done
"""
if not non_terminal_states:
non_terminal_states = self.non_terminal_states
sec = 0
while True:
time.sleep(check_interval)
sec += check_interval
try:
response = describe_function(job_name)
status = response[key]
self.log.info("Resource still running for %s seconds... current status is %s", sec, status)
except KeyError:
raise AirflowException("Could not get status of the SageMaker resource")
except ClientError:
raise AirflowException("AWS request failed, check logs for more info")
if status in self.failed_states:
raise AirflowException(f"SageMaker resource failed because {response['FailureReason']}")
elif status not in non_terminal_states:
break
if max_ingestion_time and sec > max_ingestion_time:
# ensure that the resource gets killed if the max ingestion time is exceeded
raise AirflowException(f"SageMaker resource took more than {max_ingestion_time} seconds")
self.log.info("SageMaker resource completed")
return response
[docs] def check_training_status_with_log(
self,
job_name: str,
non_terminal_states: set,
failed_states: set,
wait_for_completion: bool,
check_interval: int,
max_ingestion_time: int | None = None,
):
"""
Display logs for a given training job.
Optionally tailing them until the job is complete.
:param job_name: name of the training job to check status and display logs for
:param non_terminal_states: the set of non_terminal states
:param failed_states: the set of failed states
:param wait_for_completion: Whether to keep looking for new log entries
until the job completes
:param check_interval: The interval in seconds between polling for new log entries and job completion
:param max_ingestion_time: the maximum ingestion time in seconds. Any
SageMaker jobs that run longer than this will fail. Setting this to
None implies no timeout for any SageMaker job.
"""
sec = 0
description = self.describe_training_job(job_name)
self.log.info(secondary_training_status_message(description, None))
instance_count = description["ResourceConfig"]["InstanceCount"]
status = description["TrainingJobStatus"]
stream_names: list = [] # The list of log streams
positions: dict = {} # The current position in each stream, map of stream name -> position
job_already_completed = status not in non_terminal_states
state = LogState.TAILING if wait_for_completion and not job_already_completed else LogState.COMPLETE
# The loop below implements a state machine that alternates between checking the job status and
# reading whatever is available in the logs at this point. Note, that if we were called with
# wait_for_completion == False, we never check the job status.
#
# If wait_for_completion == TRUE and job is not completed, the initial state is TAILING
# If wait_for_completion == FALSE, the initial state is COMPLETE
# (doesn't matter if the job really is complete).
#
# The state table:
#
# STATE ACTIONS CONDITION NEW STATE
# ---------------- ---------------- ----------------- ----------------
# TAILING Read logs, Pause, Get status Job complete JOB_COMPLETE
# Else TAILING
# JOB_COMPLETE Read logs, Pause Any COMPLETE
# COMPLETE Read logs, Exit N/A
#
# Notes:
# - The JOB_COMPLETE state forces us to do an extra pause and read any items that
# got to Cloudwatch after the job was marked complete.
last_describe_job_call = time.monotonic()
last_description = description
while True:
time.sleep(check_interval)
sec += check_interval
state, last_description, last_describe_job_call = self.describe_training_job_with_log(
job_name,
positions,
stream_names,
instance_count,
state,
last_description,
last_describe_job_call,
)
if state == LogState.COMPLETE:
break
if max_ingestion_time and sec > max_ingestion_time:
# ensure that the job gets killed if the max ingestion time is exceeded
raise AirflowException(f"SageMaker job took more than {max_ingestion_time} seconds")
if wait_for_completion:
status = last_description["TrainingJobStatus"]
if status in failed_states:
reason = last_description.get("FailureReason", "(No reason provided)")
raise AirflowException(f"Error training {job_name}: {status} Reason: {reason}")
billable_seconds = SageMakerHook.count_billable_seconds(
training_start_time=last_description["TrainingStartTime"],
training_end_time=last_description["TrainingEndTime"],
instance_count=instance_count,
)
self.log.info("Billable seconds: %d", billable_seconds)
[docs] def list_training_jobs(
self, name_contains: str | None = None, max_results: int | None = None, **kwargs
) -> list[dict]:
"""
Call boto3's ``list_training_jobs``.
The training job name and max results are configurable via arguments.
Other arguments are not, and should be provided via kwargs. Note that
boto3 expects these in CamelCase, for example:
.. code-block:: python
list_training_jobs(name_contains="myjob", StatusEquals="Failed")
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.list_training_jobs`
:param name_contains: (optional) partial name to match
:param max_results: (optional) maximum number of results to return. None returns infinite results
:param kwargs: (optional) kwargs to boto3's list_training_jobs method
:return: results of the list_training_jobs request
"""
config, max_results = self._preprocess_list_request_args(name_contains, max_results, **kwargs)
list_training_jobs_request = partial(self.get_conn().list_training_jobs, **config)
results = self._list_request(
list_training_jobs_request, "TrainingJobSummaries", max_results=max_results
)
return results
[docs] def list_processing_jobs(self, **kwargs) -> list[dict]:
"""
Call boto3's `list_processing_jobs`.
All arguments should be provided via kwargs. Note that boto3 expects
these in CamelCase, for example:
.. code-block:: python
list_processing_jobs(NameContains="myjob", StatusEquals="Failed")
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.list_processing_jobs`
:param kwargs: (optional) kwargs to boto3's list_training_jobs method
:return: results of the list_processing_jobs request
"""
list_processing_jobs_request = partial(self.get_conn().list_processing_jobs, **kwargs)
results = self._list_request(
list_processing_jobs_request, "ProcessingJobSummaries", max_results=kwargs.get("MaxResults")
)
return results
def _preprocess_list_request_args(
self, name_contains: str | None = None, max_results: int | None = None, **kwargs
) -> tuple[dict[str, Any], int | None]:
"""
Preprocess arguments for boto3's ``list_*`` methods.
It will turn arguments name_contains and max_results as boto3 compliant
CamelCase format. This method also makes sure that these two arguments
are only set once.
:param name_contains: boto3 function with arguments
:param max_results: the result key to iterate over
:param kwargs: (optional) kwargs to boto3's list_* method
:return: Tuple with config dict to be passed to boto3's list_* method
and max_results parameter
"""
config = {}
if name_contains:
if "NameContains" in kwargs:
raise AirflowException("Either name_contains or NameContains can be provided, not both.")
config["NameContains"] = name_contains
if "MaxResults" in kwargs and kwargs["MaxResults"] is not None:
if max_results:
raise AirflowException("Either max_results or MaxResults can be provided, not both.")
# Unset MaxResults, we'll use the SageMakerHook's internal method for iteratively fetching results
max_results = kwargs["MaxResults"]
del kwargs["MaxResults"]
config.update(kwargs)
return config, max_results
def _list_request(
self, partial_func: Callable, result_key: str, max_results: int | None = None
) -> list[dict]:
"""
Process a list request to produce results.
All AWS boto3 ``list_*`` requests return results in batches, and if the
key "NextToken" is contained in the result, there are more results to
fetch. The default AWS batch size is 10, and configurable up to 100.
This function iteratively loads all results (or up to a given maximum).
Each boto3 ``list_*`` function returns the results in a list with a
different name. The key of this structure must be given to iterate over
the results, e.g. "TransformJobSummaries" for ``list_transform_jobs()``.
:param partial_func: boto3 function with arguments
:param result_key: the result key to iterate over
:param max_results: maximum number of results to return (None = infinite)
:return: Results of the list_* request
"""
sagemaker_max_results = 100 # Fixed number set by AWS
results: list[dict] = []
next_token = None
while True:
kwargs = {}
if next_token is not None:
kwargs["NextToken"] = next_token
if max_results is None:
kwargs["MaxResults"] = sagemaker_max_results
else:
kwargs["MaxResults"] = min(max_results - len(results), sagemaker_max_results)
response = partial_func(**kwargs)
self.log.debug("Fetched %s results.", len(response[result_key]))
results.extend(response[result_key])
if "NextToken" not in response or (max_results is not None and len(results) == max_results):
# Return when there are no results left (no NextToken) or when we've reached max_results.
return results
else:
next_token = response["NextToken"]
@staticmethod
def _name_matches_pattern(
processing_job_name: str,
found_name: str,
job_name_suffix: str | None = None,
) -> bool:
return re.fullmatch(f"{processing_job_name}({job_name_suffix})?", found_name) is not None
[docs] def count_processing_jobs_by_name(
self,
processing_job_name: str,
job_name_suffix: str | None = None,
throttle_retry_delay: int = 2,
retries: int = 3,
) -> int:
"""
Get the number of processing jobs found with the provided name prefix.
:param processing_job_name: The prefix to look for.
:param job_name_suffix: The optional suffix which may be appended to deduplicate an existing job name.
:param throttle_retry_delay: Seconds to wait if a ThrottlingException is hit.
:param retries: The max number of times to retry.
:returns: The number of processing jobs that start with the provided prefix.
"""
try:
jobs = self.get_conn().list_processing_jobs(NameContains=processing_job_name)
# We want to make sure the job name starts with the provided name, not just contains it.
matching_jobs = [
job["ProcessingJobName"]
for job in jobs["ProcessingJobSummaries"]
if self._name_matches_pattern(processing_job_name, job["ProcessingJobName"], job_name_suffix)
]
return len(matching_jobs)
except ClientError as e:
if e.response["Error"]["Code"] == "ResourceNotFound":
# No jobs found with that name. This is good, return 0.
return 0
if e.response["Error"]["Code"] == "ThrottlingException" and retries:
# If we hit a ThrottlingException, back off a little and try again.
time.sleep(throttle_retry_delay)
return self.count_processing_jobs_by_name(
processing_job_name, job_name_suffix, throttle_retry_delay * 2, retries - 1
)
raise
[docs] def delete_model(self, model_name: str):
"""
Delete a SageMaker model.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.delete_model`
:param model_name: name of the model
"""
try:
self.get_conn().delete_model(ModelName=model_name)
except Exception as general_error:
self.log.error("Failed to delete model, error: %s", general_error)
raise
[docs] def describe_pipeline_exec(self, pipeline_exec_arn: str, verbose: bool = False):
"""
Get info about a SageMaker pipeline execution.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.describe_pipeline_execution`
- :external+boto3:py:meth:`SageMaker.Client.list_pipeline_execution_steps`
:param pipeline_exec_arn: arn of the pipeline execution
:param verbose: Whether to log details about the steps status in the pipeline execution
"""
if verbose:
res = self.conn.list_pipeline_execution_steps(PipelineExecutionArn=pipeline_exec_arn)
count_by_state = Counter(s["StepStatus"] for s in res["PipelineExecutionSteps"])
running_steps = [
s["StepName"] for s in res["PipelineExecutionSteps"] if s["StepStatus"] == "Executing"
]
self.log.info("state of the pipeline steps: %s", count_by_state)
self.log.info("steps currently in progress: %s", running_steps)
return self.conn.describe_pipeline_execution(PipelineExecutionArn=pipeline_exec_arn)
[docs] def start_pipeline(
self,
pipeline_name: str,
display_name: str = "airflow-triggered-execution",
pipeline_params: dict | None = None,
) -> str:
"""
Start a new execution for a SageMaker pipeline.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.start_pipeline_execution`
:param pipeline_name: Name of the pipeline to start (this is _not_ the ARN).
:param display_name: The name this pipeline execution will have in the UI. Doesn't need to be unique.
:param pipeline_params: Optional parameters for the pipeline.
All parameters supplied need to already be present in the pipeline definition.
:return: the ARN of the pipeline execution launched.
"""
formatted_params = format_tags(pipeline_params, key_label="Name")
try:
res = self.conn.start_pipeline_execution(
PipelineName=pipeline_name,
PipelineExecutionDisplayName=display_name,
PipelineParameters=formatted_params,
)
except ClientError as ce:
self.log.error("Failed to start pipeline execution, error: %s", ce)
raise
return res["PipelineExecutionArn"]
[docs] def stop_pipeline(
self,
pipeline_exec_arn: str,
fail_if_not_running: bool = False,
) -> str:
"""
Stop SageMaker pipeline execution.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.stop_pipeline_execution`
:param pipeline_exec_arn: Amazon Resource Name (ARN) of the pipeline execution.
It's the ARN of the pipeline itself followed by "/execution/" and an id.
:param fail_if_not_running: This method will raise an exception if the pipeline we're trying to stop
is not in an "Executing" state when the call is sent (which would mean that the pipeline is
already either stopping or stopped).
Note that setting this to True will raise an error if the pipeline finished successfully before it
was stopped.
:return: Status of the pipeline execution after the operation.
One of 'Executing'|'Stopping'|'Stopped'|'Failed'|'Succeeded'.
"""
for retries in reversed(range(5)):
try:
self.conn.stop_pipeline_execution(PipelineExecutionArn=pipeline_exec_arn)
except ClientError as ce:
# this can happen if the pipeline was transitioning between steps at that moment
if ce.response["Error"]["Code"] == "ConflictException" and retries:
self.log.warning(
"Got a conflict exception when trying to stop the pipeline, "
"retrying %s more times. Error was: %s",
retries,
ce,
)
time.sleep(0.3) # error is due to a race condition, so it should be very transient
else:
# we have to rely on the message to catch the right error here, because its type
# (ValidationException) is shared with other kinds of errors (e.g. badly formatted ARN)
if (
not fail_if_not_running
and "Only pipelines with 'Executing' status can be stopped"
in ce.response["Error"]["Message"]
):
self.log.warning("Cannot stop pipeline execution, as it was not running: %s", ce)
break
else:
self.log.error(ce)
raise
else:
break
res = self.describe_pipeline_exec(pipeline_exec_arn)
return res["PipelineExecutionStatus"]
[docs] def create_model_package_group(self, package_group_name: str, package_group_desc: str = "") -> bool:
"""
Create a Model Package Group if it does not already exist.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.create_model_package_group`
:param package_group_name: Name of the model package group to create if not already present.
:param package_group_desc: Description of the model package group, if it was to be created (optional).
:return: True if the model package group was created, False if it already existed.
"""
try:
res = self.conn.create_model_package_group(
ModelPackageGroupName=package_group_name,
ModelPackageGroupDescription=package_group_desc,
)
self.log.info(
"Created new Model Package Group with name %s (ARN: %s)",
package_group_name,
res["ModelPackageGroupArn"],
)
return True
except ClientError as e:
# ValidationException can also happen if the package group name contains invalid char,
# so we have to look at the error message too
if e.response["Error"]["Code"] == "ValidationException" and e.response["Error"][
"Message"
].startswith("Model Package Group already exists"):
# log msg only so it doesn't look like an error
self.log.info("%s", e.response["Error"]["Message"])
return False
else:
self.log.error("Error when trying to create Model Package Group: %s", e)
raise
def _describe_auto_ml_job(self, job_name: str):
res = self.conn.describe_auto_ml_job(AutoMLJobName=job_name)
self.log.info("%s's current step: %s", job_name, res["AutoMLJobSecondaryStatus"])
return res
[docs] def create_auto_ml_job(
self,
job_name: str,
s3_input: str,
target_attribute: str,
s3_output: str,
role_arn: str,
compressed_input: bool = False,
time_limit: int | None = None,
autodeploy_endpoint_name: str | None = None,
extras: dict | None = None,
wait_for_completion: bool = True,
check_interval: int = 30,
) -> dict | None:
"""
Create an auto ML job to predict the given column.
The learning input is based on data provided through S3 , and the output
is written to the specified S3 location.
.. seealso::
- :external+boto3:py:meth:`SageMaker.Client.create_auto_ml_job`
:param job_name: Name of the job to create, needs to be unique within the account.
:param s3_input: The S3 location (folder or file) where to fetch the data.
By default, it expects csv with headers.
:param target_attribute: The name of the column containing the values to predict.
:param s3_output: The S3 folder where to write the model artifacts. Must be 128 characters or fewer.
:param role_arn: The ARN or the IAM role to use when interacting with S3.
Must have read access to the input, and write access to the output folder.
:param compressed_input: Set to True if the input is gzipped.
:param time_limit: The maximum amount of time in seconds to spend training the model(s).
:param autodeploy_endpoint_name: If specified, the best model will be deployed to an endpoint with
that name. No deployment made otherwise.
:param extras: Use this dictionary to set any variable input variable for job creation that is not
offered through the parameters of this function. The format is described in:
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/sagemaker.html#SageMaker.Client.create_auto_ml_job
:param wait_for_completion: Whether to wait for the job to finish before returning. Defaults to True.
:param check_interval: Interval in seconds between 2 status checks when waiting for completion.
:returns: Only if waiting for completion, a dictionary detailing the best model. The structure is that
of the "BestCandidate" key in:
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/sagemaker.html#SageMaker.Client.describe_auto_ml_job
"""
input_data = [
{
"DataSource": {"S3DataSource": {"S3DataType": "S3Prefix", "S3Uri": s3_input}},
"TargetAttributeName": target_attribute,
},
]
params_dict = {
"AutoMLJobName": job_name,
"InputDataConfig": input_data,
"OutputDataConfig": {"S3OutputPath": s3_output},
"RoleArn": role_arn,
}
if compressed_input:
input_data[0]["CompressionType"] = "Gzip"
if time_limit:
params_dict.update(
{"AutoMLJobConfig": {"CompletionCriteria": {"MaxAutoMLJobRuntimeInSeconds": time_limit}}}
)
if autodeploy_endpoint_name:
params_dict.update({"ModelDeployConfig": {"EndpointName": autodeploy_endpoint_name}})
if extras:
params_dict.update(extras)
# returns the job ARN, but we don't need it because we access it by its name
self.conn.create_auto_ml_job(**params_dict)
if wait_for_completion:
res = self.check_status(
job_name,
"AutoMLJobStatus",
# cannot pass the function directly because the parameter needs to be named
self._describe_auto_ml_job,
check_interval,
)
if "BestCandidate" in res:
return res["BestCandidate"]
return None
@staticmethod
[docs] def count_billable_seconds(
training_start_time: datetime, training_end_time: datetime, instance_count: int
) -> int:
billable_time = (training_end_time - training_start_time) * instance_count
return int(billable_time.total_seconds()) + 1
[docs] async def describe_training_job_async(self, job_name: str) -> dict[str, Any]:
"""
Return the training job info associated with the name.
:param job_name: the name of the training job
"""
async with self.async_conn as client:
response: dict[str, Any] = await client.describe_training_job(TrainingJobName=job_name)
return response
[docs] async def describe_training_job_with_log_async(
self,
job_name: str,
positions: dict[str, Any],
stream_names: list[str],
instance_count: int,
state: int,
last_description: dict[str, Any],
last_describe_job_call: float,
) -> tuple[int, dict[str, Any], float]:
"""
Return the training job info associated with job_name and print CloudWatch logs.
:param job_name: name of the job to check status
:param positions: A list of pairs of (timestamp, skip) which represents the last record
read from each stream.
:param stream_names: A list of the log stream names. The position of the stream in this list is
the stream number.
:param instance_count: Count of the instance created for the job initially
:param state: log state
:param last_description: Latest description of the training job
:param last_describe_job_call: previous job called time
"""
log_group = "/aws/sagemaker/TrainingJobs"
if len(stream_names) < instance_count:
logs_hook = AwsLogsHook(aws_conn_id=self.aws_conn_id, region_name=self.region_name)
streams = await logs_hook.describe_log_streams_async(
log_group=log_group,
stream_prefix=job_name + "/",
order_by="LogStreamName",
count=instance_count,
)
stream_names = [s["logStreamName"] for s in streams["logStreams"]] if streams else []
positions.update([(s, Position(timestamp=0, skip=0)) for s in stream_names if s not in positions])
if len(stream_names) > 0:
async for idx, event in self.get_multi_stream(log_group, stream_names, positions):
self.log.info(event["message"])
ts, count = positions[stream_names[idx]]
if event["timestamp"] == ts:
positions[stream_names[idx]] = Position(timestamp=ts, skip=count + 1)
else:
positions[stream_names[idx]] = Position(timestamp=event["timestamp"], skip=1)
if state == LogState.COMPLETE:
return state, last_description, last_describe_job_call
if state == LogState.JOB_COMPLETE:
state = LogState.COMPLETE
elif time.time() - last_describe_job_call >= 30:
description = await self.describe_training_job_async(job_name)
last_describe_job_call = time.time()
if await sync_to_async(secondary_training_status_changed)(description, last_description):
self.log.info(
await sync_to_async(secondary_training_status_message)(description, last_description)
)
last_description = description
status = description["TrainingJobStatus"]
if status not in self.non_terminal_states:
state = LogState.JOB_COMPLETE
return state, last_description, last_describe_job_call
[docs] async def get_multi_stream(
self, log_group: str, streams: list[str], positions: dict[str, Any]
) -> AsyncGenerator[Any, tuple[int, Any | None]]:
"""
Iterate over the available events coming and interleaving the events from each stream so they're yielded in timestamp order.
:param log_group: The name of the log group.
:param streams: A list of the log stream names. The position of the stream in this list is
the stream number.
:param positions: A list of pairs of (timestamp, skip) which represents the last record
read from each stream.
"""
positions = positions or {s: Position(timestamp=0, skip=0) for s in streams}
events: list[Any | None] = []
logs_hook = AwsLogsHook(aws_conn_id=self.aws_conn_id, region_name=self.region_name)
event_iters = [
logs_hook.get_log_events_async(log_group, s, positions[s].timestamp, positions[s].skip)
for s in streams
]
for event_stream in event_iters:
if not event_stream:
events.append(None)
continue
try:
events.append(await event_stream.__anext__())
except StopAsyncIteration:
events.append(None)
while any(events):
i = argmin(events, lambda x: x["timestamp"] if x else 9999999999) or 0
yield i, events[i]
try:
events[i] = await event_iters[i].__anext__()
except StopAsyncIteration:
events[i] = None