Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi merged PR #38693:
URL: https://github.com/apache/airflow/pull/38693


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1556213468


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +98,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param ensure_unique_job_name: If set to true, operator will check whether 
a model customization
+job already exists for the name in the config and append the current 
timestamp if there is a
+name conflict. (Default: True)
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks. 
(default: 120)
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion. (default: 75)
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields(
+"job_name",
+"custom_model_name",
+"role_arn",
+"base_model_id",
+"hyperparameters",
+"ensure_unique_job_name",
+"customization_job_kwargs",
+)
+
+def __init__(
+self,
+job_name: str,
+custom_model_name: str,
+role_arn: str,
+base_model_id: str,
+training_data_uri: str,
+output_data_uri: str,
+hyperparameters: dict[str, str],
+ensure_unique_job_name: bool = True,
+customization_job_kwargs: dict[str, Any] | None = None,
+wait_for_completion: bool = True,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs,
+):
+super().__init__(**kwargs)
+self.wait_for_completion = wait_for_completion
+self.waiter_delay = waiter_delay
+self.waiter_max_attempts = waiter_max_attempts
+self.deferrable = deferrable
+
+self.job_name = job_name
+self.custom_model_name = custom_model_name
+self.role_arn = role_arn
+self.base_model_id = base_model_id
+self.training_data_config = {"s3Uri": training_data_uri}
+self.output_data_config = {"s3Uri": output_data_uri}
+self.hyperparameters = hyperparameters
+self.ensure_unique_job_name = ensure_unique_job_name
+self.customization_job_kwargs = customization_job_kwargs or {}
+
+self.valid_action_if_job_exists: set[str] = {"timestamp", "fail"}
+
+def execute_complete(self, context: Context, event: dict[str, Any] | None 
= None) -> str:
+event = validate_execute_complete_event(event)
+
+if event["status"] != "success":
+raise AirflowException(f"Error while running job: {event}")
+
+self.log.info("Bedrock model customization job `%s` complete.", 
self.job_name)
+ 

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1556206876


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +98,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param ensure_unique_job_name: If set to true, operator will check whether 
a model customization
+job already exists for the name in the config and append the current 
timestamp if there is a
+name conflict. (Default: True)
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks. 
(default: 120)
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion. (default: 75)
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields(
+"job_name",
+"custom_model_name",
+"role_arn",
+"base_model_id",
+"hyperparameters",
+"ensure_unique_job_name",
+"customization_job_kwargs",
+)
+
+def __init__(
+self,
+job_name: str,
+custom_model_name: str,
+role_arn: str,
+base_model_id: str,
+training_data_uri: str,
+output_data_uri: str,
+hyperparameters: dict[str, str],
+ensure_unique_job_name: bool = True,
+customization_job_kwargs: dict[str, Any] | None = None,
+wait_for_completion: bool = True,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs,
+):
+super().__init__(**kwargs)
+self.wait_for_completion = wait_for_completion
+self.waiter_delay = waiter_delay
+self.waiter_max_attempts = waiter_max_attempts
+self.deferrable = deferrable
+
+self.job_name = job_name
+self.custom_model_name = custom_model_name
+self.role_arn = role_arn
+self.base_model_id = base_model_id
+self.training_data_config = {"s3Uri": training_data_uri}
+self.output_data_config = {"s3Uri": output_data_uri}
+self.hyperparameters = hyperparameters
+self.ensure_unique_job_name = ensure_unique_job_name
+self.customization_job_kwargs = customization_job_kwargs or {}
+
+self.valid_action_if_job_exists: set[str] = {"timestamp", "fail"}
+
+def execute_complete(self, context: Context, event: dict[str, Any] | None 
= None) -> str:
+event = validate_execute_complete_event(event)
+
+if event["status"] != "success":
+raise AirflowException(f"Error while running job: {event}")
+
+self.log.info("Bedrock model customization job `%s` complete.", 
self.job_name)
+ 

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1556206876


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +98,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param ensure_unique_job_name: If set to true, operator will check whether 
a model customization
+job already exists for the name in the config and append the current 
timestamp if there is a
+name conflict. (Default: True)
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks. 
(default: 120)
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion. (default: 75)
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields(
+"job_name",
+"custom_model_name",
+"role_arn",
+"base_model_id",
+"hyperparameters",
+"ensure_unique_job_name",
+"customization_job_kwargs",
+)
+
+def __init__(
+self,
+job_name: str,
+custom_model_name: str,
+role_arn: str,
+base_model_id: str,
+training_data_uri: str,
+output_data_uri: str,
+hyperparameters: dict[str, str],
+ensure_unique_job_name: bool = True,
+customization_job_kwargs: dict[str, Any] | None = None,
+wait_for_completion: bool = True,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs,
+):
+super().__init__(**kwargs)
+self.wait_for_completion = wait_for_completion
+self.waiter_delay = waiter_delay
+self.waiter_max_attempts = waiter_max_attempts
+self.deferrable = deferrable
+
+self.job_name = job_name
+self.custom_model_name = custom_model_name
+self.role_arn = role_arn
+self.base_model_id = base_model_id
+self.training_data_config = {"s3Uri": training_data_uri}
+self.output_data_config = {"s3Uri": output_data_uri}
+self.hyperparameters = hyperparameters
+self.ensure_unique_job_name = ensure_unique_job_name
+self.customization_job_kwargs = customization_job_kwargs or {}
+
+self.valid_action_if_job_exists: set[str] = {"timestamp", "fail"}
+
+def execute_complete(self, context: Context, event: dict[str, Any] | None 
= None) -> str:
+event = validate_execute_complete_event(event)
+
+if event["status"] != "success":
+raise AirflowException(f"Error while running job: {event}")
+
+self.log.info("Bedrock model customization job `%s` complete.", 
self.job_name)
+ 

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1556199352


##
airflow/providers/amazon/aws/sensors/bedrock.py:
##
@@ -0,0 +1,111 @@
+#
+# 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
+
+from typing import TYPE_CHECKING, Any, Sequence
+
+from airflow.configuration import conf
+from airflow.providers.amazon.aws.sensors.base_aws import AwsBaseSensor
+from airflow.providers.amazon.aws.triggers.bedrock import 
BedrockCustomizeModelCompletedTrigger
+from airflow.providers.amazon.aws.utils.mixins import aws_template_fields
+
+if TYPE_CHECKING:
+from airflow.utils.context import Context
+
+from airflow.exceptions import AirflowException, AirflowSkipException
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+
+
+class BedrockCustomizeModelCompletedSensor(AwsBaseSensor[BedrockHook]):
+"""
+Poll the state of the model customization job until it reaches a terminal 
state; fails if the job fails.
+
+.. seealso::
+For more information on how to use this sensor, take a look at the 
guide:
+:ref:`howto/sensor:BedrockCustomizeModelCompletedSensor`
+
+
+:param job_name: The name of the Bedrock model customization job.
+
+:param deferrable: If True, the sensor will operate in deferrable mode. 
This mode requires aiobotocore
+module to be installed.
+(default: False, but can be overridden in config file by setting 
default_deferrable to True)
+:param max_retries: Number of times before returning the current state. 
(default: 75)
+:param poke_interval: Polling period in seconds to check for the status of 
the job. (default: 120)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+INTERMEDIATE_STATES = ("InProgress",)
+FAILURE_STATES = ("Failed", "Stopping", "Stopped")
+SUCCESS_STATES = ("Completed",)
+FAILURE_MESSAGE = "Bedrock model customization job sensor failed."
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields("job_name")
+ui_color = "#66c3ff"
+
+def __init__(
+self,
+*,
+job_name: str,
+max_retries: int = 75,
+poke_interval: int = 120,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs: Any,
+) -> None:
+super().__init__(**kwargs)
+self.job_name = job_name
+self.poke_interval = poke_interval
+self.max_retries = max_retries
+self.deferrable = deferrable
+
+def execute(self, context: Context) -> Any:
+if self.deferrable:
+self.defer(
+trigger=BedrockCustomizeModelCompletedTrigger(
+job_name=self.job_name,
+waiter_delay=int(self.poke_interval),
+waiter_max_attempts=self.max_retries,
+aws_conn_id=self.aws_conn_id,
+),
+method_name="poke",
+)
+else:
+super().execute(context=context)
+
+def poke(self, context: Context) -> bool:
+state = self.hook.get_customize_model_job_state(self.job_name)
+
+if state in self.FAILURE_STATES:
+# TODO: remove this if block when min_airflow_version is set to 
higher than 2.7.1
+if self.soft_fail:
+raise AirflowSkipException(self.FAILURE_MESSAGE)
+raise 

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1556195811


##
airflow/providers/amazon/aws/hooks/bedrock.py:
##
@@ -19,6 +19,37 @@
 from airflow.providers.amazon.aws.hooks.base_aws import AwsBaseHook
 
 
+class BedrockHook(AwsBaseHook):
+"""
+Interact with Amazon Bedrock.
+
+Provide thin wrapper around 
:external+boto3:py:class:`boto3.client("bedrock") `.
+
+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`
+"""
+
+client_type = "bedrock"
+
+def __init__(self, *args, **kwargs) -> None:
+kwargs["client_type"] = self.client_type
+super().__init__(*args, **kwargs)
+
+def _get_job_by_name(self, job_name: str):
+return self.conn.get_model_customization_job(jobIdentifier=job_name)
+
+def get_customize_model_job_state(self, job_name: str) -> str:
+state = self._get_job_by_name(job_name)["status"]
+self.log.info("Job '%s' state: %s", job_name, state)
+return state
+
+def get_job_arn(self, job_name: str) -> str:
+return self._get_job_by_name(job_name)["jobArn"]

Review Comment:
   Yeah, they felt handy and made the Sensor unit tests a little easier to 
mock, but I'll drop them.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

vincbeck commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1556043966


##
airflow/providers/amazon/aws/sensors/bedrock.py:
##
@@ -0,0 +1,111 @@
+#
+# 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
+
+from typing import TYPE_CHECKING, Any, Sequence
+
+from airflow.configuration import conf
+from airflow.providers.amazon.aws.sensors.base_aws import AwsBaseSensor
+from airflow.providers.amazon.aws.triggers.bedrock import 
BedrockCustomizeModelCompletedTrigger
+from airflow.providers.amazon.aws.utils.mixins import aws_template_fields
+
+if TYPE_CHECKING:
+from airflow.utils.context import Context
+
+from airflow.exceptions import AirflowException, AirflowSkipException
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+
+
+class BedrockCustomizeModelCompletedSensor(AwsBaseSensor[BedrockHook]):
+"""
+Poll the state of the model customization job until it reaches a terminal 
state; fails if the job fails.
+
+.. seealso::
+For more information on how to use this sensor, take a look at the 
guide:
+:ref:`howto/sensor:BedrockCustomizeModelCompletedSensor`
+
+
+:param job_name: The name of the Bedrock model customization job.
+
+:param deferrable: If True, the sensor will operate in deferrable mode. 
This mode requires aiobotocore
+module to be installed.
+(default: False, but can be overridden in config file by setting 
default_deferrable to True)
+:param max_retries: Number of times before returning the current state. 
(default: 75)
+:param poke_interval: Polling period in seconds to check for the status of 
the job. (default: 120)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+INTERMEDIATE_STATES = ("InProgress",)
+FAILURE_STATES = ("Failed", "Stopping", "Stopped")
+SUCCESS_STATES = ("Completed",)
+FAILURE_MESSAGE = "Bedrock model customization job sensor failed."
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields("job_name")
+ui_color = "#66c3ff"
+
+def __init__(
+self,
+*,
+job_name: str,
+max_retries: int = 75,
+poke_interval: int = 120,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs: Any,
+) -> None:
+super().__init__(**kwargs)
+self.job_name = job_name
+self.poke_interval = poke_interval
+self.max_retries = max_retries
+self.deferrable = deferrable
+
+def execute(self, context: Context) -> Any:
+if self.deferrable:
+self.defer(
+trigger=BedrockCustomizeModelCompletedTrigger(
+job_name=self.job_name,
+waiter_delay=int(self.poke_interval),
+waiter_max_attempts=self.max_retries,
+aws_conn_id=self.aws_conn_id,
+),
+method_name="poke",
+)
+else:
+super().execute(context=context)
+
+def poke(self, context: Context) -> bool:
+state = self.hook.get_customize_model_job_state(self.job_name)
+
+if state in self.FAILURE_STATES:
+# TODO: remove this if block when min_airflow_version is set to 
higher than 2.7.1
+if self.soft_fail:
+raise AirflowSkipException(self.FAILURE_MESSAGE)
+raise 

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

Taragolis commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r132965


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +98,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param ensure_unique_job_name: If set to true, operator will check whether 
a model customization
+job already exists for the name in the config and append the current 
timestamp if there is a
+name conflict. (Default: True)
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks. 
(default: 120)
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion. (default: 75)
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields(
+"job_name",
+"custom_model_name",
+"role_arn",
+"base_model_id",
+"hyperparameters",
+"ensure_unique_job_name",
+"customization_job_kwargs",
+)
+
+def __init__(
+self,
+job_name: str,
+custom_model_name: str,
+role_arn: str,
+base_model_id: str,
+training_data_uri: str,
+output_data_uri: str,
+hyperparameters: dict[str, str],
+ensure_unique_job_name: bool = True,
+customization_job_kwargs: dict[str, Any] | None = None,
+wait_for_completion: bool = True,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs,
+):
+super().__init__(**kwargs)
+self.wait_for_completion = wait_for_completion
+self.waiter_delay = waiter_delay
+self.waiter_max_attempts = waiter_max_attempts
+self.deferrable = deferrable
+
+self.job_name = job_name
+self.custom_model_name = custom_model_name
+self.role_arn = role_arn
+self.base_model_id = base_model_id
+self.training_data_config = {"s3Uri": training_data_uri}
+self.output_data_config = {"s3Uri": output_data_uri}
+self.hyperparameters = hyperparameters
+self.ensure_unique_job_name = ensure_unique_job_name
+self.customization_job_kwargs = customization_job_kwargs or {}
+
+self.valid_action_if_job_exists: set[str] = {"timestamp", "fail"}
+
+def execute_complete(self, context: Context, event: dict[str, Any] | None 
= None) -> str:
+event = validate_execute_complete_event(event)
+
+if event["status"] != "success":
+raise AirflowException(f"Error while running job: {event}")
+
+self.log.info("Bedrock model customization job `%s` complete.", 
self.job_name)

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on PR #38693:
URL: https://github.com/apache/airflow/pull/38693#issuecomment-2040697189

   @o-nikolas Reworked the `while True` so it won't repeat by default


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on PR #38693:
URL: https://github.com/apache/airflow/pull/38693#issuecomment-2040644595

   > @o-nikolas @Taragolis @syedahsn - I refactored how we handle a name 
conflict. Can you please have another look and re-approve (or comment) when you 
get time?
   
   The code looks functionally correct to me. Although the `while True` loop 
worries me a bit, it's very easy for a small bug to get introduced into 
something like that and then suddenly be stuck in an infinite loop.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1554297267


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   PR to move them all into their own files is here:  
https://github.com/apache/airflow/pull/38785



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on PR #38693:
URL: https://github.com/apache/airflow/pull/38693#issuecomment-2040573795

   @o-nikolas @Taragolis @syedahsn  - I refactored how we handle a name 
conflict.  Can you please have another look and re-approve (or comment) when 
you get time?


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1554216885


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),

Review Comment:
   Alright, I decided to tweak this.  Let's leave this comment open as a 
failsafe against merging for now.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1554213131


##
tests/providers/amazon/aws/waiters/test_bedrock.py:
##
@@ -0,0 +1,71 @@
+# 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
+
+from unittest import mock
+
+import boto3
+import botocore
+import pytest
+
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.sensors.bedrock import 
BedrockCustomizeModelCompletedSensor
+
+
+class TestBedrockCustomWaiters:
+def test_service_waiters(self):
+assert "model_customization_job_complete" in 
BedrockHook().list_waiters()
+
+
+class TestBedrockCustomWaitersBase:

Review Comment:
   Yeah, the autouse made sense after your previous message. My last one was 
just referencing that you have a base class only used once, but as suspected 
and you've confirmed, more uses are coming. So it's all good :) 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1554074878


##
tests/providers/amazon/aws/waiters/test_bedrock.py:
##
@@ -0,0 +1,71 @@
+# 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
+
+from unittest import mock
+
+import boto3
+import botocore
+import pytest
+
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.sensors.bedrock import 
BedrockCustomizeModelCompletedSensor
+
+
+class TestBedrockCustomWaiters:
+def test_service_waiters(self):
+assert "model_customization_job_complete" in 
BedrockHook().list_waiters()
+
+
+class TestBedrockCustomWaitersBase:

Review Comment:
   I think maybe the autouse is throwing you off.  It creates `self.client` 
which is then inherited/used/modified in the mock_get_job in 
TestModelCustomizationJobCompleteWaiter.
   
   In addition, it will also be used by 
TestProvisionThroughputJobCompleteWaiter which I will have a PR up for either 
later today or Monday.   I know we generally don't create the helper until it 
is used a second time, but I got ahead of myself on this one, but I assure you 
it will get reused :P 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1554032096


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),

Review Comment:
   Other than this dangling thread, it looks good to me, approving :+1: 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1554031431


##
tests/providers/amazon/aws/waiters/test_bedrock.py:
##
@@ -0,0 +1,71 @@
+# 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
+
+from unittest import mock
+
+import boto3
+import botocore
+import pytest
+
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.sensors.bedrock import 
BedrockCustomizeModelCompletedSensor
+
+
+class TestBedrockCustomWaiters:
+def test_service_waiters(self):
+assert "model_customization_job_complete" in 
BedrockHook().list_waiters()
+
+
+class TestBedrockCustomWaitersBase:

Review Comment:
   Ahh I see, so this base class will be used by future waiter test classes?



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1553998433


##
airflow/providers/amazon/aws/triggers/bedrock.py:
##
@@ -0,0 +1,61 @@
+# 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
+
+from typing import TYPE_CHECKING
+
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.triggers.base import AwsBaseWaiterTrigger
+
+if TYPE_CHECKING:
+from airflow.providers.amazon.aws.hooks.base_aws import AwsGenericHook
+
+
+class BedrockCustomizeModelCompletedTrigger(AwsBaseWaiterTrigger):
+"""
+Trigger when a Bedrock model customization job is complete.
+
+:param job_name: The name of the Bedrock model customization job.
+:param waiter_delay: The amount of time in seconds to wait between 
attempts. (default: 120)
+:param waiter_max_attempts: The maximum number of attempts to be made. 
(default: 75)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+"""
+
+def __init__(
+self,
+*,
+job_name: str,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+aws_conn_id: str | None = None,
+) -> None:
+super().__init__(
+serialized_fields={"job_name": job_name},

Review Comment:
   Looks right to me.   I'll leave this open for now in case anyone else chimes 
in.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1553990772


##
airflow/providers/amazon/aws/sensors/bedrock.py:
##
@@ -0,0 +1,111 @@
+#
+# 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
+
+from typing import TYPE_CHECKING, Any, Sequence
+
+from airflow.configuration import conf
+from airflow.providers.amazon.aws.sensors.base_aws import AwsBaseSensor
+from airflow.providers.amazon.aws.triggers.bedrock import 
BedrockCustomizeModelCompletedTrigger
+from airflow.providers.amazon.aws.utils.mixins import aws_template_fields
+
+if TYPE_CHECKING:
+from airflow.utils.context import Context
+
+from airflow.exceptions import AirflowException, AirflowSkipException
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+
+
+class BedrockCustomizeModelCompletedSensor(AwsBaseSensor[BedrockHook]):
+"""
+Poll the state of the model customization job until it reaches a terminal 
state; fails if the job fails.
+
+.. seealso::
+For more information on how to use this sensor, take a look at the 
guide:
+:ref:`howto/sensor:BedrockCustomizeModelCompletedSensor`
+
+
+:param job_name: The name of the Bedrock model customization job.
+
+:param deferrable: If True, the sensor will operate in deferrable more. 
This mode requires aiobotocore

Review Comment:
   HAH!   That was actually copypasta fro SQS and I didn't catch it.  I'll fix 
this one now and fix SQS in another PR 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock - Model Customization Jobs [airflow]

[via GitHub]

syedahsn commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1553827975


##
airflow/providers/amazon/aws/triggers/bedrock.py:
##
@@ -0,0 +1,61 @@
+# 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
+
+from typing import TYPE_CHECKING
+
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.triggers.base import AwsBaseWaiterTrigger
+
+if TYPE_CHECKING:
+from airflow.providers.amazon.aws.hooks.base_aws import AwsGenericHook
+
+
+class BedrockCustomizeModelCompletedTrigger(AwsBaseWaiterTrigger):
+"""
+Trigger when a Bedrock model customization job is complete.
+
+:param job_name: The name of the Bedrock model customization job.
+:param waiter_delay: The amount of time in seconds to wait between 
attempts. (default: 120)
+:param waiter_max_attempts: The maximum number of attempts to be made. 
(default: 75)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+"""
+
+def __init__(
+self,
+*,
+job_name: str,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+aws_conn_id: str | None = None,
+) -> None:
+super().__init__(
+serialized_fields={"job_name": job_name},

Review Comment:
   It may be just something on my screen, but this looks like its indented 
incorrectly? Not a big deal either way. 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552529331


##
tests/system/providers/amazon/aws/example_bedrock.py:
##
@@ -37,30 +81,84 @@
 ) as dag:
 test_context = sys_test_context_task()
 env_id = test_context["ENV_ID"]
+bucket_name = f"{env_id}-bedrock"
+input_data_s3_key = f"{env_id}/train.jsonl"

Review Comment:
   Good catch, but the filename came straight out of the Bedrock docs page.  I 
hadn't heard of it either and thought it was odd as well, but Google said it is 
a known file format and it works in manual testing.  So I'm reasonably sure 
it's not a typo.
   
   ```
   The key difference [between json and jsonl] is in how they handle multiple 
   JSON objects.  Regular JSON files are typically a single, self-contained 
   structure, while JSON Lines use a line-by-line format, allowing for easier 
   streaming and processing of individual objects.
   ```



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552586854


##
tests/system/providers/amazon/aws/example_bedrock.py:
##
@@ -16,17 +16,61 @@
 # under the License.
 from __future__ import annotations
 
+import json
 from datetime import datetime
 
+from botocore.exceptions import ClientError
+
+from airflow.decorators import task
 from airflow.models.baseoperator import chain
 from airflow.models.dag import DAG
-from airflow.providers.amazon.aws.operators.bedrock import 
BedrockInvokeModelOperator
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.operators.bedrock import (
+BedrockCustomizeModelOperator,
+BedrockInvokeModelOperator,
+)
+from airflow.providers.amazon.aws.operators.s3 import (
+S3CreateBucketOperator,
+S3CreateObjectOperator,
+S3DeleteBucketOperator,
+)
+from airflow.providers.amazon.aws.sensors.bedrock import 
BedrockCustomizeModelCompletedSensor
+from airflow.utils.trigger_rule import TriggerRule
 from tests.system.providers.amazon.aws.utils import SystemTestContextBuilder
 
-sys_test_context_task = SystemTestContextBuilder().build()
+# Externally fetched variables:
+ROLE_ARN_KEY = "ROLE_ARN"
+sys_test_context_task = 
SystemTestContextBuilder().add_variable(ROLE_ARN_KEY).build()
 
 DAG_ID = "example_bedrock"
+
+# Creating a custom model takes nearly two hours. If SKIP_LONG_TASKS is True 
then set

Review Comment:
   I've been putting thought into using a BranchingOperator to just skip them 
entirely if the SKIP_LONG flag is true, rather than assign them an improbable 
trigger_rule.  Then maybe we can figure out some trick to get our dashboard CI 
to run the long version from time to time by setting an envvar or something 
I don't know, needs some thinking, but I think this is a decent compromise for 
the time being.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552583500


##
tests/providers/amazon/aws/waiters/test_bedrock.py:
##
@@ -0,0 +1,71 @@
+# 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
+
+from unittest import mock
+
+import boto3
+import botocore
+import pytest
+
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.sensors.bedrock import 
BedrockCustomizeModelCompletedSensor
+
+
+class TestBedrockCustomWaiters:
+def test_service_waiters(self):
+assert "model_customization_job_complete" in 
BedrockHook().list_waiters()
+
+
+class TestBedrockCustomWaitersBase:

Review Comment:
   I jumped the gun on that a little.   I'm already implementing more waiters 
and they'll be sharing that same fixture (which is used below [on 
L45](https://github.com/apache/airflow/pull/38693/files#diff-9d844f38bae728c36b62862b4197ffc0d54b238cf85d7841b40d7aa8571acf6cR45)
 )to mock the boto client.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552581272


##
tests/system/providers/amazon/aws/example_bedrock.py:
##
@@ -16,17 +16,61 @@
 # under the License.
 from __future__ import annotations
 
+import json
 from datetime import datetime
 
+from botocore.exceptions import ClientError
+
+from airflow.decorators import task
 from airflow.models.baseoperator import chain
 from airflow.models.dag import DAG
-from airflow.providers.amazon.aws.operators.bedrock import 
BedrockInvokeModelOperator
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.operators.bedrock import (
+BedrockCustomizeModelOperator,
+BedrockInvokeModelOperator,
+)
+from airflow.providers.amazon.aws.operators.s3 import (
+S3CreateBucketOperator,
+S3CreateObjectOperator,
+S3DeleteBucketOperator,
+)
+from airflow.providers.amazon.aws.sensors.bedrock import 
BedrockCustomizeModelCompletedSensor
+from airflow.utils.trigger_rule import TriggerRule
 from tests.system.providers.amazon.aws.utils import SystemTestContextBuilder
 
-sys_test_context_task = SystemTestContextBuilder().build()
+# Externally fetched variables:
+ROLE_ARN_KEY = "ROLE_ARN"
+sys_test_context_task = 
SystemTestContextBuilder().add_variable(ROLE_ARN_KEY).build()
 
 DAG_ID = "example_bedrock"
+
+# Creating a custom model takes nearly two hours. If SKIP_LONG_TASKS is True 
then set

Review Comment:
   That's very sad. Okay, this is a nice workaround if that's the case :+1: 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552580948


##
tests/providers/amazon/aws/operators/test_bedrock.py:
##
@@ -18,42 +18,153 @@
 from __future__ import annotations
 
 import json
-from typing import Generator
+from typing import TYPE_CHECKING, Generator
 from unittest import mock
 
 import pytest
 from moto import mock_aws
 
-from airflow.providers.amazon.aws.hooks.bedrock import BedrockRuntimeHook
-from airflow.providers.amazon.aws.operators.bedrock import 
BedrockInvokeModelOperator
-
-MODEL_ID = "meta.llama2-13b-chat-v1"
-PROMPT = "A very important question."
-GENERATED_RESPONSE = "An important answer."
-MOCK_RESPONSE = json.dumps(
-{
-"generation": GENERATED_RESPONSE,
-"prompt_token_count": len(PROMPT),
-"generation_token_count": len(GENERATED_RESPONSE),
-"stop_reason": "stop",
-}
+from airflow.exceptions import AirflowException
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook, 
BedrockRuntimeHook
+from airflow.providers.amazon.aws.operators.bedrock import (
+BedrockCustomizeModelOperator,
+BedrockInvokeModelOperator,
 )
 
-
-@pytest.fixture
-def runtime_hook() -> Generator[BedrockRuntimeHook, None, None]:
-with mock_aws():
-yield BedrockRuntimeHook(aws_conn_id="aws_default")
+if TYPE_CHECKING:
+from airflow.providers.amazon.aws.hooks.base_aws import BaseAwsConnection
 
 
 class TestBedrockInvokeModelOperator:
-@mock.patch.object(BedrockRuntimeHook, "conn")
-def test_invoke_model_prompt_good_combinations(self, mock_conn):
-mock_conn.invoke_model.return_value["body"].read.return_value = 
MOCK_RESPONSE
+def setup_method(self):
+self.model_id = "meta.llama2-13b-chat-v1"
+self.prompt = "A very important question."
+self.generated_response = "An important answer."

Review Comment:
   (Here and above) Yeah, it felt odd having them just hanging there outside of 
any methods.   I'll move them to class-level fields.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552580790


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),
+and "fail"
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks.
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion.
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields(
+"job_name",
+"custom_model_name",
+"role_arn",
+"base_model_id",
+"hyperparameters",
+"check_if_job_exists",
+"action_if_job_exists",
+"customization_job_kwargs",
+)
+
+def __init__(
+self,
+job_name: str,
+custom_model_name: str,
+role_arn: str,
+base_model_id: str,
+training_data_uri: str,
+output_data_uri: str,
+hyperparameters: dict[str, str],
+check_if_job_exists: bool = True,
+action_if_job_exists: str = "timestamp",
+customization_job_kwargs: dict[str, Any] | None = None,
+wait_for_completion: bool = True,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs,
+):
+super().__init__(**kwargs)
+self.wait_for_completion = wait_for_completion
+self.waiter_delay = waiter_delay
+self.waiter_max_attempts = waiter_max_attempts
+self.deferrable = deferrable
+
+self.job_name = job_name
+self.custom_model_name = custom_model_name
+self.role_arn = role_arn
+self.base_model_id = base_model_id
+self.training_data_config = {"s3Uri": training_data_uri}
+self.output_data_config = {"s3Uri": output_data_uri}
+self.hyperparameters = hyperparameters
+self.check_if_job_exists = check_if_job_exists
+self.customization_job_kwargs = customization_job_kwargs or {}
+self.action_if_job_exists = action_if_job_exists.lower()
+
+self.valid_action_if_job_exists: set[str] = {"timestamp", "fail"}
+
+def execute_complete(self, context: Context, event: dict[str, Any] | None 
= None) -> str:
+event = validate_execute_complete_event(event)
+
+if event["status"] != "success":

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552578537


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),

Review Comment:
   I'm happy either way, I don't want to be too prescriptive. But if you were 
going to change it to a bool, I much prefer your approach of 
`ensure_unique_name` (or something like that).
   



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552572554


##
tests/providers/amazon/aws/hooks/test_bedrock.py:
##
@@ -16,7 +16,70 @@
 # under the License.
 from __future__ import annotations
 
-from airflow.providers.amazon.aws.hooks.bedrock import BedrockRuntimeHook
+from unittest import mock
+
+import pytest
+from botocore.exceptions import ClientError
+
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook, 
BedrockRuntimeHook
+
+JOB_NAME = "testJobName"
+EXPECTED_STATUS = "InProgress"
+
+
+@pytest.fixture
+def mock_conn():
+with mock.patch.object(BedrockHook, "conn") as _conn:
+_conn.get_model_customization_job.return_value = {"jobName": JOB_NAME, 
"status": EXPECTED_STATUS}
+yield _conn
+
+
+class TestBedrockHook:
+def setup_method(self):
+self.hook = BedrockHook()
+
+self.validation_exception_error = ClientError(
+error_response={"Error": {"Code": "ValidationException", 
"Message": ""}},
+operation_name="GetModelCustomizationJob",
+)
+
+self.unexpected_exception = ClientError(
+error_response={"Error": {"Code": "ExpiredTokenException", 
"Message": ""}},
+operation_name="GetModelCustomizationJob",
+)

Review Comment:
   Moved to a class-level field.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552560048


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),
+and "fail"
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks.
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion.
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields(
+"job_name",
+"custom_model_name",
+"role_arn",
+"base_model_id",
+"hyperparameters",
+"check_if_job_exists",
+"action_if_job_exists",
+"customization_job_kwargs",
+)
+
+def __init__(
+self,
+job_name: str,
+custom_model_name: str,
+role_arn: str,
+base_model_id: str,
+training_data_uri: str,
+output_data_uri: str,
+hyperparameters: dict[str, str],
+check_if_job_exists: bool = True,
+action_if_job_exists: str = "timestamp",
+customization_job_kwargs: dict[str, Any] | None = None,
+wait_for_completion: bool = True,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs,
+):
+super().__init__(**kwargs)
+self.wait_for_completion = wait_for_completion
+self.waiter_delay = waiter_delay
+self.waiter_max_attempts = waiter_max_attempts
+self.deferrable = deferrable
+
+self.job_name = job_name
+self.custom_model_name = custom_model_name
+self.role_arn = role_arn
+self.base_model_id = base_model_id
+self.training_data_config = {"s3Uri": training_data_uri}
+self.output_data_config = {"s3Uri": output_data_uri}
+self.hyperparameters = hyperparameters
+self.check_if_job_exists = check_if_job_exists
+self.customization_job_kwargs = customization_job_kwargs or {}
+self.action_if_job_exists = action_if_job_exists.lower()
+
+self.valid_action_if_job_exists: set[str] = {"timestamp", "fail"}
+
+def execute_complete(self, context: Context, event: dict[str, Any] | None 
= None) -> str:
+event = validate_execute_complete_event(event)
+
+if event["status"] != "success":
+ 

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552558849


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),

Review Comment:
   Or maybe `ensure_unique_name: bool`?  It returns the job name anyway, do the 
users need to care HOW it will be uniquified?  I think this is a good one 
I'll think it over a bit more.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552557529


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),

Review Comment:
   I've actually been second-guessing this one.A parameter with only two 
possible values sounds a lot like a bool to me, in hindsight.  I can't think of 
a good succinct name for it as a bool though.
   
   To maybe shake some ideas loose, if it stays as a string, I'll add this to 
the description: 
   
   ```
If "timestamp" is used and the job name already exists, the current 
timestamp
will be appended to the name in order to make it unique.
```

But perhaps I should replace `action_if_job_exists: str` (and all the 
associated input validation for it) with `append_timestamp_on_name_conflict: 
bool`?



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552557529


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),

Review Comment:
   UI've actually been second-guessing this one.A parameter with only two 
possible values sounds a lot like a bool to me, in hindsight.  I can't think of 
a good succinct name for it as a bool though.
   
   To maybe shake some ideas loose, if it stays as a string, I'll add this to 
the description: 
   
   ```
If "timestamp" is used and a the job name already exists, the current 
timestamp
will be appended tot he name in order to make it unique.
```

But perhaps I should replace `action_if_job_exists: str` (and all the 
associated input validation for it) with `append_timestamp_on_name_conflict: 
bool`?



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552532744


##
tests/system/providers/amazon/aws/example_bedrock.py:
##
@@ -16,17 +16,61 @@
 # under the License.
 from __future__ import annotations
 
+import json
 from datetime import datetime
 
+from botocore.exceptions import ClientError
+
+from airflow.decorators import task
 from airflow.models.baseoperator import chain
 from airflow.models.dag import DAG
-from airflow.providers.amazon.aws.operators.bedrock import 
BedrockInvokeModelOperator
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook
+from airflow.providers.amazon.aws.operators.bedrock import (
+BedrockCustomizeModelOperator,
+BedrockInvokeModelOperator,
+)
+from airflow.providers.amazon.aws.operators.s3 import (
+S3CreateBucketOperator,
+S3CreateObjectOperator,
+S3DeleteBucketOperator,
+)
+from airflow.providers.amazon.aws.sensors.bedrock import 
BedrockCustomizeModelCompletedSensor
+from airflow.utils.trigger_rule import TriggerRule
 from tests.system.providers.amazon.aws.utils import SystemTestContextBuilder
 
-sys_test_context_task = SystemTestContextBuilder().build()
+# Externally fetched variables:
+ROLE_ARN_KEY = "ROLE_ARN"
+sys_test_context_task = 
SystemTestContextBuilder().add_variable(ROLE_ARN_KEY).build()
 
 DAG_ID = "example_bedrock"
+
+# Creating a custom model takes nearly two hours. If SKIP_LONG_TASKS is True 
then set

Review Comment:
   Negative.   That is the smallest possible job I can come up with (feeding 
the smallest model one datapoint to learn), there is no cancel command, and it 
can not be deleted until it is in a terminal state, whether that is a 
successful build or a failed job.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552529331


##
tests/system/providers/amazon/aws/example_bedrock.py:
##
@@ -37,30 +81,84 @@
 ) as dag:
 test_context = sys_test_context_task()
 env_id = test_context["ENV_ID"]
+bucket_name = f"{env_id}-bedrock"
+input_data_s3_key = f"{env_id}/train.jsonl"

Review Comment:
   Good catch, but the filename came straight out of the Bedrock docs page.  I 
hadn't hard of it either and thought it was odd as well, but Google said it is 
a known file format and it works in manual testing.  So I'm reasonably sure 
it's not a typo.
   
   ```
   The key difference [between json and jsonl] is in how they handle multiple 
   JSON objects.  Regular JSON files are typically a single, self-contained 
   structure, while JSON Lines use a line-by-line format, allowing for easier 
   streaming and processing of individual objects.
   ```



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552529331


##
tests/system/providers/amazon/aws/example_bedrock.py:
##
@@ -37,30 +81,84 @@
 ) as dag:
 test_context = sys_test_context_task()
 env_id = test_context["ENV_ID"]
+bucket_name = f"{env_id}-bedrock"
+input_data_s3_key = f"{env_id}/train.jsonl"

Review Comment:
   Good catch, but the filename came straight out of the Bedrock docs page.  I 
hadn't hard of it either and thought it was odd as well, but Google said it is 
a known file format and it works in manual testing.  So I'm reasonably sure 
it's not a typo.
   
   ```
   The key difference [between json and jsonl] is in how they handle multiple 
JSON objects. Regular JSON files are typically a single, self-contained 
structure, while JSON Lines use a line-by-line format, allowing for easier 
streaming and processing of individual objects.
   ```



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

o-nikolas commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552493564


##
airflow/providers/amazon/aws/hooks/bedrock.py:
##
@@ -16,9 +16,53 @@
 # under the License.
 from __future__ import annotations
 
+from botocore.exceptions import ClientError
+
 from airflow.providers.amazon.aws.hooks.base_aws import AwsBaseHook
 
 
+class BedrockHook(AwsBaseHook):
+"""
+Interact with Amazon Bedrock.
+
+Provide thin wrapper around 
:external+boto3:py:class:`boto3.client("bedrock") `.
+
+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`
+"""
+
+client_type = "bedrock"
+
+def __init__(self, *args, **kwargs) -> None:
+kwargs["client_type"] = self.client_type
+super().__init__(*args, **kwargs)
+
+def _get_job_by_name(self, job_name: str):
+return self.conn.get_model_customization_job(jobIdentifier=job_name)
+
+def get_customize_model_job_state(self, job_name) -> str:

Review Comment:
   Missing type annotation on job_name



##
tests/providers/amazon/aws/operators/test_bedrock.py:
##
@@ -18,42 +18,153 @@
 from __future__ import annotations
 
 import json
-from typing import Generator
+from typing import TYPE_CHECKING, Generator
 from unittest import mock
 
 import pytest
 from moto import mock_aws
 
-from airflow.providers.amazon.aws.hooks.bedrock import BedrockRuntimeHook
-from airflow.providers.amazon.aws.operators.bedrock import 
BedrockInvokeModelOperator
-
-MODEL_ID = "meta.llama2-13b-chat-v1"
-PROMPT = "A very important question."
-GENERATED_RESPONSE = "An important answer."
-MOCK_RESPONSE = json.dumps(
-{
-"generation": GENERATED_RESPONSE,
-"prompt_token_count": len(PROMPT),
-"generation_token_count": len(GENERATED_RESPONSE),
-"stop_reason": "stop",
-}
+from airflow.exceptions import AirflowException
+from airflow.providers.amazon.aws.hooks.bedrock import BedrockHook, 
BedrockRuntimeHook
+from airflow.providers.amazon.aws.operators.bedrock import (
+BedrockCustomizeModelOperator,
+BedrockInvokeModelOperator,
 )
 
-
-@pytest.fixture
-def runtime_hook() -> Generator[BedrockRuntimeHook, None, None]:
-with mock_aws():
-yield BedrockRuntimeHook(aws_conn_id="aws_default")
+if TYPE_CHECKING:
+from airflow.providers.amazon.aws.hooks.base_aws import BaseAwsConnection
 
 
 class TestBedrockInvokeModelOperator:
-@mock.patch.object(BedrockRuntimeHook, "conn")
-def test_invoke_model_prompt_good_combinations(self, mock_conn):
-mock_conn.invoke_model.return_value["body"].read.return_value = 
MOCK_RESPONSE
+def setup_method(self):
+self.model_id = "meta.llama2-13b-chat-v1"
+self.prompt = "A very important question."
+self.generated_response = "An important answer."

Review Comment:
   This can just be done once?



##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,155 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),
+and "fail"
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks.
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion.
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies 

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552442572


##
docs/apache-airflow-providers-amazon/operators/bedrock.rst:
##
@@ -65,6 +65,44 @@ To invoke an Amazon Titan model you would use:
 
 For details on the different formats, see `Inference parameters for foundation 
models 
`__
 
+.. _howto/operator:BedrockCustomizeModelOperator:
+
+Customize an existing Amazon Bedrock Model
+==
+
+To create a fine-tuning job to customize a base model, you can use
+:class:`~airflow.providers.amazon.aws.operators.bedrock.BedrockCustomizeModelOperator`.
+
+Model-customization jobs are asynchronous and the completion time depends on 
the base model
+and the training/validation data size. To monitor the state of the job, you 
can use the
+"model_customization_job_complete" Waiter, the
+:class:`~airflow.providers.amazon.aws.sensors.bedrock.BedrockCustomizeModelCompletedSensor`
 Sensor,
+or the 
:class:`~airflow.providers.amazon.aws.triggersBedrockCustomizeModelCompletedTrigger`
 Trigger.

Review Comment:
   missed a period in the class path, surprised the build-docs test didn't 
catch that.  Will fix.



##
docs/apache-airflow-providers-amazon/operators/bedrock.rst:
##
@@ -65,6 +65,44 @@ To invoke an Amazon Titan model you would use:
 
 For details on the different formats, see `Inference parameters for foundation 
models 
`__
 
+.. _howto/operator:BedrockCustomizeModelOperator:
+
+Customize an existing Amazon Bedrock Model
+==
+
+To create a fine-tuning job to customize a base model, you can use
+:class:`~airflow.providers.amazon.aws.operators.bedrock.BedrockCustomizeModelOperator`.
+
+Model-customization jobs are asynchronous and the completion time depends on 
the base model
+and the training/validation data size. To monitor the state of the job, you 
can use the
+"model_customization_job_complete" Waiter, the
+:class:`~airflow.providers.amazon.aws.sensors.bedrock.BedrockCustomizeModelCompletedSensor`
 Sensor,
+or the 
:class:`~airflow.providers.amazon.aws.triggersBedrockCustomizeModelCompletedTrigger`
 Trigger.
+
+
+.. exampleinclude:: /../../tests/system/providers/amazon/aws/example_bedrock.py
+:language: python
+:dedent: 4
+:start-after: [START howto_operator_customize_model]
+:end-before: [END howto_operator_customize_model]
+
+
+Sensors
+---
+
+.. _howto/sensor:BedrockCustomizeModelCompletedSensor:
+
+Wait for an Amazon Bedrock customize model job
+==
+
+To wait on the state of an AWS CloudFormation stack creation until it reaches 
a terminal state you can use

Review Comment:
   copypasta  :(   Will fix



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552191237


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   Yeah, the list just keeps growing.I don't know how you find time to do 
as much as you do for cleaning up old code.  I think I want to split the 
test_custom_waiters monlith into per-service files for now, then we can worry 
about doing some kind of baseTest class for them later.  That will at least 
make it obvious that they are expected.  But I'll do that in another PR.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

Taragolis commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552138597


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   And I guess Batch has custom waiters into the two places 藍 



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

Taragolis commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1552137707


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   I guess it is started into the custom_waiters, and after that it added into 
the separate one. So many thing need to be improved  



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1550689743


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   Added some.  I think they about cover it.  Let me know what you think.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1550640629


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   You know what, I'll admit that the fact that some are combined into 
[test_custom_waiters.py](https://github.com/apache/airflow/blob/main/tests/providers/amazon/aws/waiters/test_custom_waiters.py)
 and others were in their own files messed me up.  
   
![image](https://github.com/apache/airflow/assets/1920178/27699661-ea24-4aeb-a632-f9df4350b770)
   
   I thought we only had tests for Neptune and Databrew because the others are 
"hiding".
   
   I still don't really see the point of those ones, but I'll add them in their 
own file, and we can split the combined file out into one-file-per-service like 
everywhere else so it's more obvious in the future.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

Taragolis commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1550638175


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   > We don't test the waiters for other services
   
   TBH we test it, if we skip some of them then we should add it, some issue 
which happen in the past because we have lack of the simple tests 
https://github.com/apache/airflow/pull/33656



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1550257495


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   I'm not sure about this one.   We don't test the waiters for other services 
and it's not really testing anything.  In the Neptune one that was just added, 
we're mocking the return and not even making any asserts, just making sure they 
don't crash.  I don't see any benefit there.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1550257495


##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   I'm, not sure about this one.   We don't test the other services and it's 
not really testing anything.  In the Neptune one, we're mocking the return and 
not even making any asserts, just making sure they don't crash.  I don't see 
any benefit there.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1550254242


##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,150 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),
+and "fail"
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks.
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion.
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook
+template_fields: Sequence[str] = aws_template_fields(
+"job_name",
+"custom_model_name",
+"role_arn",
+"base_model_id",
+"hyperparameters",
+"check_if_job_exists",
+"action_if_job_exists",
+"customization_job_kwargs",
+)
+
+def __init__(
+self,
+job_name: str,
+custom_model_name: str,
+role_arn: str,
+base_model_id: str,
+training_data_uri: str,
+output_data_uri: str,
+hyperparameters: dict[str, str],
+check_if_job_exists: bool = True,
+action_if_job_exists: str = "timestamp",
+customization_job_kwargs: dict[str, Any] | None = None,
+wait_for_completion: bool = True,
+waiter_delay: int = 120,
+waiter_max_attempts: int = 75,
+deferrable: bool = conf.getboolean("operators", "default_deferrable", 
fallback=False),
+**kwargs,
+):
+super().__init__(**kwargs)
+self.wait_for_completion = wait_for_completion
+self.waiter_delay = waiter_delay
+self.waiter_max_attempts = waiter_max_attempts
+self.deferrable = deferrable
+
+self.job_name = job_name
+self.custom_model_name = custom_model_name
+self.role_arn = role_arn
+self.base_model_id = base_model_id
+self.training_data_config = {"s3Uri": training_data_uri}
+self.output_data_config = {"s3Uri": output_data_uri}
+self.hyperparameters = hyperparameters
+self.check_if_job_exists = check_if_job_exists
+self.customization_job_kwargs = customization_job_kwargs or {}
+if action_if_job_exists in {"timestamp", "fail"}:
+self.action_if_job_exists = action_if_job_exists
+else:
+raise AirflowException(
+f"Argument action_if_job_exists accepts only 'timestamp', and 
'fail'. \
+Provided value: '{action_if_job_exists}."
+)

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1550181713


##
airflow/providers/amazon/aws/hooks/bedrock.py:
##
@@ -16,9 +16,54 @@
 # under the License.
 from __future__ import annotations
 
+from botocore.exceptions import ClientError
+
 from airflow.providers.amazon.aws.hooks.base_aws import AwsBaseHook
 
 
+class BedrockHook(AwsBaseHook):
+"""
+Interact with Amazon Bedrock.
+
+Provide thin wrapper around 
:external+boto3:py:class:`boto3.client("bedrock") `.
+
+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`
+"""
+
+client_type = "bedrock"
+
+def __init__(self, *args, **kwargs) -> None:
+kwargs["client_type"] = self.client_type
+super().__init__(*args, **kwargs)
+
+def _get_job_by_name(self, job_name: str):
+return self.conn.get_model_customization_job(jobIdentifier=job_name)
+
+def get_customize_model_job_state(self, job_name) -> str:
+state = self._get_job_by_name(job_name)["status"]
+self.log.info("Job '%s' state: %s", job_name, state)
+return state
+
+def job_name_exists(self, job_name: str) -> bool:
+try:
+self._get_job_by_name(job_name)
+self.log.info("Verified that job name '%s' does exist.", job_name)
+return True
+except ClientError as e:
+if e.response["Error"]["Code"] == "ValidationException":
+self.log.info("Job name '%s' does not exist.", job_name)
+return False
+else:
+raise e

Review Comment:
   fair.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

Taragolis commented on code in PR #38693:
URL: https://github.com/apache/airflow/pull/38693#discussion_r1549147317


##
airflow/providers/amazon/aws/hooks/bedrock.py:
##
@@ -16,9 +16,54 @@
 # under the License.
 from __future__ import annotations
 
+from botocore.exceptions import ClientError
+
 from airflow.providers.amazon.aws.hooks.base_aws import AwsBaseHook
 
 
+class BedrockHook(AwsBaseHook):
+"""
+Interact with Amazon Bedrock.
+
+Provide thin wrapper around 
:external+boto3:py:class:`boto3.client("bedrock") `.
+
+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`
+"""
+
+client_type = "bedrock"
+
+def __init__(self, *args, **kwargs) -> None:
+kwargs["client_type"] = self.client_type
+super().__init__(*args, **kwargs)
+
+def _get_job_by_name(self, job_name: str):
+return self.conn.get_model_customization_job(jobIdentifier=job_name)
+
+def get_customize_model_job_state(self, job_name) -> str:
+state = self._get_job_by_name(job_name)["status"]
+self.log.info("Job '%s' state: %s", job_name, state)
+return state
+
+def job_name_exists(self, job_name: str) -> bool:
+try:
+self._get_job_by_name(job_name)
+self.log.info("Verified that job name '%s' does exist.", job_name)
+return True
+except ClientError as e:
+if e.response["Error"]["Code"] == "ValidationException":
+self.log.info("Job name '%s' does not exist.", job_name)
+return False
+else:
+raise e

Review Comment:
   ```suggestion
   raise
   ```



##
airflow/providers/amazon/aws/waiters/bedrock.json:
##


Review Comment:
   Worthwhile to tests waiter separately, time to time we have invalid 
definitions in it:
   
https://github.com/apache/airflow/tree/main/tests/providers/amazon/aws/waiters 



##
airflow/providers/amazon/aws/operators/bedrock.py:
##
@@ -91,3 +96,150 @@ def execute(self, context: Context) -> dict[str, str | int]:
 self.log.info("Bedrock %s prompt: %s", self.model_id, self.input_data)
 self.log.info("Bedrock model response: %s", response_body)
 return response_body
+
+
+class BedrockCustomizeModelOperator(AwsBaseOperator[BedrockHook]):
+"""
+Create a fine-tuning job to customize a base model.
+
+.. seealso::
+For more information on how to use this operator, take a look at the 
guide:
+:ref:`howto/operator:BedrockCustomizeModelOperator`
+
+:param job_name: A unique name for the fine-tuning job.
+:param custom_model_name: A name for the custom model being created.
+:param role_arn: The Amazon Resource Name (ARN) of an IAM role that Amazon 
Bedrock can assume
+to perform tasks on your behalf.
+:param base_model_id: Name of the base model.
+:param training_data_uri: The S3 URI where the training data is stored.
+:param output_data_uri: The S3 URI where the output data is stored.
+:param hyperparameters: Parameters related to tuning the model.
+:param check_if_job_exists: If set to true, operator will check whether a 
model customization
+job already exists for the name in the config. (Default: True)
+:param action_if_job_exists: Behavior if the job name already exists. 
Options are "timestamp" (default),
+and "fail"
+:param customization_job_kwargs: Any optional parameters to pass to the 
API.
+
+:param wait_for_completion: Whether to wait for cluster to stop. (default: 
True)
+:param waiter_delay: Time in seconds to wait between status checks.
+:param waiter_max_attempts: Maximum number of attempts to check for job 
completion.
+:param deferrable: If True, the operator will wait asynchronously for the 
cluster to stop.
+This implies waiting for completion. This mode requires aiobotocore 
module to be installed.
+(default: False)
+:param aws_conn_id: The Airflow connection used for AWS credentials.
+If this is ``None`` or empty then the default boto3 behaviour is used. 
If
+running Airflow in a distributed manner and aws_conn_id is None or
+empty, then default boto3 configuration would be used (and must be
+maintained on each worker node).
+:param region_name: AWS region_name. If not specified then the default 
boto3 behaviour is used.
+:param verify: Whether or not to verify SSL certificates. See:
+
https://boto3.amazonaws.com/v1/documentation/api/latest/reference/core/session.html
+:param botocore_config: Configuration dictionary (key-values) for botocore 
client. See:
+
https://botocore.amazonaws.com/v1/documentation/api/latest/reference/config.html
+"""
+
+aws_hook_class = BedrockHook

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on PR #38693:
URL: https://github.com/apache/airflow/pull/38693#issuecomment-2033530120

   I also didn't add a BedrockDeleteCustomModelOperator because it is literally 
just a hook call:  
`BedrockHook().conn.delete_custom_model(modelIdentifier=custom_model_arn)`.  
It's instantaneous and irreversible, so no sensors or waiters or triggers or 
anything like that, just a bulky wrapper around a hook call.  Should I add it 
for completion sake?


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Re: [PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi commented on PR #38693:
URL: https://github.com/apache/airflow/pull/38693#issuecomment-2033487899

   Just realized I didn't finish making the sensor deferrable, I'll add the 
execute_complete() to that tomorrow.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

[PR] Amazon Bedrock -Model Customization Jobs [airflow]

[via GitHub]

ferruzzi opened a new pull request, #38693:
URL: https://github.com/apache/airflow/pull/38693

   Adds support for Model Customization Jobs.  Includes changes to the hook(s), 
new Operator, Sensor, Trigger, Waiter, unit tests for the above, updates to the 
doc page, and additions to the system test.
   
   Manually tested in Breeze using (wait_for_completion=False with a Sensor), 
wait_for_completion=True, and deferrable=True
   
![image](https://github.com/apache/airflow/assets/1920178/e4d9630c-0a49-4d71-808c-69688a24e895)
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org