feat: Add detail profiler V2 options and tests

doc/api/training/debugger.rst

@@ -20,7 +20,7 @@ Debugger Rule APIs
 .. autoclass:: get_rule_container_image_uri
     :show-inheritance:
 
-.. autoclass:: get_default_profiler_rule
+.. autoclass:: get_default_profiler_processing_job
     :show-inheritance:
 
 .. class:: sagemaker.debugger.rule_configs
@@ -45,10 +45,6 @@ Debugger Rule APIs
     :show-inheritance:
     :inherited-members:
 
-.. autoclass:: ProfilerRule
-    :show-inheritance:
-    :inherited-members:
-
 Debugger Configuration APIs
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -60,43 +56,3 @@ Debugger Configuration APIs
 
 .. autoclass:: TensorBoardOutputConfig
     :show-inheritance:
-
-.. autoclass:: ProfilerConfig
-    :show-inheritance:
-
-Debugger Configuration APIs for Framework Profiling (Deprecated)
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-.. warning::
-
-  SageMaker Debugger deprecates the framework profiling feature starting from TensorFlow 2.11 and PyTorch 2.0. You can still use the feature in the previous versions of the frameworks and SDKs as follows.
-
-  * SageMaker Python SDK <= v2.130.0
-  * PyTorch >= v1.6.0, < v2.0
-  * TensorFlow >= v2.3.1, < v2.11
-
-  With the deprecation, SageMaker Debugger discontinues support for the APIs below this note.
-
-  See also `Amazon SageMaker Debugger Release Notes: March 16, 2023 <https://docs.aws.amazon.com/sagemaker/latest/dg/debugger-release-notes.html#debugger-release-notes-20230315>`_.
-
-.. autoclass:: FrameworkProfile
-    :show-inheritance:
-
-.. autoclass:: DetailedProfilingConfig
-    :show-inheritance:
-
-.. autoclass:: DataloaderProfilingConfig
-    :show-inheritance:
-
-.. autoclass:: PythonProfilingConfig
-    :show-inheritance:
-
-.. autoclass:: PythonProfiler
-    :show-inheritance:
-
-.. autoclass:: cProfileTimer
-    :show-inheritance:
-
-.. automodule:: sagemaker.debugger.metrics_config
-    :members: StepRange, TimeRange
-    :undoc-members:

doc/api/training/index.rst

@@ -5,11 +5,12 @@ Training APIs
 .. toctree::
    :maxdepth: 4
 
+   algorithm
    analytics
    automl
    debugger
    estimators
-   algorithm
    tuner
    parameter
    processing
+   profiler

doc/api/training/profiler.rst

@@ -0,0 +1,102 @@
+Profiler
+--------
+
+Amazon SageMaker Profiler provides full visibility
+into provisioned compute resources for training
+state-of-the-art deep learning models.
+The following SageMaker Profiler classes are
+for activating SageMaker Profiler while creating
+an estimator object of `:class:sagemaker.pytorch.estimator.PyTorch`
+or `:class:sagemaker.tensorflow.estimator.TensorFlow`.
+
+.. contents::
+
+.. currentmodule:: sagemaker.debugger
+
+Profiler configuration modules
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. class:: sagemaker.Profiler(cpu_profiling_duration=3600)
+
+    A configuration class to activate
+    `Amazon SageMaker Profiler <https://docs.aws.amazon.com/sagemaker/latest/dg/train-profile-computational-performance.html>`_.
+
+    To adjust the Profiler configuration instead of using the default configuration, use the following parameters.
+
+    **Parameters:**
+
+        - **cpu_profiling_duration** (*str*): Specify the time duration in seconds for
+          profiling CPU activities. The default value is 3600 seconds.
+
+    **Example usage:**
+
+    .. code:: python
+
+        import sagemaker
+        from sagemaker.pytorch import PyTorch
+        from sagemaker import ProfilerConfig, Profiler
+
+        profiler_config = ProfilerConfig(
+            profiler_params = Profiler(cpu_profiling_duration=3600)
+        )
+
+        estimator = PyTorch(
+            framework_version="2.0.0",
+            ... # Set up other essential parameters for the estimator class
+            profiler_config=profiler_config
+        )
+
+    For a complete instruction on activating and using SageMaker Profiler, see
+    `Use Amazon SageMaker Profiler to profile activities on AWS compute resources
+    <https://docs.aws.amazon.com/sagemaker/latest/dg/train-profile-computational-performance.html>`_.
+
+.. autoclass:: sagemaker.ProfilerConfig
+
+
+Profiler Rule APIs
+~~~~~~~~~~~~~~~~~~
+
+The following API is for setting up SageMaker Debugger's profiler rules
+to detect computational performance issues from training jobs.
+
+.. autoclass:: ProfilerRule
+    :inherited-members:
+
+
+Debugger Configuration APIs for Framework Profiling (Deprecated)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. warning::
+
+  In favor of `Amazon SageMaker Profiler <https://docs.aws.amazon.com/sagemaker/latest/dg/train-profile-computational-performance.html>`_,
+  SageMaker Debugger deprecates the framework profiling feature starting from TensorFlow 2.11 and PyTorch 2.0. You can still use the feature in the previous versions of the frameworks and SDKs as follows.
+
+  * SageMaker Python SDK <= v2.130.0
+  * PyTorch >= v1.6.0, < v2.0
+  * TensorFlow >= v2.3.1, < v2.11
+
+  With the deprecation, SageMaker Debugger discontinues support for the APIs below this note.
+
+  See also `Amazon SageMaker Debugger Release Notes: March 16, 2023 <https://docs.aws.amazon.com/sagemaker/latest/dg/debugger-release-notes.html#debugger-release-notes-20230315>`_.
+
+.. autoclass:: FrameworkProfile
+    :show-inheritance:
+
+.. autoclass:: DetailedProfilingConfig
+    :show-inheritance:
+
+.. autoclass:: DataloaderProfilingConfig
+    :show-inheritance:
+
+.. autoclass:: PythonProfilingConfig
+    :show-inheritance:
+
+.. autoclass:: PythonProfiler
+    :show-inheritance:
+
+.. autoclass:: cProfileTimer
+    :show-inheritance:
+
+.. automodule:: sagemaker.debugger.metrics_config
+    :members: StepRange, TimeRange
+    :undoc-members:

src/sagemaker/__init__.py

@@ -62,4 +62,6 @@
 from sagemaker.automl.automl import AutoML, AutoMLJob, AutoMLInput  # noqa: F401
 from sagemaker.automl.candidate_estimator import CandidateEstimator, CandidateStep  # noqa: F401
 
+from sagemaker.debugger import ProfilerConfig, Profiler  # noqa: F401
+
 __version__ = importlib_metadata.version("sagemaker")

src/sagemaker/debugger/__init__.py

@@ -18,7 +18,7 @@
     DEBUGGER_FLAG,
     DebuggerHookConfig,
     framework_name,
-    get_default_profiler_rule,
+    get_default_profiler_processing_job,
     get_rule_container_image_uri,
     ProfilerRule,
     Rule,
@@ -27,6 +27,7 @@
     TensorBoardOutputConfig,
 )
 from sagemaker.debugger.framework_profile import FrameworkProfile  # noqa: F401
+from sagemaker.debugger.profiler import Profiler  # noqa: F401
 from sagemaker.debugger.metrics_config import (  # noqa: F401
     DataloaderProfilingConfig,
     DetailedProfilingConfig,

src/sagemaker/debugger/debugger.py

@@ -20,8 +20,6 @@
 """
 from __future__ import absolute_import
 
-import time
-
 from abc import ABC
 
 from typing import Union, Optional, List, Dict
@@ -31,14 +29,31 @@
 import smdebug_rulesconfig as rule_configs
 
 from sagemaker import image_uris
-from sagemaker.utils import build_dict
+from sagemaker.utils import build_dict, name_from_base
 from sagemaker.workflow.entities import PipelineVariable
+from sagemaker.debugger.profiler_constants import (
+    DETAIL_PROF_PROCESSING_DEFAULT_INSTANCE_TYPE,
+    DETAIL_PROF_PROCESSING_DEFAULT_VOLUME_SIZE,
+)
 
 framework_name = "debugger"
+detailed_framework_name = "detailed-profiler"
 DEBUGGER_FLAG = "USE_SMDEBUG"
 
 
-def get_rule_container_image_uri(region):
+class DetailedProfilerProcessingJobConfig:
+    """ProfilerRule like class.
+
+    Serves as a vehicle to pass info through to the processing instance.
+
+    """
+
+    def __init__(self):
+        self.rule_name = self.__class__.__name__
+        self.rule_parameters = {"rule_to_invoke": "DetailedProfilerProcessing"}
+
+
+def get_rule_container_image_uri(name, region):
     """Return the Debugger rule image URI for the given AWS Region.
 
     For a full list of rule image URIs,
@@ -52,19 +67,28 @@ def get_rule_container_image_uri(region):
         str: Formatted image URI for the given AWS Region and the rule container type.
 
     """
+    if name is not None and name.startswith("DetailedProfilerProcessingJobConfig"):
+        # should have the format like "123456789012.dkr.ecr.us-west-2.amazonaws.com/detailed-profiler-processing:latest"
+        return image_uris.retrieve(detailed_framework_name, region)
+
     return image_uris.retrieve(framework_name, region)
 
 
-def get_default_profiler_rule():
-    """Return the default built-in profiler rule with a unique name.
+def get_default_profiler_processing_job(instance_type=None, volume_size_in_gb=None):
+    """Return the default profiler processing job (a rule) with a unique name.
 
     Returns:
         sagemaker.debugger.ProfilerRule: The instance of the built-in ProfilerRule.
 
     """
-    default_rule = rule_configs.ProfilerReport()
-    custom_name = f"{default_rule.rule_name}-{int(time.time())}"
-    return ProfilerRule.sagemaker(default_rule, name=custom_name)
+    default_rule = DetailedProfilerProcessingJobConfig()
+    custom_name = name_from_base(default_rule.rule_name)
+    return ProfilerRule.sagemaker(
+        default_rule,
+        name=custom_name,
+        instance_type=instance_type,
+        volume_size_in_gb=volume_size_in_gb,
+    )
 
 
 @attr.s
@@ -482,6 +506,8 @@ def sagemaker(
         name=None,
         container_local_output_path=None,
         s3_output_path=None,
+        instance_type=None,
+        volume_size_in_gb=None,
     ):
         """Initialize a ``ProfilerRule`` object for a *built-in* profiling rule.
 
@@ -510,13 +536,19 @@ def sagemaker(
             The instance of the built-in ProfilerRule.
 
         """
+        used_name = name or base_config.rule_name
+        if used_name.startswith("DetailedProfilerProcessingJobConfig"):
+            if volume_size_in_gb is None:
+                volume_size_in_gb = DETAIL_PROF_PROCESSING_DEFAULT_VOLUME_SIZE
+            if instance_type is None:
+                instance_type = DETAIL_PROF_PROCESSING_DEFAULT_INSTANCE_TYPE
         return cls(
-            name=name or base_config.rule_name,
+            name=used_name,
             image_uri="DEFAULT_RULE_EVALUATOR_IMAGE",
-            instance_type=None,
+            instance_type=instance_type,
             container_local_output_path=container_local_output_path,
             s3_output_path=s3_output_path,
-            volume_size_in_gb=None,
+            volume_size_in_gb=volume_size_in_gb,
             rule_parameters=base_config.rule_parameters,
         )
 

src/sagemaker/debugger/profiler.py

@@ -0,0 +1,42 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+
+"""Configuration for collecting profiler v2 metrics in SageMaker training jobs."""
+from __future__ import absolute_import
+
+from sagemaker.debugger.profiler_constants import (
+    FILE_ROTATION_INTERVAL_DEFAULT,
+    CPU_PROFILING_DURATION,
+    DETAIL_PROF_PROCESSING_DEFAULT_INSTANCE_TYPE,
+    DETAIL_PROF_PROCESSING_DEFAULT_VOLUME_SIZE,
+)
+
+
+class Profiler:
+    """A configuration class to activate SageMaker Profiler."""
+
+    def __init__(
+        self,
+        cpu_profiling_duration: str = str(CPU_PROFILING_DURATION),
+        file_rotation_interval: str = str(FILE_ROTATION_INTERVAL_DEFAULT),
+    ):
+        """To specify values to adjust the Profiler configuration, use the following parameters.
+
+        :param cpu_profiling_duration: Specify the time duration in seconds for
+            profiling CPU activities. The default value is 3600 seconds.
+        """
+        self.profiling_parameters = {}
+        self.profiling_parameters["CPUProfilingDuration"] = str(cpu_profiling_duration)
+        self.profiling_parameters["SMPFileRotationSecs"] = str(file_rotation_interval)
+        self.instanceType = DETAIL_PROF_PROCESSING_DEFAULT_INSTANCE_TYPE
+        self.volumeSizeInGB = DETAIL_PROF_PROCESSING_DEFAULT_VOLUME_SIZE