def example_run_monitor_pipeline()

in modules/monitoring/monitor_pipeline_run.py [0:0]

• 38 lines of code
• 2 McCabe index (conditional complexity)

def example_run_monitor_pipeline(workflow_arn, region, config):
    """
    execute the Workflow, which consists of four steps:

    1. Define job names for pre-processing, data, and evaluation
    2. Upload source code for pre-processing, data, and evaluation
    3. Define URLs for the input, output, and intermediary data
    4. Execute the workflow with populated parameters, and monitor the progress
    5. Inspect the evaluation result when the execution is completed
    """

    data_pipeline = get_existing_monitor_pipeline(workflow_arn)

    # Step 1 - Generate unique names for all SageMaker Jobs
    unique_id = uuid.uuid1().hex
    preprocessing_job_name = f"monitor-sm-preprocessing-{unique_id}"
    preprocessing_infer_job_name = f"monitor-sm-preprocessing-infer-{unique_id}"

    # Step 2 - Upload all source code to s3
    PREPROCESSING_SCRIPT_LOCATION = "../../src/mlmax/monitoring.py"

    sagemaker_session = sagemaker.Session()
    input_preprocessing_code = sagemaker_session.upload_data(
        PREPROCESSING_SCRIPT_LOCATION,
        bucket=sagemaker_session.default_bucket(),
        key_prefix=f"{preprocessing_job_name}/source",
    )

    # Step 3 - Define data URLs, preprocessed data URLs can be made
    # input_data = "s3://ml-proserve-nyc-taxi-data/manifest/taxi.manifest"
    input_data = config["default"]["InputSource"]
    infer_data = config["default"]["InferSource"]
    s3_monitor_bucket = config["default"]["MonitorS3Bucket"]
    s3_monitor_prefix = config["default"]["MonitorS3Prefix"]
    print("InputSource:", input_data)
    print("InferSource:", infer_data)
    print("MonitorS3Bucket:", s3_monitor_bucket)
    print("MonitorS3Prefix:", s3_monitor_prefix)
    print("InputProcessingcode", input_preprocessing_code)

    train_output_data = f"s3://{s3_monitor_bucket}/{s3_monitor_prefix}/"
    infer_output_data = f"s3://{s3_monitor_bucket}/{s3_monitor_prefix}/inference"
    execution = data_pipeline.execute(
        inputs={
            "PreprocessingJobName": preprocessing_job_name,
            "PreprocessingInferJobName": preprocessing_infer_job_name,
            "PreprocessingCodeURL": input_preprocessing_code,
            "InputDataURL": input_data,
            "InferDataURL": infer_data,
            "MonitorTrainOutputURL": train_output_data,
            "MonitorInferOutputURL": infer_output_data,
        }
    )
    execution.get_output(wait=True)
    execution.render_progress()

    # Step 5 - Inspect the output of the Workflow execution
    workflow_execution_output_json = execution.get_output(wait=True)

    print(f"output: {workflow_execution_output_json}")