Feature Request: Support for Placeholders for Sagemaker Processing container arguments

[manshahe]

This is a feature request for the newly added stepfunctions.steps.sagemaker.ProcessingStep to accept placeholders for inputs to the container_argument parameter. An example case is depicted below:

execution_input = ExecutionInput(
    schema={
        "JobName": str,
        #This argument is for the Sagemaker Processing step
        "argument": str
    }
)

processing_step = ProcessingStep(
    job_name=execution_input["JobName"],
    processor={sagemaker processor object here},
    inputs=inputs,
    outputs = outputs,
    #Argument passed here to Sagemaker Processing
    container_arguments=['--argument', execution_input.get['argument'].to_jsonpath()],
    container_entrypoint=["python3", "/opt/ml/processing/input/code/scipt.py"]
)

Rationale: This will make the generated state machines more generic, we can generate 1 state machine and pass it arguments for different workloads. Without this functionality, I need to generate n state machines for a parameter which takes n possible values.

[wong-a]

Creating an array with placeholder values for specific elements will require using the States.Array intrinsic function unless the referenced JSONPath value is an array.

"AppSpecification.ContainerArguments.$": "States.Array('--argument', $.argument)"

[ca-nguyen]

The fix to support placeholders (#155) will be available in next release.

[yanhong-zhao-ef]

I am wondering if we can re-open this issue since the original feature request is still not possible where execution_input['argument'].to_jsonpath() will get passed straight to the processing script without the variable being replaced on execution and execution_input['argument'] runs into TypeError: Object of type ExecutionInput is not JSON serializable when trying to create workflow