########################################################################################
##### dynamic model / resources content.
########################################################################################

# primary source code
import model_source_code as modelop_model_source_code
# job metadata, variables pointing to [input, output, appname,etc]
import modelop_job_metadata as modelop_job_metadata
# for printing stack trace
import traceback
# for checking number of params in function signature
from inspect import signature


########################################################################################
##### static resources content.
########################################################################################

if __name__ == "__main__":
    try:
        print("### - ModelOpPySparkExecution execution")

        ## Check for an init function to be executed; An init function is optional
        init_method = getattr(modelop_model_source_code, modelop_job_metadata.init_method_to_be_executed, None)
        if init_method is None:
            print("### - Optional init '" + modelop_job_metadata.init_method_to_be_executed + "' function was not found in model source code")
        else:
            print("### - Executing function: " + modelop_job_metadata.init_method_to_be_executed)
            init_method()

        ## Check for a scoring or metrics function to be executed; A scoring or metrics function is required
        execution_method = getattr(modelop_model_source_code, modelop_job_metadata.method_to_be_executed, None)
        if execution_method is None:
            print("### - '" + modelop_job_metadata.method_to_be_executed + "' function was not found in model source code")
            raise Exception("### - '" + modelop_job_metadata.method_to_be_executed + "' function was not found in model source code")

        ## Executing requested function
        print("### - Executing function: " + modelop_job_metadata.method_to_be_executed)
        num_params = len(signature(execution_method).parameters)
        if num_params == 0:
            execution_method()
        elif num_params == 3:
            execution_method(modelop_job_metadata.job_input, modelop_job_metadata.job_output, modelop_job_metadata.job_model_assets_input)
        else:
            traceback.print_exc()
            raise Exception("Number of function parameters, '" + str(num_params) + "', does not match expected number of parameters (0 or 3)")

    except Exception as e:
        print("### - An exception of ", e.__class__, " occurred")
        traceback.print_exc()
        print("#################### RUNTIME ERROR ####################")
        raise Exception("### - An exception occurred in the ModelOpPySparkDriver")
    print("### - PySpark code execution completed")

PySparkJobManifest

ModelOp Center decouples the transformation process that takes a MODEL_BATCH_JOB/MODEL_BATCH_TEST_JOB/MODEL_BATCH_TRAINING_JOB as input and produces a SparkLauncher as output. For this purpose, ModelOp Center uses the PySparkJobManifest intermediate component. PySparkJobManifest contains all the required values when running a spark-submit: Python source code and metadata, function name, application name (all of which are used inside ModelOpPySparkDriver). This design allows us to decouple and to keep agnostic the spark-submit.

Image Removed

The following diagram illustrates the transformation process of a MODEL_BATCH_JOB/MODEL_BATCH_TEST_JOB/MODEL_BATCH_TRAINING_JOB into a SparkLauncher:

Image Removed

ModelOpJobMonitor

The ModelOpJobMonitor has two main responsibilities:

1. Submitting jobs to the Spark cluster

2. Monitoring job statuses for existing, waiting and running, Spark jobs.

In order to successfully fulfill these responsibilities, it relies on two key services:

Image Removed

PySpark Job Input and Output Data

The spark-runtime-service supports two types of input and output data for MODEL_BATCH_JOB/MODEL_BATCH_TEST_JOB/MODEL_BATCH_TRAINING_JOB:

1. HDFS asset(s) by URL a.k.a. External asset(s)

2. Embedded asset(s) a.k.a. File asset(s)

Input Data by URL

When the user provides HDFS asset(s) by URL as input data for a job, the HDFS asset(s) have to be present at the Spark cluster before the job is submitted to the Spark cluster for execution.

Input Data Embedded

When the user provides embedded input data asset(s), the spark-runtime-service will create a temporary HDFS directory for the job and upload the file content of each embedded input data asset to HDFS. The newly created HDFS file(s) will be represented as HDFS asset(s) by URL and used as the input data when the job is submitted to the Spark cluster for execution.

The change in the input data from embedded asset(s) to HDFS asset(s) by URL is not visible to the user because the MODEL_BATCH_JOB/MODEL_BATCH_TEST_JOB/MODEL_BATCH_TRAINING_JOB is never updated to reflect this change. Instead, the change is handled internally in PySparkPreprocessorService.

Output Data by URL

When the user provides HDFS asset(s) by URL as output data for a job, the job output will be stored at the given HDFS file url.

Output Data Embedded

When the user provides embedded output data asset(s), the spark-runtime-service will create a temporary HDFS directory for the job, if one does not exist yet. Then, the following actions are performed in the given order:

1. Represent each embedded output data asset as HDFS asset by URL where the file url of the new HDFS asset is the temporary HDFS directory and the filename of the new HDFS asset is the filename of the corresponding embedded asset

2. Update MODEL_BATCH_JOB/MODEL_BATCH_TEST_JOB/MODEL_BATCH_TRAINING_JOB by storing the HDFS asset(s) by URL in its job parameters Map where the key is TMP_EXTERNAL_OUTPUT

3. Use the newly created HDFS asset(s) by URL as the output data when the job is submitted to the Spark cluster for execution

4. Download each HDFS asset by URL and store its content as the file content for the corresponding embedded output data asset in MODEL_BATCH_JOB/MODEL_BATCH_TEST_JOB/MODEL_BATCH_TRAINING_JOB

5. Remove TMP_EXTERNAL_OUTPUT from the job parameters Map

The change in the output data from embedded asset(s) to HDFS asset(s) by URL is not visible to the user because the MODEL_BATCH_JOB/MODEL_BATCH_TEST_JOB/MODEL_BATCH_TRAINING_JOB is never updated to reflect this change. Instead, the change is handled internally in PySparkPreprocessorService and ModelOpJobMonitor.

For input and output data, the user can have any combination of HDFS asset(s) by URL or embedded asset(s).

PySpark - Metrics Jobs