Using ParallelRunConfig and DockerConfig not working

[VincentMcLoughlin]

Hello, I am attempting to use the ParallelRunStep for a set of python scripts that I currently run inside a docker container. I am trying to deploy this using an AzureML pipeline that reads from the docker image saved in our ACR. However, the init and run scripts in my parallel script never seem to be able to access the files I have moved into the docker container. Particularly when I try and search for a hydra config file in my init and run scripts, I am unable to find the file, even when I look for it explicitly. I have included a snippet that is based on the example code from this repo: example of parallel run config

Any idea what I am doing wrong? I have included a snippet of what I am trying below

docker_config = DockerConfiguration(use_docker=True, arguments=docker_args)
 
environment_name = "my-environment"
environment = Environment(environment_name)
base_image_name = os.getenv("ACR_BASE_IMAGE_NAME")
base_image_tag = os.getenv("ACR_IMAGE_TAG")
environment.docker.base_image = f"{base_image_name}:{base_image_tag}"

environment.docker.base_image_registry.address = f"{acr_name}.azurecr.io"

environment.docker.base_image_registry.username = os.getenv("ACR_USER")
environment.docker.base_image_registry.password = os.getenv("ACR_PASSWORD")
environment.python.user_managed_dependencies = True
environment.docker.enabled=True

run_config = RunConfiguration()
run_config.environment = environment
run_config.docker = docker_config

parallel_run_config = ParallelRunConfig(
        source_directory=".",
        entry_script="path/to/docker/script_1.py",  
        compute_target=compute_target,
        environment=environment,
        node_count=2,
        error_threshold=10,
        output_action="append_row",
        mini_batch_size=1,
        logging_level='DEBUG'
        )
step_parallel = ParallelRunStep(
        name="parallel-step",
        parallel_run_config=parallel_run_config,
        inputs=my_inputs,
        output=output_dir,
        arguments=args,                                   
        allow_reuse=True,
    )
    
step_parallel._runconfig.docker = docker_config #Tried with and without, does not seem to make a difference

pipeline_steps = StepSequence(steps=[step_parallel])
pipeline_run = Pipeline(workspace=ws, steps=pipeline_steps)

# Submit your pipeline run
submitted_pipeline_run = Experiment(ws, "Azure Pipeline").submit(pipeline_run, regenerate_outputs=True)```

[akafle01]

Hi, your "user_managed_dependencies" should be set to false since AzureML does not install dependencies automatically inside the container. Second, you need to specify packages so import the CondaDependencies from azureml.core.conda_dependencies, and lastly you want to ensure that the file is in the appropriate location inside the container, in this case if you are using hydra simply define the configuration path in the script. This should fix all of the issues you are experiencing. However if none of these solutions work, check your logs in AzureML and run the container manually to check if files can be accessed.