Output of Python Functions lost in Dragon Backend when using Process Template

[BenCarter44]

Currently, when running 02-workload-heterogeneous.py in the /examples directory via dragon, I am unable to get the output of Python ComputeTasks that include a process template.

I noticed that mp.set_start_method("dragon") was missing in the file. Adding that in fixed stdout capture of ComputeTasks with process templates. However, for running native Python functions, the return value is missing.

Output:

Task task.000004: DONE output: '0'  Stdout: 'Single Function'  Return '0'

Expected Output:

Task task.000004: DONE output: ("my.hostname",42)  Stdout: 'Single Function'  Return ("my.hostname",42)

For reference:

async def single_function():
    import socket
    import random
    
    print("Single run", flush=True)
    
    return socket.gethostname(), random.randint(0, 100)

...

ComputeTask(
            function=single_function,
            task_backend_specific_kwargs={"process_template": {}},
        )

[BenCarter44]

Update:

What is written prior works with compute tasks which fire executables. However, for python functions, the output of the function remains lost (see Task 4 and 5 of example 02-workload-heterogeneous.py).

In ROSE, which calls Rhapsody underneath, it can also be replicated below:

@acl.simulation_task
def do_sim(a, b, c, task_description={"process_template": {} }):
    output = a + b + c
    return outputs

Including the process template causes the output to be lost. Removing the task description and process template keeps the output.