Job operations are not parallelizable

[george-moussa]

I'm trying to submit and process results of multiple jobs in parallel but I'm noticing that, in many cases, jobs are submitted sequentially.

import datetime
import imp
from multiprocessing import Pool
from azure.quantum import Workspace
from azure.quantum.optimization import Problem, ProblemType, Term
from azure.quantum.target.toshiba import SimulatedBifurcationMachine
import logging


def create_problem():
    problem = Problem('Test_Problem', problem_type=ProblemType.pubo)
    problem.add_terms([
        Term(c=1, indices=[]),
        Term(c=2, indices=[1, 2]),
        Term(c=-3, indices=[1, 2]),
        Term(c=1, indices=[0, 2])
    ])
    return problem


def exec_job(no):
    workspace = Workspace (
        resource_id = "<workspace-id>",
        location= "<location>"
    )

    problem = create_problem()
    solver = SimulatedBifurcationMachine(workspace, loops=0, timeout=10)
    print(str(datetime.datetime.utcnow()), "\t[%d] solver.submit() start" % no)
    job = solver.submit(problem)
    print(str(datetime.datetime.utcnow()), "\t[%d] job.id=%s" % (no, str(job.id)))

    print(str(datetime.datetime.utcnow()), "\t[%d] job.get_results() start" % no)
    result = job.get_results()
    print(str(datetime.datetime.utcnow()), "\t[%d] job.details=%s" % (no, str(job.details)))


if __name__ == '__main__':
    logging.basicConfig(level=logging.DEBUG)

    njobs = 2
    with Pool(processes=njobs) as pool:
        pool.map(func=exec_job, iterable=range(njobs))

For the above, output is:
2022-04-21 18:40:23.466530 [0] solver.submit() start
2022-04-21 18:40:23.542727 [1] solver.submit() start
2022-04-21 18:40:28.752803 [1] job.id=810bc3bb-c1a2-11ec-baa1-b831b575aea6
2022-04-21 18:40:28.752803 [1] job.get_results() start
..........2022-04-21 18:40:52.826348 [1] job.details={..........}
2022-04-21 18:44:19.108243 [0] job.id=81018a8e-c1a2-11ec-adf4-b831b575aea6
2022-04-21 18:44:19.109211 [0] job.get_results() start
..........2022-04-21 18:44:42.799796 [0] job.details={..............}

You can see that job 1 was submitted and awaited while job 0 hasn't been submitted at the same time and was done 4 mins after the first job.

[guenp]

Another solution to submitting multiple jobs in parallel would be through async i/o, see #94.

[anpaz]

Unfortunately, I'm seeing the same problem when using the aio library.
There seems to be some deadlock happening on a low-level http library. I have opened a ticket with the azure-sdk team to get their help: Azure/azure-sdk-for-python#24443

[anpaz]

More analysis suggests this is a service problem, which we're investigating and might have a solution soon.