Job manager #389
Job manager #389
Conversation
qiskit/providers/ibmq/jobmanager.py
Outdated
| Raises: | ||
| JobError: If a job cannot be submitted. | ||
| """ | ||
| if (all(isinstance(exp, ScheduleComponent) for exp in experiments) and |
There was a problem hiding this comment.
Double-checking - all or any?
There was a problem hiding this comment.
assemble will fail if they're not all the same type, but any is probably more efficient here.
| except JobError as err: | ||
| logger.warning("Unable to retrieve status for job %d (job ID=%s): %s", | ||
| i, job.job_id(), str(err)) | ||
| statuses.append(None) |
There was a problem hiding this comment.
Would this cause some ripples in report()? It seems it would not account for the job in the non-detailed output (which can be considered "by design"), but more importantly can you check it does not cause an error in the function (ie. when accessing status.value or similar cases)?
There was a problem hiding this comment.
Not counting in the summary report is by design; fixed status.value. We really should have a JobStatus.UNKNOWN.
test/ibmq/test_jobmanager.py
Outdated
| error_report = self._jm.error_message() | ||
| self.assertIsNotNone(error_report) | ||
| print("JobManager.error_message(): \n{}".format(error_report)) | ||
| print("\nJobManager.report(): \n{}".format(self._jm.report())) |
There was a problem hiding this comment.
Can you do a pass for the prints, which seem to be leftovers?
There was a problem hiding this comment.
They were intentional to allow someone to "eyeball" them in the Travis log. But perhaps they just pollute the log. I'll remove them.
There was a problem hiding this comment.
Thanks @jyu00 ! I think it is a solid start - merging so we can split off the remaining work in upcoming PRs and help reviewing.
| from qiskit.providers.basebackend import BaseBackend | ||
| from qiskit.providers.exceptions import JobTimeoutError | ||
| from qiskit.providers import (BaseJob, # type: ignore[attr-defined] | ||
| JobTimeoutError, BaseBackend) |
There was a problem hiding this comment.
Note I reverted this change from master during the conflict solving at the latest merge, as technically the qiskit.providers API is the one that is stable - we don't have that guarantee for qiskit.providers.xxx.yyy (at the expense of the mypy ignore).
Summary
Addresses #333 .
Details and comments
A job manager that takes a list of circuits/schedules, splits them into multiple jobs and submits the jobs. A few notes:
run()only takes a list of circuits/schedules, not a single one. The rationale is the manager is meant to manage a large number of jobs.run()takesIBMQBackend, notBaseBackendas thebackendinput. Local backends don't have circuit limits and thus shouldn't need a job manager.