[Documentation] Flux add `duration` to jobspec

[mgt16-LANL]

Hi All,

Would it be possible to add duration to the jobspecV1 for flux execution? I was trying to come up with the clean way to enforce a timeout with multiple jobs, and it seems like having a duration set from the flux-side might be a clean way to track this:

Flux-jobspec

Which seems to get referenced by the resource_dict here:
spawner_flux.py

This might clash a bit with the logic for executorlib if the idea with the flux executor is to start all these processes ("jobs") then submit multiple python processes to each one.

Either way, suggestions for how to handle timeouts / killing futures after a set duration after execution begins would be nice.

-Michael.

[jan-janssen]

Hi @mgt16-LANL ,

I like the suggestion and opened a pull request to implement this functionality:
#930

Can you give it a test? I still want to extend the test suite to make sure the future objects are also cancelled rather than remaining in the waiting state.

Best,

Jan

[jan-janssen]

I added the test, when the job is cancelled because it runs out of run time the future object is set to done() == True and the result() call raises an executorlib.api.ExecutorlibSocketError error. This is the same that comes up when the executor was not able to start the python process for the execution. While it would be nicer to mark the future object as cancelled() == True this would require the process to keep track of the time that has passed to identify the process was cancelled because it reached the time out. So I hope the error is fine for now.

[jan-janssen]

While the pull request was merges - the documentation is still missing, so I reopened the issue.