Faster Dirac.deleteJob Implementation?

[sfayer]

Hi,

When a user deletes a large batch of jobs via the API or command line, it can take a long time. We've tracked it down to this loop making a call to the server for each individual job:

DIRAC/src/DIRAC/Interfaces/API/Dirac.py

Lines 1453 to 1457 in 64f2d36

	for jobID in jobIDs:
	can_kill = JobStatus.checkJobStateTransition(jobID, JobStatus.KILLED)["OK"]
	can_del = JobStatus.checkJobStateTransition(jobID, JobStatus.DELETED)["OK"]
	if can_kill or can_del:
	jobIDsToDelete.append(jobID)

(We tried it with that loop removed and it was significantly faster).

Would it be possible to replace it with some kind of bulk version? (I was going to suggest removing those lines entirely as I think the same check is done server-side, but it seems like they were added specially here for some reason?)

Regards,
Simon

[fstagni]

This check was first introduced by me in #5752 and then edited by you in #6665. Unfortunately, my bad, I did not add in the first PR an explanation on why it was necessary but I think I remember that it was done exactly to avoid some issues with the interplay of server side check (like: the user was setting some status that was then "silently" refused server side).

It is I think possible to make it bulk. Would you give it a try, or should I?

[sfayer]

Thanks, I'll have a look at sorting it out...