make parsing jobid from sbatch output more robust #443
Conversation
FrankD412
added
the
Bugfix
There was a problem hiding this comment.
Ok, so only comment i think i have here is a slight change to the regex to ensure we can't match numbers that are part of other words (i.e. it finding the '3' in 'Stampede3'). So how about tweaking the regex to something like the following to ensure it can only find numbers attached to slurm's nominal sbatch output. Something like the following, with or without the named capture groups:
(?:Submitted batch job \b(?P<jobid>\d+)\b)
or maybe don't bother capturing the first part but require it to match:
Submitted batch job \b(?P<jobid>\d+)\b
Likely not worth the hassle to get lookahead/lookbehind constructs to filter out everything but the number as I think one of them isn't even supported in python?
|
Oh, and one more comment. Having the full string in the match means we shouldn't have to assume it's the last line either, so just using the regex.search iterator should find a match on any line; should be safer in case those plugins also allow output after the jobid (would hope not, but easy enough to guard against that). |
This adds
--parsableto thesbatchoptions when submitting a SLURM job. Then, when looking for the jobid returned by SLURM, it only examines the last line of output, which should conform to the description given in the documentation (https://slurm.schedmd.com/sbatch.html#OPT_parsable).Fixes #441.