core.py's Job class default regex not selecting correct job id

[jsmucker]

I have a python code that uses dask_jobqueue's PBSCluster to manage workers on an HPC.

Within this environment, the qsub command calls a wrapper which does several important things for the environment before actually qsubbing the job. This wrapper sometimes prints informational messages to stdout (these messages contain text, numbers, newlines, etc) before listing the job id.

When this happens, the default regex for core.py's Job class ( (?P<job_id>\d+) ) will pull out the wrong string thinking it is the job id when it is in fact not.

One solution I had to this problem was subclassing Job and making _job_id_from_submit_output apply the regex to the last line of the stdout from qsub. I suppose I could also change the regex to something more robust too.

In practice the workarounds would be sufficient, however every change made to already existing software that uses dask at my work requires recertification, which draws more resources.

[jacobtomlinson]

I suppose I could also change the regex to something more robust too.

I'm curious whether this could be solved by just making the regex more flexible, or making the regex configurable?

[jsmucker]

I suppose I could also change the regex to something more robust too.

I'm curious whether this could be solved by just making the regex more flexible, or making the regex configurable?

I think either approach would work, although I'm not adept at regex so I can't say for sure. I like the first idea more because I use a code that has already been released that doesn't do anything additional to set the regex. If I wanted to run that particular code in the environment with the qsub wrapper, then it would be nice not having to modify the source files (and instead just update the python environment to include a version of dask that addresses the issue).