Re: [Condor-devel] Questions about quill

[Ameet Kini <akini@xxxxxxxxxxx>]

Hi,

Appreciate the feedback.

> > I know Quill is very new, so it's not surprising it has a few bugs.
> >
> > 1.) "condor_q -long" (no constraints) almost always exits with a
> >     segmentation fault.  Same thing if the constraint is a username.
> >     It only seems to work when constrained to a jobid or clusterid.

I just finish running condor_q -long in a loop 1000 times and could not
recreate it.  Also I haven't encountered this behavior yet, but that
doesn't mean that a bug doesn't exist. Can you send us a core file?

> >
> > 2.) "condor_q jobid1 jobid2" only displays jobid2.  On the other hand,
> >     "condor_q user1 user2" works as expected.
> > 3.) Why does condor_quill track the job_queue.log the way it does?
> >     It would be much more reliable if it kept a file descriptor open
> >     so that it wouldn't lose any information in the small window when
> >     the file is compressed.  Logic similar to "tail --follow=name"
> >     would be much more robust.  The code seems overly complex because
> >     of how it determines if the file has been rewritten.
> >

Yes, these two will be patched soon.  In the meantime, you can simulate
the behavior of
	condor_q cid1.pid1 cid2.pid2
by
	condor_q -constraint "(ClusterId == cid1 && ProcId == pid1) ||
				(ClusterId == cid2 && ProcId == pid2)"

Regards,
Ameet

>
> Yeah, we know. A lot of it is wonkiness in the schedd - Dan B has changed
> the way the schedd compresses the job_queue to make it easier on quill, so
> I think the next release will have much simplified code to follow the job
> queue. (Dan, did you implement the new record type too?)
>
> -Erik
>