Re: [Condor-users] "Bad CONDOR_JOB_STATUS_CONSTRAINED Result" [Sec=Unclassified]

[Jaime Frey <jfrey@xxxxxxxxxxx>]

On Jul 8, 2008, at 12:40 AM, Troy Robertson wrote:

> I've posted this problem recently but had no resolution, but think it
> may have been because it was formatted HTML?
>
> Anyway,
>
> I upgraded our Condor config to Condor-C to address problems for users
> submitting from semi-permanent laptops, and now they are unable to
> receive their results.
>
> Condor_C Jobs are being submitted, executed and are showing as status
> Complete on the remote linux central manager but are not being  
> returned
> to the windows submit machines. The GridManager keeps returning the
> following error:
>
> 6/26 08:56:10 [1556] ERROR "Bad CONDOR_JOB_STATUS_CONSTRAINED  
> Result" at
> line 3808 in file ..\src\condor_gridmanager\gahp-client.C
>
> I also keep ending up with a core.C_GAHP.WIN32 core dump from the GAHP
> server.
>
> All 5 execute machines are linux, same config and dedicated.
> Submitters are Windows XP.  Central Manager is linux.
> Using 6.9.3, not sure if upgrading might help?
>
> With further testing it appears to happen when I submit more than one
> job.  If I submit one job and wait the results are returned.  If I
> submit more than one then they all just sit there on the central  
> manager
> in a Completed state.  If I then remove the first job the rest are
> returned as expected.
>
> In the gridlog below, The job was sitting Completed on the central
> manager and the gridmanager just cycled through the first 15 or so  
> lines
> repeatedly.  I then did a condor_rm on the job and the error was  
> thrown.
> A second job I had submitted, which was also sitting 'Complete' and
> blocked? by the first was then returned.


This looks like a strange bug. Can you add the following lines to your  
Condor config file:

C_GAHP_DEBUG = D_FULLDEBUG
C_GAHP_WORKER_THREAD_DEBUG = D_FULLDEBUG

Then repeat your test and send me (off-list) the resulting gridmanager  
and c-gahp logs. You can find the locations of the latter by running  
these commands:
condor_config_val C_GAHP_LOG
condor_config_val C_GAHP_WORKER_THREAD_LOG

Also, can you try examining the core file (produced by either condor_c- 
gahp or condor_c-gahp_worker_thread) and getting a stack trace?

Thanks and regards,
Jaime Frey
UW-Madison Condor Team