Conference vaxaxp::vmsnotes

673.0. "batch jobs hang in 'starting' - JOB_CONTROL uses different usernames across the cluster" by LNZALI::BACHNER (Mouse not found. Click OK to continue) Tue Jun 03 1997 05:29

A customer had a problem with the DECwindows login box not coming up and we
could track it down to the fact that process JOB_CONTROL had been restartet
interactively from an account different from the SYSTEM account.

Apparently, at least two of the three JOB_CONTROL processes were running with
username J_DOE instead of system (I'm not sure about the third one).

After they restarted the JOB_CONTROL process from SYSTEM, DECwindows worked fine
again, but new batch jobs on another node in the cluster remained in 'starting'
state (already executing jobs were unaffected). When they restarted JOB_CONTROL
from SYSTEM on that node, things were back to normal. The Queue Manager was/is
running on the node that originally had the DECwindows problem.

Is anybody sufficient familiar with Job Controiller / Queue Manager interaction
across the cluster to confirm that JOB_CONTROL processes running under different
usernames can lead to the symptoms observed ?  The customer is happy with the
solution so far (restart JOB_CONTROL as SYSTEM) but wants to know whether this
really was the source of his problem or whether he should investigate other
possible reasons.

The cluster runs OpenVMS/Alpha V6.2 (three nodes, I believe) with a bunch of
patches, including ALPQMAN03_070.

Thanks for your help,
Hans.

Normally if you restart things like JOBCTL from another account than SYSTEM
you get this warning:

$ @sys$system:startup jobctl
%STDRV-W-NOTSYSTEM, STARTUP is not running from the SYSTEM account.
-STDRV-W-UNPREDICT, Results are unpredictable.

Should'nt this be a fair warning not to do this???

Jur.

> Shouldn't this be a fair warning not to do this???

Sure it is. The customer was satisfied with this answer.

Thanks,
Hans.