Open johrstrom opened 4 years ago
I know using PBS_DEFAULT works well for us using Torque/MOAB, but my vote would be to just use the -q option. It's known to works and would save us time in investigation.
I think using command flags makes sense and is more direct. Also would be same approach taken for submitting to different clusters in slurm.
This seems like it's an issue outside our control. It was shown that we correctly set the environment variable so the only alternative for us is to use the full -q
option as it's a more explicit command (i.e., the DEFAULT implies it could be a fallback or is only implied).
We got a bug report in discourse with some strange behaviour with PBS.
What happens:
At this point you can rinse and repeat and it's always submitting the job to the previous cluster. If you wait ~60 seconds you can get back to stage 1 where it works correctly.
The fix was to submit the job with
-q <queue>@<server>
instead of just-q <queue>
.This ticket is to either figure out what's going on with the PBS_DEFAULT environment variable and why it's being cached or simply stop using it and instead force the server in the -q option.
The -q option can take any one of these 3 forms. Currently we're only using the first if
queue_name
orreservation_id
is defined.┆Issue is synchronized with this Asana task by Unito