Michael, I wasn't even aware of this behavior of falling back to jobid_name if $jobid_var is unset. Could you please file a ticket in Jira LUDOC about this, and ideally submit a patch to explain this in the manual.
Cheers, Andreas On Aug 12, 2022, at 16:26, Sternberg, Michael G. via lustre-discuss <[email protected]<mailto:[email protected]>> wrote: Einar, The strings in your $SLURM_JOB_ID values or host names are likely too long to serve as jobid for the Lustre Jobstats feature . You might try %H instead of %h in jobid_name. For reference, from the Lustre manual, https://doc.lustre.org/lustre_manual.xhtml#jobstats : %e print executable name %g print group ID number %h print fully-qualified hostname %H print short hostname %j print JobID from process environment variable named by the jobid_var parameter %p print numeric process ID %u print user ID number On my system (2.12), I use: jobid_var=PBS_JOBID jobid_name=%e.%u I get job_stats by $PBS_JOBID, as expected, from processes that actually have the variable set, and synthetic %e.%u values from all others, like processes on interactive or backup nodes. This has been working just fine to pinpoint the source of occasional trouble. Curiously, I don't think the manual spells out what happens when the variable referenced by jobid_var is unset, i.e., the above fallback logic from jobid_var to jobid_name. With best regards, -- Michael Sternberg, Ph.D. Principal Scientific Computing Administrator Center for Nanoscale Materials Argonne National Laboratory On Aug 12, 2022, at 03:37, Einar Næss Jensen <[email protected]<mailto:[email protected]>> wrote: logfiles on oss servers are full of these error messages: Invalid jobid size (37), expect(32) What does it mean? we have set this: [root@mds-1 ~]# lctl get_param jobid_var jobid_name jobid_var=SLURM_JOB_ID jobid_name=%j:%u:%h lustre version is 2.12.6(ddn) _______________________________________________ lustre-discuss mailing list [email protected]<mailto:[email protected]> http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org Cheers, Andreas -- Andreas Dilger Lustre Principal Architect Whamcloud
_______________________________________________ lustre-discuss mailing list [email protected] http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
