Resource Manager Notes → Slurm

Helper Script

Open XDMoD includes a helper script to pull data from Slurm’s sacct into Open XDMoD’s shredder system. This script can be used in place of the shredder to import data. To shred data for a Slurm cluster use this command with the name of a single cluster that would be used with sacct’s --clusters option:

$ xdmod-slurm-helper -r mycluster

The helper script doesn’t update the aggregate tables, so that must be done after the data has been shredded:

$ xdmod-ingestor

If your sacct executable isn’t in the PATH of the user that will be running the xdmod-slurm-helper command, you can specify the path by modifying the following section in your portal_settings.ini file.

[slurm]
sacct = "/path/to/sacct"

Use this command to display the help text for the Slurm helper script:

$ xdmod-slurm-helper -h

Input Format

If you’d prefer to not use the helper script, you can export data from Slurm into a file manually using the sacct command and then shred that file. The format must be the same as below. Also, the --parsable2, --noheader and --allocations are all required. Replace *cluster* with the name of your resource. It may also be possible to use other options that limit the output.

$ TZ=UTC sacct --clusters *cluster* --allusers \
    --parsable2 --noheader --allocations --duplicates \
    --format jobid,jobidraw,cluster,partition,account,group,gid,\
user,uid,submit,eligible,start,end,elapsed,exitcode,state,nnodes,\
ncpus,reqcpus,reqmem,reqgres,reqtres,timelimit,nodelist,jobname \
    --state CANCELLED,COMPLETED,FAILED,NODE_FAIL,PREEMPTED,TIMEOUT \
    --starttime 2013-01-01T00:00:00 --endtime 2013-01-01T23:59:59 \
    >/tmp/slurm.log

$ xdmod-shredder -r *cluster* -f slurm -i /tmp/slurm.log

NOTE: The time zone used in the output from sacct must be UTC to prevent ambiguities caused by clock changes due to daylight savings. The shredder will assume input times are in UTC regardless of your system time zone.

Unsupported Shredder Features

The xdmod-shredder -d/--dir option was designed to work with the accounting log naming convention used by PBS/TORQUE. If you are not using the same convention (files are named YYYYMMDD corresponding to the date jobs ended), do not use this option.