Showq

From ACENET
Jump to: navigation, search
Main page: Job Control

showq is our custom-made alternative to the Grid Engine command qstat intended to address the question, "When will more slots become available?"

The output from showq is in two sections, Running jobs and Pending jobs.

Running jobs

The first section represents jobs which have been scheduled and assigned to hosts. It looks like this:

============================== Running jobs ===============================
  JOBID   USERNAME   STATE  SLOTS   ----- ENDS BY -----          QUEUE@HOST
 106675      wally      r     16    03/17/2010 16:30:24       short.q@cl133
 106982      alice      r      8    03/17/2010 19:37:06       short.q@cl096
 106986      alice      r      8    03/18/2010 05:22:20       short.q@cl107
 106694    dilbert      r      1    03/21/2010 18:26:51      medium.q@cl026  6
 106694    dilbert      r      1    03/21/2010 18:26:51      medium.q@cl026  5
 106182        phb      r     87 R  03/22/2010 13:38:02      medium.q@cl032
 106965    catbert      r      1    04/13/2010 18:16:35        long.q@cl052

The entries are sorted such that the job which is expected to end soonest appears at the top. The fields are:

  • JOBID: The unique identifier for the job, issued by Grid Engine at qsub
  • USERNAME: Userid of the owner
  • STATE: r(unning), T(hreshold) suspended, s(uspended), S(uspended), d(eleting), t(ransferring)
  • SLOTS: Number of processors assigned to the job
  • ENDS BY: The date and time at which the job's reserved run time (h_rt) will expire. A job may end sooner than this, but will not normally end later.
  • QUEUE@HOST: Queue instance where the job is hosted. For parallel jobs there may be other hosts involved; this column will only indicate where the "master" process resides.

Some other features may appear:

  • An "R" beside the slot count indicates the job used reservation. This is more meaningful for waiting jobs, see below.
  • Numbers to the extreme right indicate sub-tasks of an array job.

Pending jobs

The second section represents jobs waiting to be scheduled, and looks like this:

============================== Pending jobs ===============================
  JOBID   USERNAME   STATE  SLOTS  ---- SUBMITTED ----   PRIORITY  TIME REQ
 106824     scotty     qw     64 R  03/17/2010 09:33:54   0.5321   48:00:00
 106788      uhura     qw     16    03/16/2010 09:18:56   0.5062  240:00:00
 106885      spock     qw      4    03/17/2010 11:12:16   0.3750    8:00:00   1-5:1
 106219       kirk     qw    150    03/12/2010 17:05:59   0.2949  9994:10:00

Jobs in this section are sorted such that the job with the highest scheduling priority appears first. This is the job that Grid Engine will first try to schedule if a running job finishes in the next few seconds.

Here are the fields:

  • JOBID: Same as above
  • USERNAME: Same as above
  • STATE: qw (waiting), Eqw (error), hqw (held)
  • SLOTS: If the job has requested a range of processor counts, only the lower limit is shown here. An "R" indicates the job is reserving.
  • SUBMITTED: Date and time the job was submitted
  • PRIORITY: Scheduling priority currently assigned to the job
  • TIME REQ: Requested run time limit (h_rt) in hh:mm:ss
  • If the job is an array job, the task count specification will appear at the extreme right

Miscellany

Because every job is represented the output can be quite long (like that from qstat -u \*). You may wish to pipe the output through a paginator:

$ showq | less