Difference between revisions of "HOWTO check up on jobs"

From HPC
m (New page: == Examining the queue == MAUI is the software application that actually decides what resources your job will run on. You can look at the queue by either using the TORQUE <code>qstat</cod...)
 
m (Deleting a job you no longer want)
 
(10 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
== Examining the queue ==
 
== Examining the queue ==
  
MAUI is the software application that actually decides what resources your job will run on. You can look at the queue by either using the TORQUE <code>qstat</code> command, or by using the MAUI <code>showq</code> command. <code>qstat</code> will display the queue ordered by JobID, whereas <code>showq</code> will display jobs grouped by their state ("running," "idle," or "hold") then ordered by priority.
+
You can look at the queue by using the <code>qstat</code> command. <code>qstat</code> will display the queue ordered by JobID.
  
 
<pre>
 
<pre>
[username@head002 ~]$ showq
+
[username@hpc1 ~]$ qstat
ACTIVE JOBS--------------------
+
Job id            Name            User              Time Use S Queue
JOBNAME            USERNAME      STATE  PROC  REMAINING            STARTTIME
+
---------------- ---------------- ---------------- -------- - -----
 
+
32.hpc1          JobName          username          351:04:3 R long
33                username    Running    1    22:07:22  Tue Jun 18 07:58:46
+
33.hpc1          JobName          username          351:06:1 R day
34                username    Running    1    22:07:22  Tue Jun 18 07:58:46
+
34.hpc1          JobName          username          390:30:2 R week
35                username    Running    1    22:07:22  Tue Jun 18 07:58:46
+
40.hpc1          JobName          username          496:38:2 R month
36                username    Running    1    22:07:22  Tue Jun 18 07:58:46
+
46.hpc1          JobName          username          506:13:5 R long
37                username    Running    1    22:07:22  Tue Jun 18 07:58:46
 
38                username    Running    1    22:07:22  Tue Jun 18 07:58:46
 
39                username    Running    1    22:07:22  Tue Jun 18 07:58:46
 
 
 
    7 Active Jobs      7 of    8 Processors Active (87.50%)
 
                        2 of    2 Nodes Active      (100.00%)
 
 
 
IDLE JOBS----------------------
 
JOBNAME            USERNAME      STATE  PROC    WCLIMIT            QUEUETIME
 
 
 
 
 
0 Idle Jobs
 
 
 
BLOCKED JOBS----------------
 
JOBNAME            USERNAME      STATE  PROC    WCLIMIT            QUEUETIME
 
 
 
 
 
Total Jobs: 7  Active Jobs: 7  Idle Jobs: 0  Blocked Jobs: 0
 
 
</pre>
 
</pre>
  
 
== Checking a specific job ==
 
== Checking a specific job ==
  
If you want to see the details of a specific job, use <code>checkjob</code> on it:
+
If you want to see the details of a specific job, use <code>qstat -fx <JobID></code> on it:
  
 
<pre>
 
<pre>
[username@head002 ~]$ checkjob 40
+
[username@hpc1 ~]$ qstat -fx 40
 +
</pre>
  
 +
If you want to look at the output of your job while it's still running, use the <code>qpeek</code> command.
  
checking job 40
+
<pre>
 +
[username@hpc1 ~]$ qpeek 40
 +
</pre>
  
State: Running
+
== Deleting a job you no longer want ==
Creds:  user:username  group:users  class:batch  qos:DEFAULT
 
WallTime: 00:09:18 of 1:00:00:00
 
SubmitTime: Tue Jun 18 09:54:34
 
  (Time Queued  Total: 00:00:01  Eligible: 00:00:01)
 
  
StartTime: Tue Jun 18 09:54:35
+
If you want to delete a job (whether it's already running or not), use the <code>qdel</code> command:
Total Tasks: 1
+
<pre>
 +
[username@hpc1 ~]$ qdel 41
 +
</pre>
  
Req[0]  TaskCount: 1  Partition: DEFAULT
+
There's no output on a successful job deletion. Keep in mind that when running jobs are killed, '''files in scratch space will not sync back to your home directory'''. Orphaned scratch space will be moved to /orphans.
Network: [NONE]  Memory >= 0  Disk >= 0  Swap >= 0
 
Opsys: [NONE]  Arch: [NONE]  Features: [NONE]
 
NodeCount: 1
 
Allocated Nodes:
 
[test001:1]
 
  
 +
== Overview of cluster usage ==
  
IWD: [NONE]  Executable:  [NONE]
+
<code>pestat</code> gives a nice overview of which nodes are busy with which jobs for which users.
Bypass: 0  StartCount: 1
 
PartitionMask: [ALL]
 
Flags:      RESTARTABLE
 
  
Reservation '40' (-00:09:10 -> 23:50:50 Duration: 1:00:00:00)
+
<pre>
PE: 1.00 StartPriority: 1
+
[username@hpc1 ~]$ pestat
 +
Queuesshort day week month long
 +
Node            state    cpu        memory  jobids/users
 +
----                  tot used    tot used
 +
comp001.hpc    busy    8    8    15G  51% 34
 +
comp002.hpc    free    64  60  126G 12% 35 36 37 38
 
</pre>
 
</pre>
  
== Asking MAUI when you job will probably start and finish ==
 
 
If you want to see a time estimate for when you job will start, use the <code>showstart</code> command:
 
 
<pre>
 
<pre>
[username@head002 ~]$ showstart 41
+
[username@hpc1 ~]$ pestat -u username
job 41 requires 1 proc for 1:00:00:00
+
Queues: short day week month long
Earliest start in        22:03:48 on Wed Jun 19 07:58:46
+
Node            state    cpu        memory  jobids/users
Earliest completion in 1:22:03:48 on Thu Jun 20 07:58:46
+
----                  tot used    tot used
Best Partition: DEFAULT
+
comp001.hpc    busy    8    8    15G 51%  34
 +
comp002.hpc    free    64  60  126G  12%  38
 
</pre>
 
</pre>
 
== Overview of cluster usage ==
 
 
<code>pestat</code> gives a nice overview of which nodes are busy with which jobs for which users.
 
  
 
<pre>
 
<pre>
[username@head002 ~]$ pestat
+
[username@hpc1 ~]$ pestat -a -u username
Node               state load   pmem ncpu  mem   resi usrs tasks NetMbit jobids/users
+
Queues:  short day week month long
comp001             excl  3.8    1877  4  5907   398 4/1    4      0!  34 username 36 username 38 username 40 username
+
Node           state    cpu        memory   jobids/users
comp002             excl  3.8   1877   4   5907    409 4/1    4      0    33 username 35 username 37 username 39 username
+
----                  tot used    tot used
 +
comp001.hpc    busy    8    8   15G  51% 34 username
 +
comp002.hpc    free   64   60   126G  12% 38 username
 
</pre>
 
</pre>

Latest revision as of 10:39, 7 March 2017

Examining the queue

You can look at the queue by using the qstat command. qstat will display the queue ordered by JobID.

[username@hpc1 ~]$ qstat
Job id            Name             User              Time Use S Queue
----------------  ---------------- ----------------  -------- - -----
32.hpc1           JobName          username          351:04:3 R long
33.hpc1           JobName          username          351:06:1 R day
34.hpc1           JobName          username          390:30:2 R week
40.hpc1           JobName          username          496:38:2 R month
46.hpc1           JobName          username          506:13:5 R long

Checking a specific job

If you want to see the details of a specific job, use qstat -fx <JobID> on it:

[username@hpc1 ~]$ qstat -fx 40

If you want to look at the output of your job while it's still running, use the qpeek command.

[username@hpc1 ~]$ qpeek 40

Deleting a job you no longer want

If you want to delete a job (whether it's already running or not), use the qdel command:

[username@hpc1 ~]$ qdel 41

There's no output on a successful job deletion. Keep in mind that when running jobs are killed, files in scratch space will not sync back to your home directory. Orphaned scratch space will be moved to /orphans.

Overview of cluster usage

pestat gives a nice overview of which nodes are busy with which jobs for which users.

[username@hpc1 ~]$ pestat
Queues:  short day week month long
Node            state    cpu        memory   jobids/users
----                   tot used    tot used
comp001.hpc     busy     8    8    15G  51%  34
comp002.hpc     free    64   60   126G  12%  35 36 37 38
[username@hpc1 ~]$ pestat -u username
Queues:  short day week month long
Node            state    cpu        memory   jobids/users
----                   tot used    tot used
comp001.hpc     busy     8    8    15G  51%  34
comp002.hpc     free    64   60   126G  12%  38
[username@hpc1 ~]$ pestat -a -u username
Queues:  short day week month long
Node            state    cpu        memory   jobids/users
----                   tot used    tot used
comp001.hpc     busy     8    8    15G  51%  34 username
comp002.hpc     free    64   60   126G  12%  38 username