[GE users] Accounting file shows submit time later than start.

reuti reuti at staff.uni-marburg.de
Mon Mar 29 18:36:19 BST 2010


Am 29.03.2010 um 17:16 schrieb griznog:

> On Mon, Mar 29, 2010 at 8:17 AM, reuti <reuti at staff.uni-marburg.de> wrote:
>> Hi,
>> 
>> which of these times are the correct ones?
> 
> I looked up another example for which the submitting user records
> submission, start and end times outside of gridengine:
> 
> jbh at tin ~ $ qacct -j 135828 | grep time
> qsub_time    Sun Mar 28 18:07:41 2010
> start_time   Sun Mar 28 15:58:58 2010
> end_time     Mon Mar 29 00:12:32 2010
> ru_utime     114183.990
> ru_stime     872.392
> 
> But the user recorded:
> 
> Submission: 2010-03-28 18:07:45
> Start: 2010-03-29 01:59:05
> End: 2010-03-29 10:12:38

Maybe the submission time is recorded by the headnode, but the start/stop-time by the exechost. Was a different timezone active when the sgeqmaster was started? Nevertheless it's different from job to job?

-- Reuti


> Which implies start and end times in gridengine are being adjusted 10
> hours into the past in the accounting file for this job.
> 
>> You can export TZ on the command line and set it to GMT+2 or GMT-5 or so and the output of qacct will change accordingly. Neverthless it should be consistent.
> 
> Yeah, all numbers off by the same offset would have made sense.
> 
>>> I've looked at the submission hosts and they all have the correct time
>>> to within a few seconds, as do the compute nodes and qmaster and all
>>> are in the same time zone. Am I reading the accounting file wrong? I'm
>>> assuming that the fields are in the same order as the accounting
>>> manpage and am using this to calculate the pending time with field 10
>>> as starttime and field 9 as submit time:
>>> 
>>> awk -F ":" '{ print $10-$9 }' accounting
>> 
>> Hence you got a negative value here?
>> 
> 
> Correct. Using the 10 hours from the above example, I tried:
> 
> awk -F ":" '{ print $10-$9+36000 }' accounting | sort -n | uniq -c | less
> 
> which produces fewer negative entries than without the adjustment, but
> shows that 10 hours is not a consistent offset among jobs with this
> accounting problem.
> 
> Also, this is 6.2u5 using the courtesy binaries and all running on
> CentOS 5.4 x86_64.
> 
> Thanks,
> 
> jbh
> 
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=251722
> 
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=251733

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list