My guess is Galaxy is deleting the directory because it believe the
job is in error because of some communication problem while polling
your DRM via DRMAA - Galaxy thinks the job has failed before it has
even been run.

You can set

cleanup_job = never

in universe_wsgi.ini's app:main section to instruct Galaxy to not
delete the working directory. I suspect this will allow the DRM to
finish running your job - but Galaxy is still going to fail it since
it cannot properly detect its status.

Can you confirm?

-John






On Thu, Jun 12, 2014 at 8:36 AM, Evan Bollig <boll0...@umn.edu> wrote:
> job_conf.xml is outside of my knowledge. Better wait to see what the
> others can tell us.
>
> -E
> -Evan Bollig
> Research Associate | Application Developer | User Support Consultant
> Minnesota Supercomputing Institute
> 599 Walter Library
> 612 624 1447
> e...@msi.umn.edu
> boll0...@umn.edu
>
>
> On Thu, Jun 12, 2014 at 8:31 AM, Shrum, Donald C <dcsh...@admin.fsu.edu> 
> wrote:
>> It's set to 600 seconds so I don't think that is the issue... Is there some 
>> sort of wait time to set in job_conf.xml
>>
>> -----Original Message-----
>> From: Evan Bollig [mailto:boll0...@umn.edu]
>> Sent: Thursday, June 12, 2014 9:27 AM
>> To: Shrum, Donald C
>> Cc: galaxy-dev@lists.bx.psu.edu
>> Subject: Re: [galaxy-dev] jobs submitted to a cluster
>>
>> Hey Donny,
>>
>> What is the value of keep_completed on your queue (from qmgr -c 'p s')? 
>> Could it be that your spool is flushing completed jobs immediately? I ran 
>> into issues the other day with libdrmaa requiring at least keep_complete = 
>> 60 seconds to properly detect completed jobs and clean up after itself.
>>
>> Cheers,
>>
>> -E
>>
>>
>> -Evan Bollig
>> Research Associate | Application Developer | User Support Consultant 
>> Minnesota Supercomputing Institute
>> 599 Walter Library
>> 612 624 1447
>> e...@msi.umn.edu
>> boll0...@umn.edu
>>
>>
>> On Thu, Jun 12, 2014 at 7:36 AM, Shrum, Donald C <dcsh...@admin.fsu.edu> 
>> wrote:
>>> I've setup galaxy to submit jobs to our HPC cluster as the logged in user.  
>>> I used the drama python module to submit the jobs to our moab server.
>>>
>>> It appears that the working directory for a submitted job is being removed 
>>> by galaxy prior to the job completing on the cluster.
>>>
>>> I can see a working directory is created in the logs:
>>> galaxy.jobs DEBUG 2014-06-12 08:21:03,786 (15) Working directory for
>>> job is:
>>> /panfs/storage.local/software/galaxy-dist/database/job_working_directo
>>> ry/000/15
>>>
>>> I've confirmed the directory is created by watching on the file system and 
>>> within about two seconds of the folder being created it is deleted.
>>> [root@admin 000]# watch -d ls -lR
>>> Every 2.0s: ls -lR                                                          
>>>                                                             Thu Jun 12 
>>> 08:21:06 2014
>>> total 64
>>> drwxrwxrwx 2 dcshrum dcshrum 4096 Jun 12 08:21 15
>>>
>>>
>>> I see the job sent via DRMAA:
>>> galaxy.jobs.handler DEBUG 2014-06-12 08:21:03,795 (15) Dispatching to
>>> drmaa runner galaxy.jobs.runners.drmaa DEBUG 2014-06-12 08:21:05,566
>>> (15) submitting file
>>> /panfs/storage.local/software/galaxy-dist/database/job_working_directo
>>> ry/000/15/galaxy_15.sh galaxy.jobs.runners.drmaa DEBUG 2014-06-12
>>> 08:21:05,566 (15) native specification is: -N galaxyjob -l
>>> nodes=1,walltime=2:00 -q genacc_q galaxy.jobs.runners.drmaa DEBUG
>>> 2014-06-12 08:21:05,892 (15) submitting with credentials: dcshrum
>>> [uid: 232706] galaxy.jobs.runners.drmaa INFO 2014-06-12 08:21:06,196
>>> (15) queued as 7570705.moab.local
>>>
>>> The job fails:
>>> galaxy.jobs.runners.drmaa DEBUG 2014-06-12 08:21:06,698
>>> (15/7570705.moab.local) state change: job finished, but failed 
>>> galaxy.jobs.runners DEBUG 2014-06-12 08:21:07,124 (15/7570705.moab.local) 
>>> Unable to cleanup 
>>> /panfs/storage.local/software/galaxy-dist/database/job_working_directory/000/15/galaxy_15.sh:
>>>  [Errno 2] No such file or directory: 
>>> '/panfs/storage.local/software/galaxy-dist/database/job_working_directory/000/15/galaxy_15.sh'
>>>
>>>
>>> I can see the same error in my moab log:
>>> *** error from copy
>>> /bin/cp: cannot create regular file
>>> `/panfs/storage.local/software/galaxy-dist/database/job_working_direct
>>> ory/000/15/galaxy_15.o': No such file or directory
>>> *** end error output
>>>
>>>
>>> Any idea as to why galaxy removes the working directory?  Is there a 
>>> setting in the job_conf.xml that would resolve this?
>>>
>>> Thanks for any pointers.
>>>
>>> Donny
>>> FSU Research Computing Center
>>>
>>>
>>> ___________________________________________________________
>>> Please keep all replies on the list by using "reply all"
>>> in your mail client.  To manage your subscriptions to this and other
>>> Galaxy lists, please use the interface at:
>>>   http://lists.bx.psu.edu/
>>>
>>> To search Galaxy mailing lists use the unified search at:
>>>   http://galaxyproject.org/search/mailinglists/
>>
>>
>> ___________________________________________________________
>> Please keep all replies on the list by using "reply all"
>> in your mail client.  To manage your subscriptions to this
>> and other Galaxy lists, please use the interface at:
>>   http://lists.bx.psu.edu/
>>
>> To search Galaxy mailing lists use the unified search at:
>>   http://galaxyproject.org/search/mailinglists/
>
> ___________________________________________________________
> Please keep all replies on the list by using "reply all"
> in your mail client.  To manage your subscriptions to this
> and other Galaxy lists, please use the interface at:
>   http://lists.bx.psu.edu/
>
> To search Galaxy mailing lists use the unified search at:
>   http://galaxyproject.org/search/mailinglists/

___________________________________________________________
Please keep all replies on the list by using "reply all"
in your mail client.  To manage your subscriptions to this
and other Galaxy lists, please use the interface at:
  http://lists.bx.psu.edu/

To search Galaxy mailing lists use the unified search at:
  http://galaxyproject.org/search/mailinglists/

Reply via email to