On Fri, 17 Apr 2009, Douglas McNab wrote:
> I have been noticing that occasionally that I get broken pipes from glexec
> when I submit a job i.e
>
> 2009-04-17 23:24:07,810 FATAL - MethodName=[jobRegister] Timestamp=[Fri 17
> Apr 2009 23:24:07] ErrorCode=[0] Description=[system error]
> FaultCause=[cannot write the job wrapper (jobId = CREAM736347427)! The
> problem seems to be related to glexec which reported: Broken pipe]
>
> I looked in the glexec_lcas_lcmaps.log and found nothing strange, even with
> the verbosity turned up.
>
> I then discovered in /opt/glite/var/cream_sandbox/ the pool account
> directories. I was running this job in the scotgrid VO and as such was
> mapped to scotg.
> I must have been previously mapped to a different grid account as glexec was
> currently mapping me to scotg001 but there was file already in this
> directory:
>
> drwx------ 3 scotg074 scotg 4096 Mar 3 17:22
> C_UK_O_eScience_OU_Glasgow_L_Compserv_CN_douglas_mcnab_vo.scotgrid.ac.uk_Role_NULL_Capability_NULL
>
> I decided to remove this sandbox and this fixed job submission as it was
> then able to create:
>
> drwx------ 3 scotg001 scotg 4096 Apr 17 23:30
> C_UK_O_eScience_OU_Glasgow_L_Compserv_CN_douglas_mcnab_vo.scotgrid.ac.uk_Role_NULL_Capability_NULL
>
> Is this the desired behaviour? I thought pool account mappings in the
> gridmapdir were recycled?
Yes, but there is an open bug:
https://savannah.cern.ch/bugs/?47278
Were _all_ scotgrid accounts in use? If so, you should create more:
for good security practices recycling should be a rare event...
|