Hi, > Am 23.02.2017 um 14:50 schrieb Szoke Igor <i...@darkfuture.cz>: > > Hi, thanks Reuti. > > I just want to log job stdout / stderr to some nice job related directories, > which I cannot create in prior as I have noNFS on the target machine.
How do you get the results of the jobs then? When all is written on the nodes, how will the users access the results? Also all input files and programs need to be present on the nodes then too. Or you need some means to copy them between the machines: https://arc.liv.ac.uk/SGE/howto/filestaging/ Submitting jobs with DRMAA and not by a plain `qsub` will set some variables to allow easier file staging: http://gridscheduler.sourceforge.net/howto/filestaging/filestaging6.html > I'll study the JSV. Then this won't work too. The server side JSV will be executed on the qmaster, and when he has no access to the nodes directories, he can't create anything there on. It was only an approach to cover non-existing directories which can also happen in case you have NFS on the nodes. -- Reuti > Thanks! > -Igor > > > On 02/23/2017 02:25 PM, Reuti wrote: >> Hi, >> >>> Am 23.02.2017 um 13:55 schrieb Szoke Igor <i...@darkfuture.cz>: >>> >>> Hi, >>> >>> I'm running SGE in noNFS mode. >>> There is obvious problem, that if I submit a job with -e -o pointing to a >>> directory, which does not exists, the job fails (Error state). >> This will happen in NFS mode too. >> >> So there is also no NFS for the home directories of the user on the nodes? >> Whereto should the stdout/stderr be written then? >> >> >>> I have tried to use prolog, but the result is the same. (it make sense >>> according to the documentation) >>> >>> I'm using a /tmp/ direcotry now, but I would like to use some other one >>> directory (depending on job name etc... ). >> But it should still be local on the node only? >> >> Essentially this would mean to have one directory per job (unless jobs have >> the same name). The default consisting of the job name and the job number is >> not sufficient? There are some pseudo variables to assemble the proper name >> on your own at runtime (see `man qsub` for the "-e" option). >> >> >>> Is that a way how to: >>> 1] make a -o -e directory for logging prior to SGE tries to make a file >>> there? >>> 2] force SGE to create the directory if it does not exists? (which is an >>> interesting feature in my opinion) >>> >>> (I do not want 2 jobs solution, where I submit 1. job making these >>> directories and then submit 2. one with -e -o to these directories). >> Depending on the actual need, I could think of putting this logic in an JSV, >> which will create the directory during job submission by reading the the >> "-e"/"-o" options. Sure, if the user deletes the directory before the job >> starts you are out of luck with this approach. >> >> -- Reuti >> >> >>> Thanks! >>> >>> -Igor >>> _______________________________________________ >>> users mailing list >>> users@gridengine.org >>> https://gridengine.org/mailman/listinfo/users >
signature.asc
Description: Message signed with OpenPGP using GPGMail
_______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
