Am 13.12.2011 um 16:50 schrieb Christoph Müller: > Hi Reuti, > >> -----Ursprüngliche Nachricht----- >> Von: Reuti [mailto:[email protected]] >> Gesendet: Dienstag, 13. Dezember 2011 16:46 >> An: Christoph Müller >> Cc: [email protected] >> Betreff: Re: [gridengine users] Access complex resources from prolog script > > >>> My prolog script (with write access) parses the machine file and opens a >> SSH session to perform the changes. The problem is that this SSH session >> does not have the environment of the final job, i. e. I do not have variables >> like $JOB_ID etc. I have implemented a nasty hack that passes selected >> variables into the session, but it does not work for variables that differ >> from >> node to node. Therefore, I have two questions: >> >> from where to where are you making the ssh connection. The spool directory >> should be writable on the node where the job runs anyway. The prolog is >> executed there too. When you run the prolog as the admin user of SGE, >> there shouldn't be any access problem. > > From the prolog script. My problem is that the script seems to be executed on > the first node (where I make my mpirun call),
Correct. > but I need to change all nodes in a different way. I do not have any file > access problems, the problem is the context/the SGE envvars, which are not > available in my SSH session. Mhh - I'm still puzzeled at what point you need the ssh. To start the slaves of the parallel job? To change the environment file on the slave nodes? If you setup SGE's configuration to use ssh (in case you really need it), all variables should be inherited from the sge_shepherd. -- Reuti _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
