Also want to add that qrsh with command (qrsh with command and without command are actually 2 different execution paths) accepts environment variables passed from the submission side. You can do something like:
% qrsh -v PATH=hahaha /usr/bin/printenv PATH /tmp/17.1.all.q:hahaha Rayson On Fri, Apr 20, 2012 at 5:27 PM, Reuti <[email protected]> wrote: > Am 20.04.2012 um 22:50 schrieb Burian, John: > >> When I run: >> >> $ qrsh set >> >> The PATH that appears in the output invariably contains whatever the TMPDIR > > It's a scratch directory for the job created and removed by SGE in "tmpdir > /tmp" location of the queue definition on the exechost. It's often used for > scratch data of the job, and it's nice that the users don't have to think > about removing the stuff from the nodes after their job ran. > > >> is, and "/usr/local/bin:/bin:/usr/bin". It seemingly ignores both the system >> and user profile, bashrc, csh.login, etc. Where does qrsh get it's path? >> Jobs submitted with qsub get the correct PATH. > > The user's profiles will be sourced only for `qsub` if: > > - queue is set to "shell_start_mode posix_compliant" > - the used shell (-S or the setting in the queue definition: "shell /bin/sh") > is in "login_shells sh,ksh,csh,tcsh" in SGE's configuration > > otherwise a minimal $PATH is used, like for `qrsh` with command. > > == > > I prefer "shell_start_mode unix_behavior" in the queue definition and > self-contained scripts. Otherwise a changed setting in the profile might > trash the job execution and it's hard to investigate if you rely on the > settings therein (the same job script might run for user A but not for B, as > he changed something in his profile). > > -- Reuti > _______________________________________________ > users mailing list > [email protected] > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
