Hi All, Strange behavior that I noticed for previous post for simple jobs occurs only if a user has a complex, multi-part profile (but without any input). If profile is minimal (single file) or default for RHEL, then everything works fine.
Strange behavior of execd (inability to communicate with the master for installations which use sgeadmin account) is connected with the fact that on execution host RPM creates sgeadmin user account using regular useradd command without coordination with the values on master host. This probably should be made into a manual step as such help is worse then no help at all. If discrepancy occurs, execd can't access properly $SGE_ROOT/$SGE_CELL subtree and this behavior occurs. This bug can be reproduced if you shut down execution daemon on the node, change value of GID and UID to non-matching with master host and start the daemon. One minor nuisance. Even with -nobincheck option you still need to copy several files from the master/opt/sge/utilbin/lx-amd64/ directory (only two files are needed) to run execution host installation: spooldefaults spoolinit Those two files probably should be added to execd RPM. Regards, Nikolai Nikolai N Bezroukov/BASF-CORP/BASF wrote on 11/07/2014 09:17:49 AM: > From: Nikolai N Bezroukov/BASF-CORP/BASF > To: Reuti <re...@staff.uni-marburg.de>@BASF-COM > Cc: users@gridengine.org > Date: 11/07/2014 09:17 AM > Subject: Re: [gridengine users] Strange error output on testing of > Son of Grid Engine 8.1.8 after install > > Hi Reuti, > > > You can execute `hostname` when you log in to a node? Is it working > > if you submit a script calling `hostname` instead of submitting it > > as a binary? > > > > -- Reuti> > > No. It's the same behavior for scripts with this version as with > commands (I use simple.sh test for scripts). > > I also checked 8.1.7 it demonstrated the same behavior too. What I > noticed that this bug appears only if you install SoGE as root . > > And BTW in this case unless you share whole /opt/sge tree, execution > daemon installer complains about missed executables in in bin and > utilbin directories, despite the fact that those missing are not > needed on execution node and that's why they were not installed by RPMs. > > But if you install it as sgeadmin then computational node in my case > can't connect to the master, while on the master (if you install > execution daemon on it) both test run OK. qping said that > communication rejected. > > Regards, > Nikolai > > Reuti <re...@staff.uni-marburg.de> wrote on 11/07/2014 05:39:57 AM: > > > From: Reuti <re...@staff.uni-marburg.de> > > To: Nikolai N Bezroukov <nikolai.bezrou...@basf.com> > > Cc: users@gridengine.org > > Date: 11/07/2014 05:40 AM > > Subject: Re: [gridengine users] Strange error output on testing of > > Son of Grid Engine 8.1.8 after install > > > > Hi, > > > > Am 06.11.2014 um 23:54 schrieb Nikolai N Bezroukov: > > > > > After the install of SoGE 8.1.8 on RHEL 6.6 a simple submission of > > the command via qsub produces error output: > > > > > > qsub -b y -cwd hostname > > > > > > There is no output and error messages tell something about ioctl. > > But hostname command does not read anything from the terminal. > > > > > > -rw-r--r--. 1 test test 290 Nov 6 17:47 hostname.e5 > > > -rw-r--r--. 1 test test 0 Nov 6 17:47 hostname.o5 > > > > > > $ cat hostname.e5 > > > stty: standard input: Inappropriate ioctl for device > > > stty: standard input: Inappropriate ioctl for device > > > stty: standard input: Inappropriate ioctl for device > > > stty: standard input: Inappropriate ioctl for device > > > > You can execute `hostname` when you log in to a node? Is it working > > if you submit a script calling `hostname` instead of submitting it > > as a binary? > > > > -- Reuti > > > > > > > Regards, > > > - Nikolai > > > _______________________________________________ > > > users mailing list > > > users@gridengine.org > > > https://gridengine.org/mailman/listinfo/users > >
_______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
