>-----Original Message-----
>From: Reuti [mailto:[email protected]]
>Sent: Tuesday, March 01, 2011 9:00 AM

>HI,

>Am 01.03.2011 um 17:50 schrieb Adams, Jeb:

>> On our dev box we have installed GE 6.2u5 three or four times over the last 
>> year or so, in various locations. We do not run as root, so we have to 
>> apportion a user/group to own the installation. At one point last year it 
>> was installed using ge62tst as the owner and group
>>
>>
>> (from /etc/passwd)
>> ge62tst:x:59820:412:GE62_tst_acct:/home/ge62tst:/bin/bash

>you are running local only? The output of:

>$ id ge62tst
>$ id sgeadmin

>show proper values?

uid=59820(ge62tst) gid=412(ge62tst) groups=412(ge62tst) 
context=user_u:system_r:unconfined_t:s0
uid=411(sgeadmin) gid=411(sgeadmin) groups=411(sgeadmin) 
context=user_u:system_r:unconfined_t:s0
These look OK.


>> It was installed in /tmp and a cron job ate most of the installation. Flash 
>> forward and we have re-installed using sgeadmin 
>> (sgeadmin:x:411:411::/home/sgeadmin:/bin/bash). Part of our install has the 
>> whole directory chown’d and chgrp’d recursively to this account
>>
>>  drwxr-xr-x  3 sgeadmin sgeadmin 4.0K Feb 24 15:54 ge62/
>>
>> We are finding though, that permissions issues plague us—you can only run in 
>> a 777 directory and that ge62tst is doing the read/writes.
>>
>>
>> -rw-r--r-- 1 ge62tst   24001    0 Feb 28 16:55 simple.sh.e28
>>
>> -rw-r--r-- 1 ge62tst   24001   58 Feb 28 16:55 simple.sh.o28

>It's the group of the user when the job started. This has nothing to do with 
>the owner of the SGE installation. You started the sgeexecd from the root 
>account after making the changes?

>-- Reuti
(from ll /etc/init.d |grep sge)
-rwxr-xr-x 1 root root  9114 Oct  6 17:55 sgeexecd
-rwxr-xr-x 1 root root 11149 Oct 18 11:11 sgeexecd.pkdmsbx
-rwxr-xr-x 1 root root 22859 Feb 24 15:33 sgemaster.pkdms1
-rwxr-xr-x 1 root root 22867 Oct 18 10:52 sgemaster.pkdmsbx

The problem is definitely in here somewhere:
/home/jeba > sudo /sbin/service sgeexecd softstop
/etc/init.d/sgeexecd: line 83: /opt/sge/util/arch: No such file or directory
/etc/init.d/sgeexecd: line 89: /opt/sge/util/arch: No such file or directory
/etc/init.d/sgeexecd: line 94: =$:/opt/sge/lib/: No such file or directory
export LANG="en_US.UTF-8"
export OLDPWD
export PATH="/bin:/usr/bin:/sbin:/usr/sbin"
export PWD="/"
export SGE_CELL="default"
export SGE_ROOT="/opt/sge"
export SHLVL="1"
export TERM="vt100"
can't determine path to Grid Engine binaries

/home/jeba > echo $SGE_CELL
sbx
/home/jeba > echo $SGE_ROOT
/opt/ge62/ge6.2u5

sgeexecd is using an old config.
You can see the detritus from earlier installs there. I think the first was a 
6.1 build from more than 18 months ago.
<snip>
>From: Laotsao [mailto:[email protected]]
>Sent: Tuesday, March 01, 2011 9:25 AM


>Please read the installation guide one more time
>Files are own by uid but one need to install by root, otherwise only uid can 
>submit jobs
>You can install multiple versions at different CELL and  with different PORT 
>for master and  comm
>You use the shell scripts under SGE_CELL/common to set the correct PATH etc
>Different version can co-exist in the same cluster
>Keep in mind different version all share the same CPU in exced-hosts unless 
>you configure different exec-hosts for different versions

This advice, while accurate, is not germane. I don't want more than one 
installation of Grid Engine. The shell scripts under common are fine, and are 
resposible for my accurate $SGE_CELL and $SGE_ROOT echos above. The issue is 
the service is still "owned" by an earlier install and I can't seem to softstop 
it anymore, as it's desynched from the current executables.

All uids in operators can submit jobs currently, but the GE itself is not 
passing the uid to the execution layer, it's using the ge62tst account, which 
needs to be able to r/w wherever you are running.

Perhaps my best option is to unistall GE and start over? I am worried that the 
"zombie" services will be left behind.

Jeb Adams
PKDM IS Liaison
~~~~~~~~~~~~~~~~~~
[email protected]<mailto:[email protected]>
805.313.5022
PKDM IS 
Portal<http://myteams.amgen.com/sites/TS/PKDM/PKDMDept/PKDMFunctions/IS/default.aspx>
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to