[GE users] TMPDIR

Reuti reuti at staff.uni-marburg.de
Thu Sep 13 14:49:50 BST 2007


Hi,

Am 13.09.2007 um 15:03 schrieb Colin Thomas:

> Many thanks for your reply.
>
> Is KEEP_ACTIVE an execd parameter ? If so I can see how (via qmon) it
> can be changed in the cluster configuration.
>
> So I "modify" a single machine, add KEEP_ACTIVE=1 to the
> execd_parameters , click okay , but it then says that the "host  
> already
> exists" even though I was modifying it :-(
>
> What is an alternative way of altering a machines's execd parameter  
> (not
> through qmon).

I would use the value TRUE instead. But the real reason is this:

http://gridengine.sunsource.net/issues/show_bug.cgi?id=2080

-- Reuti


> Best regards
>
> Colin Thomas
>
> -----Original Message-----
> From: Reuti [mailto:reuti at staff.uni-marburg.de]
> Sent: 13 September 2007 13:37
> To: users at gridengine.sunsource.net
> Subject: Re: [GE users] TMPDIR
>
> Hi,
>
> Am 13.09.2007 um 14:03 schrieb Colin Thomas:
>
>> I have a question about TMPDIR.
>>
>> When job is started the TMPDIR is set to
>>
>> <temp dir from queue config>/<gridId>.<queue Name>
>>
>> We have a qsh job, which creates an xterm.
>
> how? Is it jumping out of the process tree - like started with & ?
> You can try to set ENABLE_ADDGRP_KILL in the SGE configuration for
> the execd_params so that the xterm is killed when the main job
> completed.
>
> -- Reuti
>
> PS: There is the option KEEP_ACTIVE to avoid the removal, but this
> way you would end up a) with many orphaned directories on the nodes
> over time, and b) violating SGE's policy, as the xterm might still
> generate load on the system but SGE will put already another job on  
> it.
>
> ...
>
>> The xterm is inheriting  the
>> TMPDIR that the qsh set up. Then the qsh job completes (with the  
>> xterm
>> still running), and the TMPDIR is removed. The xterm then has no
>> TMPDIR
>> to write to. This did start a conversation saying the the qsh job
>> should
>> have know that the xterm child process was still active, and so  
>> should
>> not have cleaned up anyway..
>>
>> I have looked (and failed) to find a switch to disable the TMPDIR  
>> from
>> being deleted : is there such a thing ?
>>
>> I did note some talk about in messages concerning the notify setting
>> from the queue configuration (which is set to 60sec by default) but
>> changing this does not seem to postpone the TMPDIR from being  
>> cleaned.
>
> PPS: Depends on the handling of the signal by the main job. If it
> can't handle it, it will simply quit and tell SGE as a result that it
> finished already. If you ignore the generated signal in the
> jobscript, then during the grace period before being killed finally,
> the TMPDIR should still exist.
>
>> I also tried a prolog script to try and set it to a value that I
>> wanted,
>> but this route also did not work.
>>
>> Any thoughts ? (we are 6.03u)
>>
>> Best regards
>>
>> Colin Thomas
>>
>>
>
>
> .
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
>

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list