Print

Print


On Mon, Nov 13, 2006 at 10:27:54PM +0300, Valery Mitsyn wrote:

> On Mon, 13 Nov 2006, Rod Walker wrote:
> 
> >
> >Given that anyone can setsid() in a job, is it an option to look at the 
> >torque
> >code to fix both problems once and for all?
> 
> This problem with setsid and new session has been discussed
> in the torque mail-list without any real solution. It's known
> that torque can not deal properly with such jobs.

SGE (optionally) uses group id's to track processes, when the job
starts it gets en extra group id and when it's killed shepherd checks
all running jobs and kills every job that has that group id.

It is a bit of a hack (what do you do if a user is a member of the
maximum allowed groups? do you drop a random one? you do nothing?)
but it is far better than nothing.

It should not be that hard to do something similar in torque if you
are familiar with the code (I am not).

Kostas