[torqueusers] Prologue/Epilogue script

Adam Emerich aemerich at us.ibm.com
Mon Apr 23 09:05:16 MDT 2007


Jeff,

Here is what we use:

#!/bin/ksh

if [ "$1" = "root" ]
then
      echo "$0: cannot kill root!"
      exit 1
fi

if [ -z "$1" ]
then
      echo "$0: kill who?"
      exit 1
fi

HN=$(hostname)

for i in $(ps aux | grep "^$1 " | grep -v grep | awk '{print $2}')
do
      if ps -p $i >/dev/null 2>&1
      then
            echo "killing $HN $i"
            kill $i 2>&1
      fi
      if ps -p $i >/dev/null 2>&1
      then
            echo "KILLING $HN $i"
            kill -9 $i 2>&1
      fi
done

Hope that helps.

Adam Emerich


"Insanity: doing the same thing over and over again and expecting different
results."   -Albert Einstein


                                                                           
             "Jeffrey B.                                                   
             Layton"                                                       
             <laytonjb at charter                                          To 
             .net>                     Adam Emerich/Rochester/IBM at IBMUS    
                                                                        cc 
             04/23/2007 08:36          torqueusers-bounces at supercluster.or 
             AM                        g, torqueusers at supercluster.org     
                                                                   Subject 
                                       Re: [torqueusers] Prologue/Epilogue 
             Please respond to         script                              
             laytonjb at charter.                                             
                    net                                                    
                                                                           
                                                                           
                                                                           
                                                                           




Adam Emerich wrote:
> Jeff,
>
> We have scripts that kill all the user's processes as the job exits.
This
> is how we ensure that only the user who currently has the node scheduled
to
> him is running on the node.  As for killing jobs in the prologue, unless
> you do not have the ability to have multiple users on a node, this would
> seem like it could cause problems.
>

I'm assuming that even if you get only 1 core on the node, you "own" the
node
(you may need the memory per core for a code).

> Let me know if you are interested in the 'killuser' script.
>

You bet!

Thanks!

Jeff





More information about the torqueusers mailing list