On 19/08/16 20:32, Dr. Thomas Orgis wrote: > The (for a Torque/Maui user) simple idea of appending to the user's job > log file(s) from prologue/epilogue scripts seems not to be part of > Slurm's mindset. I consider it only natural that printouts to stdout > and stderr from the prologue/epilogue scripts end up somehow in the job > log files.
Unfortunately you can only do this as part of the taskprolog, so prepending to the users stdout. We've been doing prepending forever with Slurm on our Blue Gene/Q, our taskprolog.sh has this in it: # Add a banner to the job standard output echo "print Scratch directory /scratch/avoca/$SLURM_JOB_ID has been allocated" echo "print $SLURM_BG_NUM_NODES Blue Gene/Q compute nodes have been allocated" That file is triggered with: TaskProlog=/usr/local/slurm/etc/taskprolog.sh Just be warned that on an x86-64 cluster that will get triggered for every task in every job step (i.e. srun) in the job, so you might want to be careful how you use that! Fortunately that's not an issue on BG/Q as it only runs once on the launch node as srun talks to the BG/Q control system to launch code. I've tested on our test x86-64 cluster and it looks like it's the only place where that's implemented (so far at least). Hope that helps! Chris -- Christopher Samuel Senior Systems Administrator VLSCI - Victorian Life Sciences Computation Initiative Email: [email protected] Phone: +61 (0)3 903 55545 http://www.vlsci.org.au/ http://twitter.com/vlsci
