Hey Chris,
On 07/12/15 19:04, Christopher Samuel wrote:
Hi Danny,
On 10/07/15 22:30, Danny Auble wrote:
What is your jobaccountgather set to in your slurm.conf?
I think this is what you mean, I believe it's the default.
JobAcctGatherFrequency=30
We're using cgroups to gather the info:
JobAcctGatherType=jobacct_gather/cgroup
I would recommend jobacct_gather/linux over cgroup. cgroup adds quite a
bit of overhead with very little benefit, but this was what I was
looking for.
Stats should only show up from jobs launched from srun.
Do you mean during runtime, or at completion?
sstat will work while the step is running and sacct will start to
display the stats after completion.
MPI jobs launched either directly with srun or via Open-MPI's mpirun
(which uses srun to invoke orted to launch MPI ranks) do not report any
stats until completion.
Expected, no stats are sent to the database until after the step
completes, use sstat beforehand.
We do get stats for jobs that just use sbatch (and no srun) once they
complete.
Expected, for the batch step anyway.
The manual page for sbatch says that memory stats are collected by
default every 30 seconds.
Correct, they are. Just use sstat if you want to see any running step's
info.
All the best,
Chris
