Hi everyone,
We recently started to use a priority-based scheduling and after solving some
final issues (see this post:
https://groups.google.com/forum/m/#!topic/slurm-users/N8r8MoyjQAU), everything
seems to be running quite smoothly now. However, we realized that the data
shown by sshare, e.g.
Account User RawShares NormShares RawUsage
EffectvUsage FairShare
root 0.000000
8484544 1.000000
root root 1 0.500000
0 0.000000 1.000000
iasteam 1 0.500000
8484544 1.000000
iasteam carvalho 1 0.250000
1550368 0.182729 0.400000
iasteam hany 1 0.250000
0 0.000000 0.800000
iasteam pascal 1 0.250000
6934176 0.817271 0.200000
iasteam stark 1 0.250000
0 0.000000 0.800000
is only updated in very long intervals. This means that the current RawUsage of
e.g. user ‚pascal‘ stays a very long time on 6934176, and then jumps to the
next value, say 7238923, where it then again waits a long time until it is
updated. Different from this behavior, the data shown by sacct is updated every
second.
We already tried reducing the update interval of sshare by adjusting the
JobAcctGatherFrequency, but this did not help in our case. Also my attempts to
look for similar questions had no success. Can anybody help us out here and
point us to the correct option that we need to change to get everything running
smoothly?
P.S.: Our config is the same as in the post that I linked (except for the
proposed fix in the corresponding thread obviously).
Best,
Pascal
