I rescheduled most of my cronjobs to 22 and 23 hours (7 and 6 jobs).
21.03.2019 7:46, Giovanni Tirloni пишет:
While trying to troubleshoot performance issues on our NFS server, it
became apparent that we have a peak of requests durings hours 0-3
every day. This is currently hard to properly manage and results in
slower performance for everybody.
Here's a breakdown of the number of cronjobs scheduled to start in
the new SGE cluster, per hour:
473 * # hourly
You can see that, apart from hourly jobs, we have most cronjobs
starting in hours 1-3.
My advice is that, if your workload permits, *schedule your cronjobs
outside of that time window so you are not competing for resources
with everybody else*. The more spread throughout the day your cronjobs
are, the better.
Also beware of step values like "*/2" which will make your cronjob
run on hours 0 and 2 too.
If you have any questions, please find us on #wikimedia-cloud