Currently, the general recommendation is to change the scheduler back to wpq instead of mclock, this allows you to control backfill behaviour with the usual configs like osd_recovery_max_active, osd_recovery_max_active_hdd, osd_recovery_max_active_ssd, osd_max_backfills. Zitat von Nicola Mori <mori@fi.infn.it>:

Dear Ceph users

yesterday I added a new host to my cluster. This triggered a rebalance and now backfill is in progress. What I'm seeing is that many PGs are in backfill_wait state, and just a few are actually backfilling:

pgs: 35081735/215211118 objects misplaced (16.301%) 348 active+remapped+backfill_wait 174 active+clean 7 active+remapped+backfilling

I prioritized the recovery by setting the high_recovery_ops scheduler profile, but this just increased the recovery rate a bit while not incrementing the number of actually backfilling PGs.

Is there a way to increment the number of PGs that backfill at the same time? I guess this should speed up the operations, but I didn't find a way to do so. Currently I have very little load from users so I can devote all the resources to the recovery.

Using Ceph 19.2.0 managed by cephadm.

Thanks,

Nicola _______________________________________________ Dev mailing list -- dev@ceph.io To unsubscribe send an email to dev-leave@ceph.io