Feature #11202
open
Maybe we could use "ceph osd set noscrub" or "ceph osd unset noscrub" to achieve this.
- Status changed from New to Fix Under Review
- Assignee set to Xinze Chi
- Status changed from Fix Under Review to 12
- Status changed from 12 to New
We have set "ceph osd set noscrub" and "ceph osd set no-deep-scrub" flags:
noscrub,nodeep-scrub flag(s) set
And PGs are keep getting deep scrubbed. There is a rebalance (backfilling) going on. IIRC deep scrubs get started to verify integrity of the PG on the new OSD. Is that (still) true? I don't think this makes sense anymore with bluestore ... as it's already doing CRC checks etc. while doing the backfilling, right? If not, it would make sense to do this instead of doing this work twice.
Anyway ... when those flags are set the cluster should NOT keep scheduling deep-scrubs ... and it's still doing this ... really annoying because it slows down backfilling, potentially triggering SLOW_OPS (in our case it does) and or even cause the "OSD::osd_op_tp thread" to timeout ... leading to issue #41255 for at least Mimic (13.2.8). When backfilling is done the balancer might want to tune things ... but won't run because (deep-)scrubs ...
Also available in: Atom
PDF