Zakhar, try to look to top of slow ops in daemon socket for this osd, you may find 'snapc' operations, for example. By rbd head you can find rbd image, and then try to look how much snapshots in chain for this image. More than 10 snaps for one image can increase client ops latency to tens milliseconds even for NVMe drives, that usually operates at usec or 1-2msec k Sent from my iPhone > On 7 Oct 2022, at 14:35, Zakhar Kirpichenko <zakhar@xxxxxxxxx> wrote: > > The drive doesn't show increased utilization on average, but it does > sporadically get more I/O than other drives, usually in short bursts. I am > now trying to find a way to trace this to a specific PG, pool and object > (s) – not sure if that is possible. _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |