On Thu, Nov 3, 2016 at 11:57 AM, Wido den Hollander <wido@xxxxxxxx> wrote: > >> Op 3 november 2016 om 10:46 schreef Wido den Hollander <wido@xxxxxxxx>: >> >> >> >> > Op 3 november 2016 om 10:42 schreef Dan van der Ster <dan@xxxxxxxxxxxxxx>: >> > >> > >> > Hi Wido, >> > >> > AFAIK mon's won't trim while a cluster is in HEALTH_WARN. Unset >> > noscrub,nodeep-scrub, get that 3rd mon up, then it should trim. >> > >> >> The 3rd MON is back, but afaik the MONs trim when all PGs are active+clean. A cluster can go into WARN state for almost any reason, eg old CRUSH tunables. >> >> Will give it a try though. > > No, it didn't. Health is OK, but still, the MON stores will not trim. A manual compaction actually grew the store from 25GB to 39GB. > > They keep having a high amount of Paxos keys in the MON stores. > How does the min_last_epoch_clean (from ceph pg dump) compare with your current osdmap epoch? That will confirm if it *should* trim. Note that, AFAICT, ceph pg dump only gives the correct min_last_epoch_clean if you ask the mon leader. Peons always return 0, for some reason. Cheers, Dan _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |