Hello, this question was brought many times before, and also solved in a various ways - snap trimmer, scheduler` priorities and persistent fix (for a ReplicatedPG issue), but it seems that the current Ceph versions may suffer as well during the rollback operations on large images and on large scale. Given CFQ scheduler for rotating media and ~10 percentage of an utilization as an initial preconditions, the rollback operation of an one-fourth terabyte image over 100 OSDs may result in a significant latency impact and, for such configuration, breakage of a 30s request completion barrier. Although recent improvements did very well in means of the congestion control for a slow media for many kind of non-client ops, this exact issue remains. I think it can be solved by another sleeper knob but unsure where its proper place should be. Thanks for suggestions! _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |