Hi
Analyzing a single node crash in a PG, I've noticed that CRUSH might allocate an OSD which already exists in the PG to replace the failed OSD.
For Example, say this is how the PG is allocated: [1,2,3,4]
And say node 1 crashed, the initial allocation can be [2,2,3,4], meaning 2 was assigned instead of 1.
Once CRUSH will detect 2 is assigned twice, it will replace it as well.
Each such replacement is regarded as another crash, requiring additional traffic.
Is there a way to force CRUSH to allocate only non-used OSDs to a failed node in a PG?
Thanks
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
