Hi all,
I can now add another data point as well. We upgraded our production cluster
from mimic to octopus with the procedure of
- set quick-fix-on-start=false in all ceph.conf files and the mon config store
- set nosnaptrim
- upgrade all daemons
- set require-osd-release=octopus
- host by host:
Hosts buckets with no disks don't hurt, but they do influence the crush map.
To avoid unnecessary data movement, you might want to change your server
replacement method. Since mimic, instead of purging OSDs, you can destroy OSDs,
which is the same as purge but keeps the ID intact. This way, you