Infrastructure » sepia

To replace a disk (from memory, may be missing a step or have wrong order):

• unweight the OSD if it's still viable (perhaps gradually over time): ceph osd reweight <float-less-than-1.0> <osdnum>
• once data is off, mark OSD out (sets weight to 0): ceph osd out <osdnum>
• remove from crush: ceph osd crush remove <osd.num>
• remove from cluster: ceph osd rm <osdnum>
• remove from auth: ceph auth del <osd.num>

The last three steps might be doable with ceph-disk osd destroy, which still isn't part of ceph-deploy AFAICT

• unmount the osd data directory

and then it should be ready to remove.