Purpose
Replacing an OSD is routine and is the operation most often botched under time pressure. The two classic mistakes are purging the OSD before backfill has finished, which removes a copy the cluster was still relying on, and replacing the disk without ever asking why it died, which schedules the same replacement again next month. This checklist is the guard rail around both.
Order of work
This list is sequential, not thematic, and the frontmatter is in execution
order. identify fixes which OSD and which physical device - do it from the
tree and the device path together, because pulling the wrong disk from a
degraded cluster is how a single failure becomes data loss. out-wait,
stop-purge, disk-replace and recreate are the replacement proper, and
the wait inside out-wait is a hard gate rather than a courtesy. verify
and monitor close the loop on placement and recovery. rca is last and is
not optional: without it, the drive model that just failed is still in the
purchase order.
Before you start
Have the OSD id, the host, the device path, the drive serial, and confirmation that no other OSD is currently out.
Verification
Done when the new OSD is up and in, every affected PG is active+clean, and the cluster is HEALTH_OK - all three, in that order. Mark an item N/A rather than leaving it unchecked when it does not apply to this cluster.
Escalation
If a second OSD fails while the first is being replaced, stop and escalate. Two concurrent failures in one failure domain is an incident, not maintenance.