Purpose
CephFS puts a metadata server in front of RADOS, and that changes the failure model: clients block on the MDS, not on the OSDs. Most CephFS incidents are a single active MDS with no standby to fail over to, or a metadata pool that ran out of room while the data pool still looked fine. This checklist is the gate for handing a filesystem to real users.
Order of work
The list runs in four passes. The daemon pass is mds-active-standby, and it
comes first because nothing below it matters on a filesystem that cannot
survive one MDS restart. The storage pass is pools: metadata and data are
separate pools with separate utilisation and often separate device classes, so
both are inspected. The tenancy pass is caps and quotas, which together
decide what a client can reach and how much of it a client can consume. The
data-protection pass is snapshots and backup - a snapshot policy is a
rollback mechanism, not an off-cluster copy, and the two items are not
substitutes for each other.
Before you start
Have the client inventory with the mount path each one uses, the tenant list the quotas are meant to divide, and the device class each pool sits on.
Verification
The filesystem is ready when every critical item is checked and the standby MDS
has been proven by an actual failover, not by its presence in ceph mds stat.
Mark an item N/A rather than leaving it unchecked when it does not
apply to this cluster.
Escalation
A filesystem with no standby MDS does not go to production. Take that to the storage lead before scheduling the handover.