Purpose
Proxmox makes Ceph easy to stand up and easy to stand up badly. The hyperconverged layout means guest traffic, cluster traffic and OSD replication traffic all start out on the same wire, and the same three nodes are both the compute failure domain and the storage failure domain. This checklist is the gate before guests are migrated onto RBD storage.
Order of work
The list runs in four passes. The isolation pass is storage-networks and
crush-failure-domain: separate the storage path from the guest path, and
confirm the CRUSH rule spreads replicas across something that is not a single
rack. These come first because both are painful to change once guests are
running. The pool pass is pools and vm-pg-affinity, which cover how guest
disks are divided and how they are placed. The caveat pass is multipath,
which is not a defect to fix but an assumption to record - RBD has no multipath
layer, and the operator needs to know that rather than discover it. The
operations pass is monitoring.
Before you start
Have the node inventory with rack positions, the VLAN plan for guest and storage traffic, and the tenant-to-pool mapping.
Verification
Ready when every critical item is checked and a live migration has been tested with a guest actually running on RBD storage. Mark an item N/A rather than leaving it unchecked when it does not apply to this cluster.
Escalation
Shared guest and storage networking is an accepted risk that needs recording with the virtualisation owner before any guest is migrated.