A failover cluster fails in a few seams: quorum math, Cluster Shared Volume access, the
live-migration path, and the storage under CSV. Read the cluster's own evidence before
anyone moves a role.
Climb the Troubleshooting Ladder base skill first, with these specifics. History: a recent
patch (clustering is sensitive to mixed build levels during CAU), a storage, firmware or
networking change. Documentation: node count (it drives quorum, so establish it
first), Windows Server build, witness type — disk, file share or cloud — the storage
backend, CSV layout, and the migration network. Liongard, where present, gives node and
cluster state, dated (Inspector Read Discipline base skill).
Evidence: node, role and resource states (Get-ClusterNode, Get-ClusterGroup,
Get-ClusterResource), CSV states, quorum and votes (Get-ClusterQuorum, DynamicWeight), the
cluster log, and System and FailoverClustering events.
1. Quorum loss or the cluster won't form — count votes: nodes plus witness must exceed half.
A lost witness with an even split, or too many nodes down, drops it below quorum and it
stops on purpose. Restore the missing votes rather than forcing. Start-ClusterNode
-FixQuorum is a last resort that can cause split-brain: use it only when you know which
partition holds the authoritative data, and escalate that decision.
2. CSV redirected or offline — Redirected Access means IO routes over the network to the
coordinator node instead of direct: often a backup snapshot in progress, a storage-path
loss on one node, or antivirus scanning the CSV. Confirm no backup job is running before
treating a redirect as a fault. No Access or offline is a real connectivity or
reservation problem — check MPIO, iSCSI and HBA paths per node. SAN and fabric faults
are the storage owner's and the vendor's.
3. Live-migration failures — read the exact error: authentication or delegation (CredSSP or
Kerberos constrained delegation), a migration-network problem, insufficient memory or
NUMA on the target, or a processor-compatibility mismatch. A failed migration usually
leaves the VM on the source — verify before retrying.
4. A drain or pause that won't complete — a role won't move: anti-affinity, a resource that
won't come online elsewhere, misconfigured possible owners, or a VM pending. Read which
role blocks it and why. Never force-stop a node mid-drain to get patching moving;
that can hard-stop VMs.
Don't fail over or restart nodes during a storage event — stabilize storage first. Success is the cluster's own report: all nodes Up, roles Online on intended owners,
CSVs in direct access, quorum healthy, and a clean test migration if that was the fault.
Cluster Validation reports a config problem, it does not fix one. Note in plain text (PSA
Note Discipline base skill): build, quorum state, evidence, branch, action, verification.
Magic Library