HA vs maintenance
The HA stack guards nodes with a watchdog and can fence (hard-reset) a node on trouble. During planned work — firmware updates, re-cabling network or storage, power tests — that behaviour can be dangerous: a maintenance action may look like a failure and trigger a node reset.
Release the watchdogs cluster-wide
Disarming HA (disarm) releases all CRM and LRM watchdogs across the whole cluster at once. That lets you do the work safely without risking any node being fenced. It's a cluster-wide approach, different from the existing per-node maintenance mode.
Freeze vs Ignore
When disarming HA you choose how HA-managed VMs are handled:
- Freeze — new commands and state changes aren't applied. Services stay where they are, but HA doesn't react to failures. The safest choice when all nodes should stay up.
- Ignore — resources are suspended from HA tracking; you can manually start, stop and migrate them like ordinary VMs. Pick this when you need to relocate services during the work.
Commands and coming back
You run the operation from the command line:
| Action | Command |
|---|---|
| Disarm HA (Ignore mode) | ha-manager crm-command disarm-ha ignore |
| Arm again | ha-manager crm-command arm-ha |
On re-arming (arm) the CRM rechecks service locations against the configuration, so it picks up any manual migrations done during the work.
Remember to arm HA again once the work is done — a disarmed cluster offers no failure protection. Best to bake it into the maintenance runbook.
We'll plan cluster work safely
We'll build HA maintenance procedures (disarm/arm), resource modes and a work order so servicing never triggers fencing — in line with your SLA.
⚡ Free consultation → HA affinity rules