9. Exclude Workstation Nodes from Longhorn Storage

Status: Accepted

Context

Workstation nodes (tagged workstation: true in hosts.yml) may reboot for updates or user activity. Longhorn replica data stored on these nodes is at risk during unplanned reboots, degrading volume health until the node returns.

Decision

Disable Longhorn scheduling on workstation nodes and request eviction of existing replicas. This is done via the Longhorn Node CR:

spec:
  allowScheduling: false
  evictionRequested: true

The Kubernetes workstation=true:NoSchedule taint already prevents Longhorn’s instance-manager pods from scheduling on workstation nodes after a fresh install, making this the natural default. The explicit Longhorn node setting handles the transition for existing clusters where replicas were placed before the taint was applied.

Longhorn DaemonSet components (manager, engine-image, csi-plugin) continue to run on all nodes as they tolerate all taints by default.

Consequences

• Longhorn will not place new replicas on workstation nodes

• Existing replicas are evicted and rebuilt on always-on nodes

• Volumes pinned to workstation nodes via PV node affinity will update to include other nodes once replicas are rebuilt

• Services that previously tolerated the workstation taint for storage access should have those tolerations removed once a dedicated worker is available