autonomy ha split-brain recover

Recover from a split-brain condition on this node

Synopsis

Sends a recovery request to the control-plane split-brain recovery endpoint.

Two strategies are supported:

promote-leader — Forcibly clears the local node’s stale leadership claim by calling ForceResign on its elector. The node stops believing it holds write authority. The legitimate leader (which already acquired the advisory lock after the failover) continues unimpeded.

                  Use this when 'split-brain detect' reports risk=detected
                  (epoch divergence) or risk=possible (holder mismatch or
                  unclosed epochs) on the node you are targeting.

manual-reconcile — Dry-run: runs detection, plans the recovery actions, and returns a description of what promote-leader would do — without executing anything. Use this to verify conditions before committing.

Preconditions:

  • The target node must report risk != “none” on split-brain detect, unless –force is specified.

  • operator is required for audit trail.

Recovery is performed on the node that receives the request. To recover a stale leader, send this request to that specific node’s control-plane URL.

After recovery, run ‘autonomy ha status’ to verify the cluster has converged to a single leader with a consistent epoch.

Usage

autonomy ha split-brain recover [flags]

Options

      --force                     override the risk=none guard; allow recovery even when no split-brain is detected
      --operator string           operator identity for audit trail (required)
      --orchestrator-url string   orchestrator base URL (env: AUTONOMY_ORCHESTRATOR_URL)
      --output string             output format: "text" (human-readable) or "json" (machine-readable) (default "text")
      --reason string             reason for recovery (included in audit trail)
      --strategy string           recovery strategy: "promote-leader" (execute) or "manual-reconcile" (dry-run) (required)

See also