Level-triggered automatic rebalancing (opt-in). A periodic reconcile that, on the lease holder, commits the current rebalancing plan once it has stayed the same for stabilization consecutive ticks, so a transient SWIM membership flap (a node briefly suspected, then alive again) never triggers a vnode move. This is the k8s-controller pattern: SWIM does the fast, event-driven detection; the decision to move vnodes reconciles desired-vs-actual and converges.

It is policy on top of the Malachi.Cluster.RebalanceCoordinator mechanism (which stays manual-by-default): the seams :plan_fun / :commit_fun / :leader? wire to the coordinator and the lease holder, so the timing logic is testable by driving one reconcile with the timer out of the way. Off by default: nothing moves unless it is enabled and this node holds the lease.

Options

• :plan_fun - (-> [Rebalance.change()]), the current plan (wired to RebalanceCoordinator.plan/1)
• :commit_fun - (-> commit_result), commit under the lease (wired to RebalanceCoordinator.commit/1)
• :leader? - (-> boolean()), whether this node holds the lease (only the leader drives)
• :interval - ms between reconciles (default 30_000)
• :stabilization - consecutive identical, non-empty plans required before committing (default 3)
• :on_result - (commit_result -> any), called with each commit result (default: logs)
• :name - optional registered name

The tick, the validated interval and the rule that an unknown message is counted and survived rather than fatal all come from Malachi.Cluster.PeriodicWorker.