committed

A hot Flow of committed instructions in index order, delivered as Committed values.

Every node in the cluster — leader and followers alike — emits the same sequence here. Collect this flow to drive a state machine: apply Committed.Entry entries and reset to Committed.Install snapshots as they arrive.

Raft's own bookkeeping — the internal §5.4.2 election no-op (LogEntry.isNoOp) and §6 membership-change entries (LogEntry.config) — is withheld: its contents never appear here, so a consumer applying Committed.Entry sees only application commands and need not filter. Application-proposed entries always surface, including those with an empty LogEntry.command.

Those entries still occupy a committed index, so each surfaces as a payload-free Committed.Internal marker in its own position in the stream. Fold it by advancing your applied index and doing nothing else — then every index commitIndex counts is accounted for, which is what makes an applied prefix comparable to readIndex's fence (#1718).

Replay=0; late collectors miss entries emitted before they subscribed. To resume without gaps after subscribing late (e.g. on crash recovery), use committedFrom.