Struct ClusterSupervisor 

pub struct ClusterSupervisor<L: PeerLiveness, A: ShardAdopter> { /* private fields */ }

Watches peer liveness and auto-adopts a dead peer’s shards (SS-5b).

Implementations

impl<L: PeerLiveness, A: ShardAdopter> ClusterSupervisor<L, A>

pub fn new( liveness: Arc<L>, adopter: Arc<A>, peers: Vec<WatchedPeer>, config: SupervisorConfig, ) -> Self

Build a supervisor over peers, polling liveness and calling adopter.adopt_shards on confirmed peer death. Peers with no owned shards are dropped from the watch set (nothing to adopt for them).

pub fn with_publisher( self, publisher: Arc<ClusterEventPublisher>, self_node: impl Into<String>, ) -> Self

Attach the WS3 cluster-event publisher and this node’s name so tick() emits topology deltas. Pure builder addition — a supervisor without it behaves exactly as before (every existing test passes new only).

pub fn watches_any(&self) -> bool

Whether this supervisor watches any peer (false when no peer declared owned shards — the loop would do nothing, so the caller can skip spawning).

pub fn adopter(&self) -> &A

Borrow the adopter (the engine, in production) this supervisor drives. Lets a test inspect the engine it auto-adopts onto after the loss.

pub async fn tick(&mut self) -> Vec<String>

Run ONE poll tick: observe every watched peer’s liveness, advance the debounce counters, and adopt the shards of any peer that has now been down for confirmations consecutive ticks and is not yet adopted.

Returned is the list of peer names adopted on THIS tick (empty on a quiet tick), so a test can assert exactly when adoption fires. Extracted from the loop so the debounce decision is unit-testable without real time.

pub async fn run(self, shutdown: Receiver<bool>)

Drive the poll loop until shutdown flips true, ticking every poll_interval. Consumes self; spawn it as a background task.