Struct NodeHandle 

pub struct NodeHandle {
    pub instance_id: NodeId,
    pub gossip_addr: String,
    pub controller: Arc<ClusterController>,
    /* private fields */
}

One instance of the mini-cluster: a gossip discovery + controller.

Fields

instance_id: NodeId

This node’s identity.

gossip_addr: String

Gossip address this node listens on.

controller: Arc<ClusterController>

The cluster control facade. Arc-shared so tests can observe leader status while the harness retains ownership.

Implementations

impl NodeHandle

pub async fn kill(self)

Gracefully stop this node. Before shutting down the gossip socket, publishes state = Left via chitchat so peers see the status change via the next gossip round (~100 ms), rather than waiting for phi-accrual to flag the missing heartbeats.

This matches the contract other production gossip stacks (Serf, Akka cluster) provide — clean leave is a protocol primitive, not a timing hack. Tests that want to simulate a crash (no clean leave) should use Self::crash instead.

pub async fn crash(self)

Simulate a crash: shut down this node without announcing a Left state. Peers rely on phi-accrual to eventually detect the failure, which in the default chitchat config can take tens of seconds — exactly the behavior a real crash produces. Use Self::kill for scenarios that expect prompt failover.

Unlike a plain drop, this calls discovery.stop().await to shut down the chitchat background task and release the UDP socket. Skipping that was previously leaking tasks and ports between crash-based tests; the protocol-visible behavior matches a real crash either way because stop does not emit any Left announcement (that’s kill’s job).