Struct FailureDetector 

pub struct FailureDetector { /* private fields */ }

Heartbeat-based failure detector.

Tracks node health via heartbeat messages and detects failures.

Implementations

impl FailureDetector

pub fn new() -> Self

Create a new failure detector with default config

pub fn with_config(config: FailureDetectorConfig) -> Self

Create with custom config

pub fn on_failure<F>(self, f: F) -> Self

where F: Fn(u64) + Send + Sync + 'static,

Set failure callback

pub fn on_recovery<F>(self, f: F) -> Self

where F: Fn(u64) + Send + Sync + 'static,

Set recovery callback

pub fn heartbeat(&self, node_id: u64, addr: SocketAddr)

Record a heartbeat from a node

Previously the recovery callback was invoked inside entry().and_modify(...), which holds the DashMap shard’s write lock. A user-supplied callback that re-entered the same shard (or any structure ordered against it) deadlocked; even without deadlock, every concurrent heartbeat hashing to the same shard stalled while the callback ran. The fix collects a “should I notify?” flag inside the closure and fires the callback after the and_modify returns, releasing the shard lock.

pub fn check_all(&self) -> Vec<u64>

Check all nodes for failures

Callbacks are now invoked after the iter_mut loop has dropped its shard locks. Previously cb(*entry.key()) ran inside the iteration, with the per-shard write lock still held — a user-supplied callback that touched another DashMap entry on the same shard (or re-entered the failure detector itself via heartbeat / status) would deadlock. We collect the failed ids first, release the iteration locks, then fire the callbacks.

pub fn status(&self, node_id: u64) -> NodeStatus

Get node status

pub fn failed_nodes(&self) -> Vec<u64>

Get all failed nodes

pub fn suspected_nodes(&self) -> Vec<u64>

Get all suspected nodes

pub fn healthy_nodes(&self) -> Vec<u64>

Get all healthy nodes

pub fn remove(&self, node_id: u64)

Remove a node from tracking

pub fn cleanup(&self) -> usize

Clean up stale entries (nodes that have been failed for too long)

pub fn stats(&self) -> FailureStats

Get statistics

nodes_tracked reads the O(1) counter; the per-status tally is still a single pass over the entries. That scan is observability-only (not on any hot path) and is deliberately NOT replaced by per-status counters: node status is mutated in-place via get_mut().status = ... in tests and could be elsewhere, which would silently drift maintained counters. The scan is always exact.

pub fn node_count(&self) -> usize

Get node count

Trait Implementations

impl Default for FailureDetector

fn default() -> Self

Returns the “default value” for a type.