Struct GossipHandler 

pub struct GossipHandler { /* private fields */ }

Authoritative owner of PeerState transitions for the gossip plane.

The handler holds an Arc<ServerPool> and feeds the per-peer phi-accrual failure detectors as gossip frames arrive. A periodic tick re-evaluates phi for every non-local peer and toggles PeerState between Normal and Down based on the configured threshold:

• a peer is Normal once at least one heartbeat has been recorded AND phi(now) <= threshold,
• a peer is Down when no heartbeat has ever been recorded OR phi(now) > threshold.

The handler is the single place that mutates peer.state once gossip is wired; the supervisor loop that owns the TCP link no longer publishes peer-state transitions of its own.

Examples

use std::sync::Arc;
use dynomite::cluster::gossip::GossipHandler;
use dynomite::cluster::peer::{Peer, PeerEndpoint};
use dynomite::cluster::pool::{PoolConfig, ServerPool};
use dynomite::hashkit::DynToken;

let cfg = PoolConfig::default();
let local = Peer::new(
    0, PeerEndpoint::tcp("h".into(), 1), "r".into(), "d".into(),
    vec![DynToken::from_u32(0)], true, true, false,
);
let pool = Arc::new(ServerPool::new(cfg, vec![local]));
let handler = GossipHandler::new(pool);
assert!((handler.threshold() - 8.0).abs() < f64::EPSILON);

Implementations

impl GossipHandler

pub fn new(pool: Arc<ServerPool>) -> Self

Build a fresh handler over pool using the default phi-accrual threshold (crate::cluster::failure_detector::DEFAULT_THRESHOLD).

pub fn with_failure_metrics(self, metrics: Arc<FailureMetrics>) -> Self

Attach a crate::stats::FailureMetrics handle.

When set, Self::evaluate emits a peer_state_transitions_total counter tick and a peer_state_current gauge update for every transition it applies, plus a gossip_phi_score gauge update for every non-local peer regardless of whether its state changed. Default behaviour is unchanged when no metrics handle is supplied.

pub fn with_events(self, events: Arc<EventManager>) -> Self

Attach an EventManager handle.

When set, every peer-state transition the handler applies surfaces a ClusterEvent::PeerUp or ClusterEvent::PeerDown payload on the manager’s broadcast. Default behaviour is unchanged when no event manager is supplied.

pub fn events(&self) -> Option<&Arc<EventManager>>

Borrow the installed event manager, if any.

pub fn with_threshold(self, threshold: f64) -> Self

Override the phi threshold (default 8.0).

pub fn with_interval(self, interval: Duration) -> Self

Override the gossip interval used by the periodic tick when the handler is driven by the binary’s run loop. The in-process tests do not depend on this value.

pub fn threshold(&self) -> f64

Phi threshold the handler is configured with.

pub fn interval(&self) -> Duration

Configured gossip interval.

pub fn pool(&self) -> &Arc<ServerPool>

Borrow the underlying pool.

pub fn record_heartbeat_pname(&self, pname: &str, now: Instant)

Record an inbound gossip heartbeat from the peer identified by pname (a host:port string matching the peer’s crate::cluster::peer::PeerEndpoint::pname).

Mutates the peer’s failure detector and immediately promotes the peer’s state to PeerState::Normal when phi(now) is below the threshold; this gives gossip a snappy first-contact transition without waiting for the next periodic tick.

Unknown pnames are ignored.

pub fn record_heartbeat_idx(&self, peer_idx: u32, now: Instant)

Record an inbound gossip heartbeat against a known peer index. Used by tests and by callers that already resolved the originating peer.

pub fn evaluate(&self, now: Instant) -> Vec<(u32, PeerState)>

Walk every non-local peer and reconcile its PeerState with the failure detector’s current view of phi(now). Returns the list of (peer_idx, new_state) transitions the call applied (handy in tests).

This is the failure-detector tick the binary runs on a periodic timer. Calling it never panics and it never blocks on I/O.

pub fn mark_down_pname(&self, pname: &str)

Mark the peer identified by pname as PeerState::Down without consulting the failure detector. Used by the gossip-shutdown path so the dispatcher can short-circuit routing to a peer that announced its own departure.

pub fn reset_detector(&self, peer_idx: u32)

Reset the per-peer failure detector. Used when a peer is removed and re-added so historical jitter does not bias the new suspicion value.

Trait Implementations

impl Debug for GossipHandler

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.