+autoremoval of stale discovery records

api/pkg/discovery/registry.go

@@ -181,6 +181,26 @@ func (r *Registry) Delete(key string) bool {
 	return true
 }
 
+// DeleteStale removes entries whose heartbeat timeout has elapsed.
+func (r *Registry) DeleteStale(now time.Time) []string {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	removed := make([]string, 0)
+	for key, entry := range r.entries {
+		if entry == nil || !entry.isStaleAt(now) {
+			continue
+		}
+		delete(r.entries, key)
+		r.unindexEntry(key, entry)
+		removed = append(removed, key)
+	}
+	if len(removed) == 0 {
+		return nil
+	}
+	return removed
+}
+
 func (r *Registry) List(now time.Time, onlyHealthy bool) []RegistryEntry {
 	r.mu.Lock()
 	defer r.mu.Unlock()
@@ -789,3 +809,17 @@ func (e *RegistryEntry) isHealthyAt(now time.Time) bool {
 	}
 	return now.Sub(e.LastHeartbeat) <= timeout
 }
+
+func (e *RegistryEntry) isStaleAt(now time.Time) bool {
+	if e == nil {
+		return false
+	}
+	if e.LastHeartbeat.IsZero() {
+		return true
+	}
+	timeout := time.Duration(e.Health.TimeoutSec) * time.Second
+	if timeout <= 0 {
+		timeout = time.Duration(DefaultHealthTimeoutSec) * time.Second
+	}
+	return now.Sub(e.LastHeartbeat) > timeout
+}