fix(gateway): close Phase 3C audit gaps

2026-07-30 01:46:00 +07:00
parent 040ca73ce9
commit d3852d15f3
23 changed files with 1619 additions and 219 deletions
@@ -90,6 +90,8 @@ func heartbeatLoop(ctx context.Context, client *gateway.ControlPlaneClient, serv
 	ticker := time.NewTicker(2 * time.Second)
 	defer ticker.Stop()
 	var sequence int64
+	var sampler heartbeatSampler
+	_, _ = sampler.sample(time.Now(), server.Metrics())
 	for {
 		select {
 		case <-ctx.Done():
@@ -104,11 +106,82 @@ func heartbeatLoop(ctx context.Context, client *gateway.ControlPlaneClient, serv
 				state = "draining"
 			}
 			metrics := server.Metrics()
-			_ = client.Heartbeat(ctx, protocol.GatewayHeartbeat{Version: "1", GatewayID: registration.GatewayID, Sequence: sequence, ObservedAt: time.Now().UTC().Format(time.RFC3339Nano), ActiveConnections: metrics.ActiveSessions, EgressKbps: registration.BandwidthCapacityKbps, State: state})
+			observedAt := time.Now()
+			egressKbps, telemetry := sampler.sample(observedAt, metrics)
+			_ = client.Heartbeat(ctx, protocol.GatewayHeartbeat{
+				Version: "1", GatewayID: registration.GatewayID, Sequence: sequence,
+				ObservedAt: observedAt.UTC().Format(time.RFC3339Nano), ActiveConnections: metrics.ActiveSessions,
+				EgressKbps: egressKbps, State: state, Telemetry: telemetry,
+			})
 		}
 	}
 }

+type heartbeatSampler struct {
+	observedAt time.Time
+	mediaBytes uint64
+}
+
+func (s *heartbeatSampler) sample(observedAt time.Time, metrics gateway.MetricsSnapshot) (int64, protocol.GatewayTelemetry) {
+	egressKbps := int64(0)
+	elapsedMillis := observedAt.Sub(s.observedAt).Milliseconds()
+	if !s.observedAt.IsZero() && elapsedMillis > 0 && metrics.MediaBytes >= s.mediaBytes {
+		delta := metrics.MediaBytes - s.mediaBytes
+		milliseconds := uint64(elapsedMillis)
+		whole, remainder := delta/milliseconds, delta%milliseconds
+		if whole > 125_000_000 {
+			egressKbps = 1_000_000_000
+		} else {
+			rate := whole*8 + remainder*8/milliseconds
+			if rate > 1_000_000_000 {
+				rate = 1_000_000_000
+			}
+			egressKbps = int64(rate)
+		}
+	}
+	s.observedAt, s.mediaBytes = observedAt, metrics.MediaBytes
+	return egressKbps, protocol.GatewayTelemetry{
+		AdmittedSessions: boundedMetric(metrics.AdmittedSessions), AdmissionRejects: boundedMetric(metrics.AdmissionRejects),
+		Reconnects: boundedMetric(metrics.Reconnects), DrainTransitions: boundedMetric(metrics.DrainTransitions),
+		MediaDrops: boundedMetric(metrics.MediaDrops), MediaPackets: boundedMetric(metrics.MediaPackets),
+		MediaBytes: boundedMetric(metrics.MediaBytes), QueueDelayMicros: boundedMetric(metrics.QueueDelayNanos / 1000),
+		ProcessingDelayMicros: boundedMetric(metrics.ProcessingDelayNanos / 1000), ProcessingSamples: boundedMetric(metrics.ProcessingSamples),
+		PacingDelayMicros: boundedMetric(metrics.PacingDelayNanos / 1000), ProviderErrors: boundedMetric(metrics.ProviderErrors),
+		InputRejected: boundedMetric(metrics.InputRejected), ControlRttMicros: boundedMetric(metrics.ControlRTTNanos / 1000),
+		ControlJitterMicros: boundedMetric(metrics.ControlJitterNanos / 1000), ControlLossPpm: boundedMetric(metrics.ControlLossPPM),
+		PendingReliable: boundedMetric(metrics.PendingReliable), ProviderState: providerStateName(metrics.ProviderState),
+	}
+}
+
+func boundedMetric(value uint64) int64 {
+	const maximum = uint64(^uint64(0) >> 1)
+	if value > maximum {
+		return int64(maximum)
+	}
+	return int64(value)
+}
+
+func providerStateName(value uint64) string {
+	switch value {
+	case 1:
+		return gateway.ProviderStateStarting
+	case 2:
+		return gateway.ProviderStateReady
+	case 3:
+		return gateway.ProviderStateDisconnected
+	case 4:
+		return gateway.ProviderStateTerminating
+	case 5:
+		return gateway.ProviderStateTerminated
+	case 6:
+		return gateway.ProviderStateCleanup
+	case 7:
+		return gateway.ProviderStateFailed
+	default:
+		return "unknown"
+	}
+}
+
 func loadTLS(certFile, keyFile, clientCAFile string) (*tls.Config, *tls.Config, error) {
 	certificate, err := tls.LoadX509KeyPair(certFile, keyFile)
 	if err != nil {