feat(go): improve realtime pipeline resilience

go/vehicle-gateway/internal/gateway/mqtt_client.go

@@ -179,6 +179,12 @@ func (c *MQTTClient) buildTLSConfig() (*tls.Config, error) {
 }
 
 func (c *MQTTClient) handleMessage(ctx context.Context, topic string, payload []byte) {
+	started := time.Now()
+	frameStatus := envelope.ParseBadFrame
+	defer func() {
+		c.recordFrameDuration(frameStatus, time.Since(started))
+	}()
+
 	messageCtx, cancelMessage := context.WithTimeout(context.WithoutCancel(ctx), frameOperationTimeout)
 	defer cancelMessage()
 
@@ -214,6 +220,7 @@ func (c *MQTTClient) handleMessage(ctx context.Context, topic string, payload []
 			c.recordIdentityMetric(identityStatus(env))
 		}
 	}
+	frameStatus = env.ParseStatus
 	c.recordFrameMetric(env.ParseStatus)
 	if err := c.cfg.Sink.PublishRaw(messageCtx, env); err != nil {
 		c.recordPublishMetric("raw", "error")
@@ -252,6 +259,16 @@ func (c *MQTTClient) recordFrameMetric(status envelope.ParseStatus) {
 	})
 }
 
+func (c *MQTTClient) recordFrameDuration(status envelope.ParseStatus, elapsed time.Duration) {
+	if c.cfg.Metrics == nil {
+		return
+	}
+	c.cfg.Metrics.SetGauge("vehicle_gateway_frame_duration_ms", metrics.Labels{
+		"protocol": string(envelope.ProtocolYutongMQTT),
+		"status":   string(status),
+	}, float64(elapsed.Milliseconds()))
+}
+
 func (c *MQTTClient) recordPublishMetric(kind string, status string) {
 	if c.cfg.Metrics == nil {
 		return

go/vehicle-gateway/internal/gateway/mqtt_client_test.go

@@ -79,6 +79,7 @@ func TestMQTTClientRecordsMessageMetrics(t *testing.T) {
 		`vehicle_gateway_frames_total{protocol="YUTONG_MQTT",status="OK"} 1`,
 		`vehicle_gateway_identity_total{protocol="YUTONG_MQTT",status="resolved"} 1`,
 		`vehicle_gateway_publish_total{kind="raw",protocol="YUTONG_MQTT",status="ok"} 1`,
+		`vehicle_gateway_frame_duration_ms{protocol="YUTONG_MQTT",status="OK"}`,
 	} {
 		if !strings.Contains(text, want) {
 			t.Fatalf("metrics missing %s:\n%s", want, text)

go/vehicle-gateway/internal/gateway/tcp_server.go

@@ -195,6 +195,12 @@ func (s *TCPServer) handleConnection(ctx context.Context, conn net.Conn) {
 }
 
 func (s *TCPServer) handleFrame(ctx context.Context, conn net.Conn, raw []byte, source string, state *connectionState) {
+	started := time.Now()
+	frameStatus := envelope.ParseBadFrame
+	defer func() {
+		s.recordFrameDuration(frameStatus, time.Since(started))
+	}()
+
 	frameCtx, cancelFrame := context.WithTimeout(context.WithoutCancel(ctx), frameOperationTimeout)
 	defer cancelFrame()
 
@@ -229,6 +235,7 @@ func (s *TCPServer) handleFrame(ctx context.Context, conn net.Conn, raw []byte,
 		}
 		enrichConnectionPlatform(&env, state)
 	}
+	frameStatus = env.ParseStatus
 	s.recordFrameMetric(env.ParseStatus)
 
 	if err := s.sink.PublishRaw(frameCtx, env); err != nil {
@@ -307,6 +314,16 @@ func (s *TCPServer) recordFrameMetric(status envelope.ParseStatus) {
 	})
 }
 
+func (s *TCPServer) recordFrameDuration(status envelope.ParseStatus, elapsed time.Duration) {
+	if s.metrics == nil {
+		return
+	}
+	s.metrics.SetGauge("vehicle_gateway_frame_duration_ms", metrics.Labels{
+		"protocol": string(s.protocol.Protocol),
+		"status":   string(status),
+	}, float64(elapsed.Milliseconds()))
+}
+
 func (s *TCPServer) recordPublishMetric(kind string, status string) {
 	if s.metrics == nil {
 		return

go/vehicle-gateway/internal/gateway/tcp_server_test.go

@@ -81,6 +81,7 @@ func TestTCPServerRecordsFrameMetrics(t *testing.T) {
 		`vehicle_gateway_frames_total{protocol="JT808",status="OK"} 1`,
 		`vehicle_gateway_identity_total{protocol="JT808",status="unresolved"} 1`,
 		`vehicle_gateway_publish_total{kind="raw",protocol="JT808",status="ok"} 1`,
+		`vehicle_gateway_frame_duration_ms{protocol="JT808",status="OK"}`,
 	} {
 		if !strings.Contains(text, want) {
 			t.Fatalf("metrics missing %s:\n%s", want, text)