feat(go): expose bridge batch pressure metrics

This commit is contained in:
lingniu
2026-07-03 18:58:37 +08:00
parent 16ee09180c
commit 018015f9de
6 changed files with 97 additions and 0 deletions

View File

@@ -240,20 +240,29 @@ func bridgeBatch(ctx context.Context, registry *metrics.Registry, writer kafkaBa
if len(messages) == 0 {
return nil
}
started := time.Now()
status := "ok"
defer func() {
recordBridgeBatchDuration(registry, status, time.Since(started))
recordBridgeBatchPending(registry, 0, 0)
}()
kafkaMessages := make([]kafka.Message, 0, len(messages))
for _, message := range messages {
addBridgeSubjectMetric(registry, "vehicle_bridge_messages_total", message.subject, "received")
topic, ok := route[message.subject]
if !ok || topic == "" {
addBridgeSubjectMetric(registry, "vehicle_bridge_messages_total", message.subject, "route_error")
status = "error"
return fmt.Errorf("kafka topic not configured for nats subject %q", message.subject)
}
kafkaMessages = append(kafkaMessages, kafkaMessage(topic, message.data))
}
recordBridgeBatchPending(registry, len(messages), len(kafkaMessages))
if err := writer.WriteMessages(ctx, kafkaMessages...); err != nil {
for _, message := range kafkaMessages {
addBridgeTopicMetric(registry, "vehicle_bridge_kafka_writes_total", message.Topic, "error")
}
status = "error"
return err
}
for _, message := range kafkaMessages {
@@ -265,6 +274,7 @@ func bridgeBatch(ctx context.Context, registry *metrics.Registry, writer kafkaBa
}
if err := message.ack(); err != nil {
addBridgeSubjectMetric(registry, "vehicle_bridge_nats_acks_total", message.subject, "error")
status = "error"
return err
}
addBridgeSubjectMetric(registry, "vehicle_bridge_nats_acks_total", message.subject, "ok")
@@ -272,6 +282,8 @@ func bridgeBatch(ctx context.Context, registry *metrics.Registry, writer kafkaBa
return nil
}
var bridgeBatchDurationBucketsMS = []float64{1, 5, 10, 25, 50, 100, 250, 500, 1000, 5000}
func addBridgeSubjectMetric(registry *metrics.Registry, name string, subject string, status string) {
if registry == nil {
return
@@ -286,6 +298,23 @@ func addBridgeTopicMetric(registry *metrics.Registry, name string, topic string,
registry.IncCounter(name, metrics.Labels{"topic": topic, "status": status})
}
func recordBridgeBatchPending(registry *metrics.Registry, messages int, kafkaMessages int) {
if registry == nil {
return
}
registry.SetGauge("vehicle_bridge_batch_pending_messages", nil, float64(messages))
registry.SetGauge("vehicle_bridge_batch_pending_kafka_messages", nil, float64(kafkaMessages))
}
func recordBridgeBatchDuration(registry *metrics.Registry, status string, elapsed time.Duration) {
if registry == nil {
return
}
registry.ObserveHistogram("vehicle_bridge_batch_duration_ms_histogram", metrics.Labels{
"status": status,
}, bridgeBatchDurationBucketsMS, float64(elapsed.Milliseconds()))
}
func recordNATSConsumerInfoMetrics(registry *metrics.Registry, cfg config, info *nats.ConsumerInfo) {
if registry == nil || info == nil {
return

View File

@@ -91,6 +91,49 @@ func TestBridgeBatchRecordsMetrics(t *testing.T) {
}
}
func TestBridgeBatchExposesPendingAndDurationMetrics(t *testing.T) {
env := envelope.FrameEnvelope{Protocol: envelope.ProtocolGB32960, VIN: "VIN001", MessageID: "0x02"}
payload, err := json.Marshal(env)
if err != nil {
t.Fatal(err)
}
registry := metrics.NewRegistry()
writer := &recordingBridgeWriter{
onWrite: func() {
text := registry.Render()
for _, want := range []string{
`vehicle_bridge_batch_pending_messages 2`,
`vehicle_bridge_batch_pending_kafka_messages 2`,
} {
if !strings.Contains(text, want) {
t.Fatalf("pending bridge metric missing %s during write:\n%s", want, text)
}
}
},
}
err = bridgeBatch(context.Background(), registry, writer, []bridgeMessage{
{subject: "vehicle.raw.go.gb32960.v1", data: payload, ack: func() error { return nil }},
{subject: "vehicle.raw.go.gb32960.v1", data: payload, ack: func() error { return nil }},
}, map[string]string{"vehicle.raw.go.gb32960.v1": "vehicle.raw.go.gb32960.v1"})
if err != nil {
t.Fatalf("bridgeBatch() error = %v", err)
}
text := registry.Render()
for _, want := range []string{
`vehicle_bridge_batch_pending_messages 0`,
`vehicle_bridge_batch_pending_kafka_messages 0`,
`vehicle_bridge_batch_duration_ms_histogram_bucket{le="+Inf",status="ok"} 1`,
`vehicle_bridge_batch_duration_ms_histogram_count{status="ok"} 1`,
`vehicle_bridge_batch_duration_ms_histogram_sum{status="ok"}`,
} {
if !strings.Contains(text, want) {
t.Fatalf("bridge batch metric missing %s:\n%s", want, text)
}
}
}
func TestRecordNATSConsumerInfoMetrics(t *testing.T) {
registry := metrics.NewRegistry()
cfg := config{NATSStream: "VEHICLE_INGEST", NATSDurable: "vehicle-kafka-bridge"}
@@ -148,9 +191,13 @@ func TestLoadConfigIncludesUnifiedOnlyWhenExplicitlyConfigured(t *testing.T) {
type recordingBridgeWriter struct {
messages []kafka.Message
err error
onWrite func()
}
func (w *recordingBridgeWriter) WriteMessages(_ context.Context, messages ...kafka.Message) error {
if w.onWrite != nil {
w.onWrite()
}
if w.err != nil {
return w.err
}