feat(go): expose async sink publish latency histogram

docs/ops/100k-capacity-baseline.md

@@ -64,6 +64,7 @@ systemd gateway 已配置 `LimitNOFILE=1048576`，需要持续保持。
 | Gateway async sink | `vehicle_async_sink_queue_depth{sink}` | 队列深度不持续增长 |
 | Gateway async sink | `vehicle_async_sink_enqueue_total{sink,kind,status}` | `timeout/closed` 不增长 |
 | Gateway async sink | `vehicle_async_sink_publish_total{sink,kind,status}` | `error` 不增长 |
+| Gateway async sink | `vehicle_async_sink_publish_duration_ms_histogram_bucket` | p99 不持续上升 |
 | Gateway | `vehicle_gateway_publish_total{status="ok"}` | 持续增长 |
 | Gateway | `vehicle_gateway_frame_duration_ms_histogram_bucket` | p99 小于容量目标 |
 | History writer | `vehicle_history_batch_flush_total{status}` | `error` 不增长 |

docs/ops/go-service-observability.md

@@ -37,6 +37,7 @@ curl -fsS http://127.0.0.1:20200/readyz
 | `vehicle_gateway_frames_total` | Protocol frames received and parsed by the gateway. Labels: `protocol`, `status`. |
 | `vehicle_gateway_frame_duration_ms` | Last observed Gateway frame handling duration and histogram buckets for p95/p99. Labels: `protocol`, `status`. |
 | `vehicle_gateway_publish_total` | Gateway publish result for raw events. Unified appears only when the compatibility stream is explicitly enabled. Labels: `protocol`, `kind`, `status`. |
+| `vehicle_async_sink_publish_duration_ms_histogram` | Async sink publish duration histogram covering worker publish calls to the delegate sink. Labels: `sink`, `kind`, `status`. |
 | `vehicle_bridge_messages_total` | NATS messages fetched by the bridge. Labels: `subject`, `status`. |
 | `vehicle_bridge_kafka_writes_total` | Bridge writes to Kafka. Labels: `topic`, `status`. |
 | `vehicle_bridge_nats_acks_total` | NATS ack results after Kafka write. Labels: `subject`, `status`. |

docs/ops/go-vehicle-ingest-memory.md

@@ -393,6 +393,9 @@ net.ipv4.tcp_tw_reuse = 1
 vehicle_async_sink_enqueue_total{sink,kind,status}
 vehicle_async_sink_publish_total{sink,kind,status}
 vehicle_async_sink_publish_duration_ms{sink,kind,status}
+vehicle_async_sink_publish_duration_ms_histogram_bucket{sink,kind,status,le}
+vehicle_async_sink_publish_duration_ms_histogram_count{sink,kind,status}
+vehicle_async_sink_publish_duration_ms_histogram_sum{sink,kind,status}
 vehicle_async_sink_queue_depth{sink}
 ```
 
@@ -404,7 +407,7 @@ vehicle_async_sink_queue_depth{sink}
 - NATS bridge ack pending 为 `0`
 - Gateway 启动后窗口无 `error|failed|panic|fatal` 日志
 
-这些指标是进入带帧率压测前的关键保护栏：如果 `queue_depth` 持续增长或 `enqueue_total{status="timeout"}` 增长，说明入口 publish 队列已成为瓶颈。
+这些指标是进入带帧率压测前的关键保护栏：如果 `queue_depth` 持续增长、`enqueue_total{status="timeout"}` 增长，或 `vehicle_async_sink_publish_duration_ms_histogram_bucket` 的 p99 持续上升，说明入口 publish 队列或下游 NATS/Kafka 写入已成为瓶颈。
 
 ### NATS Kafka bridge batch 指标
 

docs/ops/vehicle-ingest-runbook.md

@@ -120,6 +120,7 @@ go run ./cmd/load-sim \
 | `vehicle_async_sink_queue_depth{sink="nats"}` | 持续增长且不回落 | Gateway 到 NATS/Kafka 的异步 publish 队列开始积压。 |
 | `vehicle_async_sink_enqueue_total{status="timeout"}` | 任意增长 | Gateway publish 队列已满或 worker 长时间阻塞，入口可能开始丢实时性。 |
 | `vehicle_async_sink_publish_total{status="error"}` | 连续增长 | NATS/Kafka publish 失败，需要先查中间件连接和日志。 |
+| `vehicle_async_sink_publish_duration_ms_histogram_bucket` | p99 连续 5 分钟上升 | Gateway async worker 写 NATS/Kafka 变慢，通常会带动 queue depth 增长。 |
 | `vehicle_history_batch_flush_total{status="error"}` | 任意增长 | TDengine 批写失败；Kafka offset 不会提交，应先查 TDengine 和 SQL 错误。 |
 | `vehicle_history_batch_pending_messages` | 持续非 0 或 burst 后不回落 | history-writer 已拉取但未完成写入/提交，可能卡在 TDengine 或 Kafka commit。 |
 | `vehicle_history_batch_pending_rows` | 持续非 0 或 burst 后不回落 | TDengine 有批量写入积压，通常早于 Kafka lag 放大。 |

go/vehicle-gateway/internal/eventbus/async_sink.go

@@ -40,6 +40,8 @@ type asyncJob struct {
 
 var ErrAsyncSinkClosed = errors.New("async sink is closed")
 
+var asyncSinkPublishDurationBucketsMS = []float64{1, 5, 10, 25, 50, 100, 250, 500, 1000, 5000}
+
 func NewAsyncSink(delegate Sink, cfg AsyncConfig) *AsyncSink {
 	if delegate == nil {
 		panic("async delegate sink must not be nil")
@@ -169,7 +171,9 @@ func (s *AsyncSink) recordPublish(kind string, status string, elapsed time.Durat
 		"status": status,
 	}
 	s.metrics.IncCounter("vehicle_async_sink_publish_total", labels)
-	s.metrics.SetGauge("vehicle_async_sink_publish_duration_ms", labels, float64(elapsed.Milliseconds()))
+	elapsedMS := float64(elapsed.Milliseconds())
+	s.metrics.SetGauge("vehicle_async_sink_publish_duration_ms", labels, elapsedMS)
+	s.metrics.ObserveHistogram("vehicle_async_sink_publish_duration_ms_histogram", labels, asyncSinkPublishDurationBucketsMS, elapsedMS)
 }
 
 func (s *AsyncSink) recordQueueDepth() {

go/vehicle-gateway/internal/eventbus/async_sink_test.go

@@ -83,6 +83,9 @@ func TestAsyncSinkRecordsQueueMetrics(t *testing.T) {
 		`vehicle_async_sink_queue_depth{sink="nats"}`,
 		`vehicle_async_sink_publish_total{kind="raw",sink="nats",status="ok"} 1`,
 		`vehicle_async_sink_publish_total{kind="fields",sink="nats",status="ok"} 1`,
+		`vehicle_async_sink_publish_duration_ms_histogram_bucket{le="+Inf",kind="raw",sink="nats",status="ok"} 1`,
+		`vehicle_async_sink_publish_duration_ms_histogram_count{kind="raw",sink="nats",status="ok"} 1`,
+		`vehicle_async_sink_publish_duration_ms_histogram_sum{kind="raw",sink="nats",status="ok"}`,
 	} {
 		if !strings.Contains(text, want) {
 			t.Fatalf("async sink metric missing %s:\n%s", want, text)