fix(chat): 优化后台限流与 SSE 日志

backend/internal/modules/chathub/handler.go

@@ -1,6 +1,7 @@
 package chathub
 
 import (
+	"context"
 	"fmt"
 	"time"
 
@@ -8,16 +9,21 @@ import (
 	"hfb_sys/backend/pkg/response"
 
 	"github.com/gin-gonic/gin"
+	"go.uber.org/zap"
 )
 
 const heartbeatInterval = 30 * time.Second
 
 type Handler struct {
-	hub *Hub
+	hub    *Hub
+	logger *zap.Logger
 }
 
-func NewHandler(hub *Hub) *Handler {
-	return &Handler{hub: hub}
+func NewHandler(hub *Hub, logger *zap.Logger) *Handler {
+	if logger == nil {
+		logger = zap.NewNop()
+	}
+	return &Handler{hub: hub, logger: logger}
 }
 
 // UserEvents 处理用户端 SSE 连接: GET /api/chats/events
@@ -51,8 +57,23 @@ func (h *Handler) AdminEvents(c *gin.Context) {
 }
 
 func (h *Handler) serveSSE(c *gin.Context, pType string, pID uint64) {
+	startedAt := time.Now()
+	disconnectReason := "handler_completed"
+	var disconnectErr error
 	ch := h.hub.Subscribe(pType, pID)
-	defer h.hub.Unsubscribe(pType, pID, ch)
+	h.logger.Info("SSE 连接建立", h.connectionLogFields(c, pType, pID)...)
+	defer func() {
+		h.hub.Unsubscribe(pType, pID, ch)
+		fields := h.connectionLogFields(c, pType, pID)
+		fields = append(fields,
+			zap.String("disconnect_reason", disconnectReason),
+			zap.Float64("lifetime_ms", float64(time.Since(startedAt).Microseconds())/1000),
+		)
+		if disconnectErr != nil {
+			fields = append(fields, zap.Error(disconnectErr))
+		}
+		h.logger.Info("SSE 连接断开", fields...)
+	}()
 
 	c.Header("Content-Type", "text/event-stream")
 	c.Header("Cache-Control", "no-cache")
@@ -60,7 +81,11 @@ func (h *Handler) serveSSE(c *gin.Context, pType string, pID uint64) {
 	c.Header("X-Accel-Buffering", "no")
 
 	// 发送初始连接确认
-	fmt.Fprintf(c.Writer, "event: connected\ndata: {\"ok\":true}\n\n")
+	if _, err := fmt.Fprintf(c.Writer, "event: connected\ndata: {\"ok\":true}\n\n"); err != nil {
+		disconnectReason = "initial_write_error"
+		disconnectErr = err
+		return
+	}
 	c.Writer.Flush()
 
 	heartbeat := time.NewTicker(heartbeatInterval)
@@ -71,17 +96,55 @@ func (h *Handler) serveSSE(c *gin.Context, pType string, pID uint64) {
 	for {
 		select {
 		case <-clientGone:
+			disconnectReason = contextDisconnectReason(c.Request.Context().Err())
 			return
 		case <-heartbeat.C:
-			fmt.Fprintf(c.Writer, ":heartbeat\n\n")
+			if _, err := fmt.Fprintf(c.Writer, ":heartbeat\n\n"); err != nil {
+				disconnectReason = "heartbeat_write_error"
+				disconnectErr = err
+				return
+			}
 			c.Writer.Flush()
 		case event, ok := <-ch:
 			if !ok {
+				disconnectReason = "subscription_closed"
 				return
 			}
 			data := MarshalEvent(event)
-			fmt.Fprintf(c.Writer, "event: %s\ndata: %s\n\n", event.Type, data)
+			if _, err := fmt.Fprintf(c.Writer, "event: %s\ndata: %s\n\n", event.Type, data); err != nil {
+				disconnectReason = "event_write_error"
+				disconnectErr = err
+				return
+			}
 			c.Writer.Flush()
 		}
 	}
 }
+
+func (h *Handler) connectionLogFields(c *gin.Context, pType string, pID uint64) []zap.Field {
+	fields := []zap.Field{
+		zap.String("request_id", middleware.GetRequestID(c)),
+		zap.String("principal_type", pType),
+		zap.Uint64("principal_id", pID),
+		zap.String("client_ip", c.ClientIP()),
+		zap.Int("active_sse_connections", h.hub.OnlineCount()),
+		zap.Int("active_principal_type_connections", h.hub.OnlineCountByType(pType)),
+	}
+	if pType == "admin" {
+		fields = append(fields, zap.Uint64("admin_id", pID))
+	} else {
+		fields = append(fields, zap.Uint64("user_id", pID))
+	}
+	return fields
+}
+
+func contextDisconnectReason(err error) string {
+	switch err {
+	case context.Canceled:
+		return "context_canceled"
+	case context.DeadlineExceeded:
+		return "context_deadline_exceeded"
+	default:
+		return "context_closed"
+	}
+}

backend/internal/modules/chathub/handler_test.go

@@ -0,0 +1,60 @@
+package chathub
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"hfb_sys/backend/internal/middleware"
+
+	"github.com/gin-gonic/gin"
+	"go.uber.org/zap"
+	"go.uber.org/zap/zaptest/observer"
+)
+
+func TestSSELifecycleLogsConnectionCountReasonAndLifetime(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	core, observed := observer.New(zap.InfoLevel)
+	hub := NewHub(nil)
+	handler := NewHandler(hub, zap.New(core))
+
+	requestContext, cancel := context.WithCancel(context.Background())
+	cancel()
+	request := httptest.NewRequest(http.MethodGet, "/api/admin/chats/events", nil).WithContext(requestContext)
+	request.RemoteAddr = "42.49.131.69:34567"
+	recorder := httptest.NewRecorder()
+	ginContext, _ := gin.CreateTestContext(recorder)
+	ginContext.Request = request
+	ginContext.Set(middleware.ContextAdminID, uint64(88))
+	ginContext.Set(middleware.ContextRequestID, "sse-test-request")
+
+	handler.AdminEvents(ginContext)
+
+	connected := observed.FilterMessage("SSE 连接建立").All()
+	if len(connected) != 1 {
+		t.Fatalf("connection logs = %d, want 1", len(connected))
+	}
+	connectedFields := connected[0].ContextMap()
+	if connectedFields["admin_id"] != uint64(88) || connectedFields["active_sse_connections"] != int64(1) {
+		t.Fatalf("unexpected connection fields: %v", connectedFields)
+	}
+
+	disconnected := observed.FilterMessage("SSE 连接断开").All()
+	if len(disconnected) != 1 {
+		t.Fatalf("disconnection logs = %d, want 1", len(disconnected))
+	}
+	disconnectedFields := disconnected[0].ContextMap()
+	if disconnectedFields["disconnect_reason"] != "context_canceled" {
+		t.Fatalf("disconnect reason = %v, want context_canceled", disconnectedFields["disconnect_reason"])
+	}
+	if disconnectedFields["active_sse_connections"] != int64(0) {
+		t.Fatalf("active connections after disconnect = %v, want 0", disconnectedFields["active_sse_connections"])
+	}
+	if _, ok := disconnectedFields["lifetime_ms"]; !ok {
+		t.Fatalf("missing lifetime_ms field: %v", disconnectedFields)
+	}
+	if got := hub.OnlineCount(); got != 0 {
+		t.Fatalf("online connection count = %d, want 0", got)
+	}
+}

backend/internal/modules/chathub/hub.go

@@ -190,3 +190,16 @@ func (h *Hub) OnlineCount() int {
 	}
 	return count
 }
+
+// OnlineCountByType 返回当前进程内指定主体类型的 SSE 连接数。
+func (h *Hub) OnlineCountByType(pType string) int {
+	h.mu.RLock()
+	defer h.mu.RUnlock()
+	count := 0
+	for key, clients := range h.clients {
+		if key.Type == pType {
+			count += len(clients)
+		}
+	}
+	return count
+}

backend/internal/modules/chathub/hub_test.go

@@ -7,6 +7,7 @@ func TestDisconnectUserClosesAllConnections(t *testing.T) {
 	first := hub.Subscribe("user", 7)
 	second := hub.Subscribe("user", 7)
 	hub.Subscribe("user", 8)
+	hub.Subscribe("admin", 9)
 
 	hub.DisconnectUser(7)
 	for name, ch := range map[string]<-chan *ChatEvent{"first": first, "second": second} {
@@ -19,7 +20,13 @@ func TestDisconnectUserClosesAllConnections(t *testing.T) {
 			t.Fatalf("%s connection did not close immediately", name)
 		}
 	}
-	if got := hub.OnlineCount(); got != 1 {
-		t.Fatalf("online connection count = %d, want 1", got)
+	if got := hub.OnlineCount(); got != 2 {
+		t.Fatalf("online connection count = %d, want 2", got)
+	}
+	if got := hub.OnlineCountByType("user"); got != 1 {
+		t.Fatalf("online user connection count = %d, want 1", got)
+	}
+	if got := hub.OnlineCountByType("admin"); got != 1 {
+		t.Fatalf("online admin connection count = %d, want 1", got)
 	}
 }