feat: 完善服务监控与下游重投

gateway/cmd/gateway/main.go

@@ -5,13 +5,19 @@ import (
 	"log"
 	"net/http"
 	"os"
+	"strconv"
+	"strings"
+	"time"
 
 	"cmpp-platform/gateway/internal/control"
 	"cmpp-platform/gateway/internal/health"
 	"cmpp-platform/gateway/internal/inbound"
+	platformmetrics "cmpp-platform/gateway/internal/metrics"
 	"cmpp-platform/gateway/internal/ratelimit"
 	"cmpp-platform/gateway/internal/submitworker"
 	"cmpp-platform/gateway/internal/upstream"
+
+	"github.com/redis/go-redis/v9"
 )
 
 func main() {
@@ -25,6 +31,7 @@ func main() {
 	}
 	apiBaseURL := os.Getenv("API_BASE_URL")
 	upstreamManager := &upstream.Manager{APIBaseURL: apiBaseURL}
+	var worker *submitworker.Worker
 	channelLimiter, err := ratelimit.New(os.Getenv("REDIS_URL"))
 	if err != nil {
 		log.Fatalf("gateway channel rate limiter init failed: %v", err)
@@ -53,7 +60,7 @@ func main() {
 	}()
 
 	if os.Getenv("GATEWAY_SUBMIT_WORKER_DISABLED") != "true" {
-		worker, err := submitworker.New(os.Getenv("REDIS_URL"), upstreamManager)
+		worker, err = submitworker.New(os.Getenv("REDIS_URL"), upstreamManager)
 		if err != nil {
 			log.Printf("gateway submit worker init failed: %v", err)
 		} else {
@@ -72,6 +79,39 @@ func main() {
 
 	mux := http.NewServeMux()
 	mux.Handle("/health", health.Handler())
+	mux.Handle("/metrics", platformmetrics.Handler(func(ctx context.Context) platformmetrics.Snapshot {
+		desired, connected := upstreamManager.ConnectionCounts()
+		snapshot := platformmetrics.Snapshot{
+			UpstreamDesired: desired, UpstreamConnected: connected,
+			DownstreamConnected: inbound.ActiveConnectionCount(), SubmitWorkerUp: worker != nil,
+		}
+		if worker == nil || worker.Redis == nil {
+			return snapshot
+		}
+		pending, err := worker.Redis.XPending(ctx, worker.Stream, worker.Group).Result()
+		if err != nil {
+			return snapshot
+		}
+		snapshot.QueueAvailable = true
+		snapshot.QueuePending = pending.Count
+		groups, err := worker.Redis.XInfoGroups(ctx, worker.Stream).Result()
+		if err == nil {
+			for _, group := range groups {
+				if group.Name == worker.Group {
+					snapshot.QueueLag = group.Lag
+					break
+				}
+			}
+		}
+		entries, err := worker.Redis.XPendingExt(ctx, &redis.XPendingExtArgs{Stream: worker.Stream, Group: worker.Group, Start: "-", End: "+", Count: 1}).Result()
+		if err == nil && len(entries) > 0 {
+			milliseconds, parseErr := strconv.ParseInt(strings.SplitN(entries[0].ID, "-", 2)[0], 10, 64)
+			if parseErr == nil {
+				snapshot.QueueOldestAgeSeconds = max(0, time.Since(time.UnixMilli(milliseconds)).Seconds())
+			}
+		}
+		return snapshot
+	}))
 	control.Register(mux, control.Server{
 		APIBaseURL: apiBaseURL,
 		Upstream:   upstreamManager,