feat: implement rate limiting retry mechanism for stat forwarder

Replace fixed delay retry with Kubernetes workqueue rate limiter

- Add TypedItemFastSlowRateLimiter for adaptive retry delays
- Implement fast/slow retry strategy (100ms for first 5, 500ms after)
- Add automatic state tracking and cleanup
- Update maybeRetry() to use rate limiter instead of fixed sleep
- Clean up rate limiter state on successful processing

Resolves TODO: Use RateLimitingInterface and NewItemFastSlowRateLimiter

Author: yansaitao

pkg/autoscaler/statforwarder/forwarder.go

@@ -23,6 +23,7 @@ import (
 	"time"
 
 	"go.uber.org/zap"
+	"k8s.io/client-go/util/workqueue"
 	"knative.dev/pkg/hash"
 	"knative.dev/pkg/logging"
 	"knative.dev/pkg/logging/logkey"
@@ -38,8 +39,18 @@ const (
 	retryInterval      = 100 * time.Millisecond
 
 	// Retry at most 15 seconds to process a stat.
-	maxProcessingRetry      = 30
+	maxProcessingRetry = 30
+	// retryProcessingInterval is kept for backward compatibility but no longer used.
+	// Use rateLimiter instead.
 	retryProcessingInterval = 500 * time.Millisecond
+
+	// Rate limiting configuration for retries.
+	// fastRetryDelay is used for the first maxFastRetryAttempts retries.
+	fastRetryDelay = 100 * time.Millisecond
+	// slowRetryDelay is used after maxFastRetryAttempts retries.
+	slowRetryDelay = 500 * time.Millisecond
+	// maxFastRetryAttempts is the number of fast retries before switching to slow retries.
+	maxFastRetryAttempts = 5
 )
 
 var svcURLSuffix = fmt.Sprintf("svc.%s:%d", network.GetClusterDomainName(), autoscalerPort)
@@ -75,6 +86,9 @@ type Forwarder struct {
 
 	statCh chan stat
 	stopCh chan struct{}
+
+	// rateLimiter controls the retry backoff strategy using fast/slow retry delays.
+	rateLimiter workqueue.TypedRateLimiter[string]
 }
 
 // New creates a new Forwarder.
@@ -84,11 +98,12 @@ type Forwarder struct {
 func New(ctx context.Context, bs *hash.BucketSet) *Forwarder {
 	bkts := bs.Buckets()
 	f := &Forwarder{
-		logger:     logging.FromContext(ctx),
-		bs:         bs,
-		processors: make(map[string]bucketProcessor, len(bkts)),
-		statCh:     make(chan stat, 1000),
-		stopCh:     make(chan struct{}),
+		logger:      logging.FromContext(ctx),
+		bs:          bs,
+		processors:  make(map[string]bucketProcessor, len(bkts)),
+		statCh:      make(chan stat, 1000),
+		stopCh:      make(chan struct{}),
+		rateLimiter: workqueue.NewTypedItemFastSlowRateLimiter[string](fastRetryDelay, slowRetryDelay, maxFastRetryAttempts),
 	}
 
 	f.processingWg.Add(1)
@@ -142,23 +157,37 @@ func (f *Forwarder) process() {
 			if err := p.process(s.sm); err != nil {
 				l.Errorw("Error while processing stat", zap.Error(err))
 				f.maybeRetry(l, s)
+			} else {
+				// Successfully processed, forget the retry state for this stat.
+				f.rateLimiter.Forget(rev)
 			}
 		}
 	}
 }
 
 func (f *Forwarder) maybeRetry(logger *zap.SugaredLogger, s stat) {
-	if s.retry > maxProcessingRetry {
-		logger.Warn("Exceeding max retries. Dropping the stat.")
+	rev := s.sm.Key.String()
+
+	// Check current retry count before calling When (which increments the count).
+	numRequeues := f.rateLimiter.NumRequeues(rev)
+	if numRequeues >= maxProcessingRetry {
+		logger.Warnf("Exceeding max retries (%d). Dropping the stat.", maxProcessingRetry)
+		// Clean up the rate limiter state for this stat.
+		f.rateLimiter.Forget(rev)
+		return
 	}
 
-	s.retry++
+	// Get the retry delay from the rate limiter (fast for initial retries, slow after).
+	// This will increment the internal retry count.
+	retryDelay := f.rateLimiter.When(rev)
+
 	f.retryWg.Add(1)
 	go func() {
 		defer f.retryWg.Done()
-		// TODO(yanweiguo): Use RateLimitingInterface and NewItemFastSlowRateLimiter.
-		time.Sleep(retryProcessingInterval)
-		logger.Debug("Enqueuing stat for retry.")
+		time.Sleep(retryDelay)
+		logger.Debugf("Enqueuing stat for retry (attempt %d, delay %v).", numRequeues+1, retryDelay)
+		// Increment retry count for tracking (though rate limiter also tracks this).
+		s.retry++
 		f.statCh <- s
 	}()
 }