fix(libbeat): mitigate race condition in ratelimit processor

CHANGELOG.next.asciidoc

@@ -21,6 +21,7 @@ https://github.com/elastic/beats/compare/v8.8.1\...main[Check the HEAD diff]
 - Fix templates and docs to use correct `--` version of command line arguments. {issue}42038[42038] {pull}42060[42060]
 - removed support for a single `-` to precede multi-letter command line arguments.  Use `--` instead. {issue}42117[42117] {pull}42209[42209]
 - Removed encryption from diskqueue V2 for fips compliance {issue}4534[4534]{pull}42848[42848]
+- Fix race conditions in the global ratelimit processor that could drop events or apply rate limiting incorrectly.
 
 *Auditbeat*
 

libbeat/processors/ratelimit/rate_limit_test.go

@@ -176,3 +176,16 @@ func TestRateLimit(t *testing.T) {
 		})
 	}
 }
+
+func BenchmarkRateLimit(b *testing.B) {
+	p, err := new(conf.MustNewConfigFrom(mapstr.M{
+		"limit": "100/s",
+	}))
+	require.NoError(b, err)
+	event := beat.Event{Fields: mapstr.M{"field": 1}}
+
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		p.Run(&event) //nolint:errcheck // ignore
+	}
+}

libbeat/processors/ratelimit/token_bucket.go

@@ -35,10 +35,36 @@ func init() {
 }
 
 type bucket struct {
+	mu sync.Mutex
+
 	tokens        float64
 	lastReplenish time.Time
 }
 
+func (b *bucket) withdraw() bool {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+
+	if b.tokens < 1 {
+		return false
+	}
+
+	b.tokens--
+	return true
+}
+
+func (b *bucket) replenish(rate rate, clock clockwork.Clock) float64 {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+
+	secsSinceLastReplenish := clock.Now().Sub(b.lastReplenish).Seconds()
+	tokensToReplenish := secsSinceLastReplenish * rate.valuePerSecond()
+
+	b.tokens += tokensToReplenish
+	b.lastReplenish = clock.Now()
+	return b.tokens
+}
+
 type tokenBucket struct {
 	mu unison.Mutex
 
@@ -122,35 +148,20 @@ func (t *tokenBucket) setClock(c clockwork.Clock) {
 }
 
 func (t *tokenBucket) getBucket(key uint64) *bucket {
-	v, exists := t.buckets.LoadOrStore(key, &bucket{
-		tokens:        t.depth,
-		lastReplenish: t.clock.Now(),
-	})
-	//nolint:errcheck // ignore
-	b := v.(*bucket)
-
+	v, exists := t.buckets.Load(key)
 	if exists {
+		//nolint:errcheck // ignore
+		b := v.(*bucket)
 		b.replenish(t.limit, t.clock)
 		return b
 	}
 
-	return b
-}
-
-func (b *bucket) withdraw() bool {
-	if b.tokens < 1 {
-		return false
-	}
-	b.tokens--
-	return true
-}
-
-func (b *bucket) replenish(rate rate, clock clockwork.Clock) {
-	secsSinceLastReplenish := clock.Now().Sub(b.lastReplenish).Seconds()
-	tokensToReplenish := secsSinceLastReplenish * rate.valuePerSecond()
-
-	b.tokens += tokensToReplenish
-	b.lastReplenish = clock.Now()
+	v, _ = t.buckets.LoadOrStore(key, &bucket{
+		tokens:        t.depth,
+		lastReplenish: t.clock.Now(),
+	})
+	//nolint:errcheck // ignore
+	return v.(*bucket)
 }
 
 func (t *tokenBucket) runGC() {
@@ -177,9 +188,8 @@ func (t *tokenBucket) runGC() {
 			//nolint:errcheck // ignore
 			b := v.(*bucket)
 
-			b.replenish(t.limit, t.clock)
-
-			if b.tokens >= t.depth {
+			tokens := b.replenish(t.limit, t.clock)
+			if tokens >= t.depth {
 				toDelete = append(toDelete, key)
 			}
 
@@ -193,7 +203,7 @@ func (t *tokenBucket) runGC() {
 		}
 
 		// Reset GC metrics
-		t.gc.metrics.numCalls = atomic.Uint64{}
+		t.gc.metrics.numCalls.Store(0)
 
 		gcDuration := time.Since(gcStartTime)
 		numBucketsDeleted := len(toDelete)