[exporterhelper] Fix potential deadlock in the batch sender

Concurrent handling of the flush timeouts can run into a deadlock when a batch is simultaneously sent by reaching the minimum size and flush timeout. The deadlock can happen on the following lines:
- https://github.com/open-telemetry/opentelemetry-collector/blob/115bc8e28e009ca93565dc4deb4cf6608fa63622/exporter/exporterhelper/batch_sender.go#L131
- https://github.com/open-telemetry/opentelemetry-collector/blob/115bc8e28e009ca93565dc4deb4cf6608fa63622/exporter/exporterhelper/batch_sender.go#L87

.chloggen/batchseder-fix-potential-deadlock.yaml

@@ -0,0 +1,20 @@
+# Use this changelog template to create an entry for release notes.
+
+# One of 'breaking', 'deprecation', 'new_component', 'enhancement', 'bug_fix'
+change_type: bug_fix
+
+# The name of the component, or a single word describing the area of concern, (e.g. otlpreceiver)
+component: exporterhelper
+
+# A brief description of the change.  Surround your text with quotes ("") if it needs to start with a backtick (`).
+note: Fix potential deadlock in the batch sender
+
+# One or more tracking issues or pull requests related to the change
+issues: [10315]
+
+# Optional: The change log or logs in which this entry should be included.
+# e.g. '[user]' or '[user, api]'
+# Include 'user' if the change is relevant to end users.
+# Include 'api' if there is a change to a library API.
+# Default: '[user]'
+change_logs: [user]

exporter/exporterhelper/batch_sender.go

@@ -33,10 +33,9 @@ type batchSender struct {
 	concurrencyLimit uint64
 	activeRequests   atomic.Uint64
 
-	resetTimerCh chan struct{}
-
 	mu          sync.Mutex
 	activeBatch *batch
+	lastFlushed time.Time
 
 	logger *zap.Logger
 
@@ -57,7 +56,6 @@ func newBatchSender(cfg exporterbatcher.Config, set exporter.CreateSettings,
 		shutdownCh:         nil,
 		shutdownCompleteCh: make(chan struct{}),
 		stopped:            &atomic.Bool{},
-		resetTimerCh:       make(chan struct{}),
 	}
 	return bs
 }
@@ -85,16 +83,17 @@ func (bs *batchSender) Start(_ context.Context, _ component.Host) error {
 				return
 			case <-timer.C:
 				bs.mu.Lock()
+				nextFlush := bs.cfg.FlushTimeout
 				if bs.activeBatch.request != nil {
-					bs.exportActiveBatch()
+					sinceLastFlush := time.Since(bs.lastFlushed)
+					if sinceLastFlush >= bs.cfg.FlushTimeout {
+						bs.exportActiveBatch()
+					} else {
+						nextFlush = bs.cfg.FlushTimeout - sinceLastFlush
+					}
 				}
 				bs.mu.Unlock()
-				timer.Reset(bs.cfg.FlushTimeout)
-			case <-bs.resetTimerCh:
-				if !timer.Stop() {
-					<-timer.C
-				}
-				timer.Reset(bs.cfg.FlushTimeout)
+				timer.Reset(nextFlush)
 			}
 		}
 	}()
@@ -123,15 +122,10 @@ func (bs *batchSender) exportActiveBatch() {
 		b.err = bs.nextSender.send(b.ctx, b.request)
 		close(b.done)
 	}(bs.activeBatch)
+	bs.lastFlushed = time.Now()
 	bs.activeBatch = newEmptyBatch()
 }
 
-func (bs *batchSender) resetTimer() {
-	if !bs.stopped.Load() {
-		bs.resetTimerCh <- struct{}{}
-	}
-}
-
 // isActiveBatchReady returns true if the active batch is ready to be exported.
 // The batch is ready if it has reached the minimum size or the concurrency limit is reached.
 // Caller must hold the lock.
@@ -168,7 +162,6 @@ func (bs *batchSender) sendMergeSplitBatch(ctx context.Context, req Request) err
 		batch := bs.activeBatch
 		if bs.isActiveBatchReady() || len(reqs) > 1 {
 			bs.exportActiveBatch()
-			bs.resetTimer()
 		}
 		bs.mu.Unlock()
 		<-batch.done
@@ -208,7 +201,6 @@ func (bs *batchSender) sendMergeBatch(ctx context.Context, req Request) error {
 	batch := bs.activeBatch
 	if bs.isActiveBatchReady() {
 		bs.exportActiveBatch()
-		bs.resetTimer()
 	}
 	bs.mu.Unlock()
 	<-batch.done

exporter/exporterhelper/batch_sender_test.go

@@ -535,6 +535,35 @@ func TestBatchSenderWithTimeout(t *testing.T) {
 	assert.EqualValues(t, 12, sink.itemsCount.Load())
 }
 
+func TestBatchSenderTimerResetNoConflict(t *testing.T) {
+	bCfg := exporterbatcher.NewDefaultConfig()
+	bCfg.MinSizeItems = 8
+	bCfg.FlushTimeout = 50 * time.Millisecond
+	be, err := newBaseExporter(defaultSettings, defaultDataType, newNoopObsrepSender,
+		WithBatcher(bCfg, WithRequestBatchFuncs(fakeBatchMergeFunc, fakeBatchMergeSplitFunc)))
+	require.NoError(t, err)
+	require.NoError(t, be.Start(context.Background(), componenttest.NewNopHost()))
+
+	sink := newFakeRequestSink()
+
+	// Send 2 concurrent requests that should be merged in one batch in the same interval as the flush timer
+	go func() {
+		require.NoError(t, be.send(context.Background(), &fakeRequest{items: 4, sink: sink}))
+	}()
+	time.Sleep(50 * time.Millisecond)
+	go func() {
+		require.NoError(t, be.send(context.Background(), &fakeRequest{items: 4, sink: sink}))
+	}()
+
+	// The batch should be sent either with the flush interval or by reaching the minimum items size with no conflict
+	assert.EventuallyWithT(t, func(c *assert.CollectT) {
+		assert.LessOrEqual(c, uint64(1), sink.requestsCount.Load())
+		assert.EqualValues(c, 8, sink.itemsCount.Load())
+	}, 200*time.Millisecond, 10*time.Millisecond)
+
+	require.NoError(t, be.Shutdown(context.Background()))
+}
+
 func queueBatchExporter(t *testing.T, batchOption Option) *baseExporter {
 	be, err := newBaseExporter(defaultSettings, defaultDataType, newNoopObsrepSender, batchOption,
 		WithRequestQueue(exporterqueue.NewDefaultConfig(), exporterqueue.NewMemoryQueueFactory[Request]()))