diff --git a/pipeline/event.go b/pipeline/event.go
index 661b95cc..045e9f2c 100644
--- a/pipeline/event.go
+++ b/pipeline/event.go
@@ -391,11 +391,13 @@ again:
 	})
 
 	// Wait until we fit in the capacity.
+	p.slowWaiters.Inc()
 	p.getCond.L.Lock()
-	if int(p.inUseEvents.Load()) > p.capacity {
+	if p.eventsAvailable() {
 		p.getCond.Wait()
 	}
 	p.getCond.L.Unlock()
+	p.slowWaiters.Dec()
 	goto again
 }
 
@@ -430,10 +432,14 @@ func (p *eventSyncPool) wakeupWaiters() {
 
 		time.Sleep(5 * time.Second)
 		waiters := p.slowWaiters.Load()
-		eventsAvailable := p.inUseEvents.Load() <= int64(p.capacity)
+		eventsAvailable := p.eventsAvailable()
 		if waiters > 0 && eventsAvailable {
 			// There are events in the pool, wake up waiting goroutines.
 			p.getCond.Broadcast()
 		}
 	}
 }
+
+func (p *eventSyncPool) eventsAvailable() bool {
+	return int(p.inUseEvents.Load()) >= p.capacity
+}
diff --git a/pipeline/event_test.go b/pipeline/event_test.go
index 01c85c16..4eddb7f1 100644
--- a/pipeline/event_test.go
+++ b/pipeline/event_test.go
@@ -4,6 +4,7 @@ import (
 	"runtime"
 	"sync"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/require"
 )
@@ -19,69 +20,154 @@ func TestEventPoolDump(t *testing.T) {
 }
 
 func BenchmarkEventPoolOneGoroutine(b *testing.B) {
-	const capacity = 32
-
-	p := newEventPool(capacity, DefaultAvgInputEventSize)
-
-	for i := 0; i < b.N; i++ {
-		p.back(p.get())
+	bench := func(b *testing.B, p pool) {
+		for i := 0; i < b.N; i++ {
+			p.back(p.get())
+		}
 	}
+	const capacity = 32
+	b.Run("eventPool", func(b *testing.B) {
+		p := newEventPool(capacity, DefaultAvgInputEventSize)
+		bench(b, p)
+	})
+	b.Run("syncPool", func(b *testing.B) {
+		p := newSyncPool(capacity)
+		bench(b, p)
+	})
 }
 
 func BenchmarkEventPoolManyGoroutines(b *testing.B) {
+	bench := func(b *testing.B, p pool) {
+		workers := runtime.GOMAXPROCS(0)
+		for i := 0; i < b.N; i++ {
+			wg := &sync.WaitGroup{}
+			wg.Add(workers)
+			for j := 0; j < workers; j++ {
+				go func() {
+					defer wg.Done()
+
+					for k := 0; k < 1000; k++ {
+						p.back(p.get())
+					}
+				}()
+			}
+			wg.Wait()
+		}
+	}
 	const capacity = 32
+	b.Run("eventPool", func(b *testing.B) {
+		p := newEventPool(capacity, DefaultAvgInputEventSize)
+		bench(b, p)
+	})
+	b.Run("syncPool", func(b *testing.B) {
+		p := newSyncPool(capacity)
+		bench(b, p)
+	})
+}
 
-	p := newEventPool(capacity, DefaultAvgInputEventSize)
-
-	for i := 0; i < b.N; i++ {
+func BenchmarkEventPoolSlowestPath(b *testing.B) {
+	bench := func(b *testing.B, p pool) {
 		wg := &sync.WaitGroup{}
-		for j := 0; j < runtime.GOMAXPROCS(0); j++ {
-			wg.Add(1)
-			go func() {
-				for k := 0; k < 1000; k++ {
-					p.back(p.get())
-				}
-				wg.Done()
-			}()
+		for i := 0; i < b.N; i++ {
+			const concurrency = 1_000
+			wg.Add(concurrency)
+			for j := 0; j < concurrency; j++ {
+				go func() {
+					defer wg.Done()
+					e := p.get()
+					p.back(e)
+				}()
+			}
+			wg.Wait()
 		}
-		wg.Wait()
 	}
-}
 
-func BenchmarkEventPoolSlowestPath(b *testing.B) {
 	const capacity = 32
+	b.Run("eventPool", func(b *testing.B) {
+		p := newEventPool(capacity, DefaultAvgInputEventSize)
+		bench(b, p)
+	})
+	b.Run("syncPool", func(b *testing.B) {
+		p := newSyncPool(capacity)
+		bench(b, p)
+	})
+}
 
-	p := newEventPool(capacity, DefaultAvgInputEventSize)
-
-	for i := 0; i < b.N; i++ {
-		wg := &sync.WaitGroup{}
-		for j := 0; j < 100_000; j++ {
-			wg.Add(1)
-			go func() {
-				e := p.get()
-				p.back(e)
-				wg.Done()
-			}()
+func TestLowMemPool(t *testing.T) {
+	t.Parallel()
+	r := require.New(t)
+	test := func(capacity, batchSize int) {
+		p := newSyncPool(capacity)
+		for i := 0; i < batchSize; i++ {
+			batch := make([]*Event, batchSize)
+			for j := 0; j < batchSize; j++ {
+				batch[j] = p.get()
+			}
+			r.Equal(int64(batchSize), p.inUse())
+			r.Equal(int64(0), p.waiters())
+
+			for j := 0; j < batchSize; j++ {
+				p.back(batch[j])
+			}
 		}
-		wg.Wait()
 	}
+
+	test(1, 1)
+	test(8, 7)
+	test(64, 63)
+	test(64, 64)
+	test(1024, 128)
 }
 
-func BenchmarkEventSyncPoolSlowestPath(b *testing.B) {
-	const capacity = 32
+func TestLowMemPoolSlowWait(t *testing.T) {
+	t.Parallel()
+	r := require.New(t)
 
-	p := newEventPool(capacity, DefaultAvgInputEventSize)
+	test := func(p pool) {
+		event := p.get() // Empty the pool.
+		r.Equal(int64(1), p.inUse())
 
-	for i := 0; i < b.N; i++ {
-		wg := &sync.WaitGroup{}
-		for j := 0; j < 100_000; j++ {
+		eventReleased := false
+
+		const waiters = 16
+		// Create 16 goroutines to wait on new events.
+		wg := new(sync.WaitGroup)
+		for i := 0; i < waiters; i++ {
 			wg.Add(1)
 			go func() {
+				defer wg.Done()
 				e := p.get()
+				if !eventReleased {
+					r.FailNowf("pool have to be empty", "event: %v", e)
+				}
 				p.back(e)
-				wg.Done()
 			}()
 		}
+
+		// Wait for all goroutines to be waiting.
+		for i := 0; i < 50; i++ {
+			if p.waiters() == waiters {
+				break
+			}
+			time.Sleep(time.Millisecond * 100)
+		}
+
+		r.Equal(p.waiters(), int64(waiters))
+		r.Equal(int64(1), p.inUse())
+
+		// Release events to wakeup waiters.
+		eventReleased = true
+		p.back(event)
+
 		wg.Wait()
+
+		r.Equal(int64(0), p.waiters())
+		r.Equal(int64(0), p.inUse())
 	}
+
+	t.Run("syncPool", func(t *testing.T) {
+		pool := newSyncPool(1)
+		test(pool)
+	})
+	// TODO: add test for eventPool after #685.
 }
diff --git a/pipeline/pipeline_test.go b/pipeline/pipeline_test.go
index 675f99a1..7bafa6b3 100644
--- a/pipeline/pipeline_test.go
+++ b/pipeline/pipeline_test.go
@@ -1,7 +1,6 @@
 package pipeline_test
 
 import (
-	"reflect"
 	"testing"
 
 	"github.com/ozontech/file.d/pipeline"
@@ -25,61 +24,6 @@ func getFakeInputInfo() *pipeline.InputPluginInfo {
 	}
 }
 
-func TestInUnparsableMessages(t *testing.T) {
-	name := "invalid_json"
-	message := []byte("{wHo Is Json: YoU MeAn SoN oF JoHn???")
-	pipelineSettings := &pipeline.Settings{
-		Capacity:           5,
-		Decoder:            "json",
-		MetricHoldDuration: pipeline.DefaultMetricHoldDuration,
-	}
-	offset := int64(666)
-	sourceID := pipeline.SourceID(3<<16 + int(10))
-
-	t.Run(name, func(t *testing.T) {
-		pipe := pipeline.New("test_pipeline", pipelineSettings, prometheus.NewRegistry())
-
-		pipe.SetInput(getFakeInputInfo())
-
-		seqID := pipe.In(sourceID, "kafka", offset, message, false, nil)
-		require.Equal(t, pipeline.EventSeqIDError, seqID)
-
-		refPipe := reflect.ValueOf(pipe)
-		eventPool := reflect.Indirect(refPipe).FieldByName("eventPool")
-
-		free1slice := reflect.
-			Indirect(
-				reflect.
-					Indirect(eventPool).
-					FieldByName("free1")).
-			Slice(0, pipelineSettings.Capacity)
-		free2slice := reflect.
-			Indirect(
-				reflect.
-					Indirect(eventPool).
-					FieldByName("free2")).
-			Slice(0, pipelineSettings.Capacity)
-
-		for i := 0; i < pipelineSettings.Capacity; i++ {
-			// free1, free2 are []atomic.Bool which underlying v is really uint32
-			// so if v val == uint32(1) event was released.
-			free1idxUint := reflect.
-				Indirect(free1slice.Index(i)).
-				FieldByName("v").
-				FieldByName("v").
-				Uint()
-			require.EqualValues(t, uint32(1), free1idxUint)
-
-			free2idxUint := reflect.
-				Indirect(free2slice.Index(i)).
-				FieldByName("v").
-				FieldByName("v").
-				Uint()
-			require.EqualValues(t, uint32(1), free2idxUint)
-		}
-	})
-}
-
 func TestInInvalidMessages(t *testing.T) {
 	cases := []struct {
 		name             string