diff --git a/connection_impl.go b/connection_impl.go
index 1fa1a8e4..eb99f1c1 100644
--- a/connection_impl.go
+++ b/connection_impl.go
@@ -43,7 +43,6 @@ type connection struct {
 	writeTrigger    chan error
 	inputBuffer     *LinkBuffer
 	outputBuffer    *LinkBuffer
-	inputBarrier    *barrier
 	outputBarrier   *barrier
 	supportZeroCopy bool
 	maxSize         int // The maximum size of data between two Release().
@@ -323,7 +322,7 @@ func (c *connection) init(conn Conn, opts *options) (err error) {
 	c.writeTrigger = make(chan error, 1)
 	c.bookSize, c.maxSize = pagesize, pagesize
 	c.inputBuffer, c.outputBuffer = NewLinkBuffer(pagesize), NewLinkBuffer()
-	c.inputBarrier, c.outputBarrier = barrierPool.Get().(*barrier), barrierPool.Get().(*barrier)
+	c.outputBarrier = barrierPool.Get().(*barrier)
 
 	c.initNetFD(conn) // conn must be *netFD{}
 	c.initFDOperator()
diff --git a/connection_reactor.go b/connection_reactor.go
index fa485be1..cd5d717c 100644
--- a/connection_reactor.go
+++ b/connection_reactor.go
@@ -71,7 +71,6 @@ func (c *connection) closeBuffer() {
 	// so we need to check the buffer length, and if it's an "unclean" close operation, let's give up to reuse the buffer
 	if c.inputBuffer.Len() == 0 || onConnect != nil || onRequest != nil {
 		c.inputBuffer.Close()
-		barrierPool.Put(c.inputBarrier)
 	}
 	if c.outputBuffer.Len() == 0 || onConnect != nil || onRequest != nil {
 		c.outputBuffer.Close()
diff --git a/connection_test.go b/connection_test.go
index 6de6f017..782e85c2 100644
--- a/connection_test.go
+++ b/connection_test.go
@@ -25,6 +25,7 @@ import (
 	"net"
 	"os"
 	"runtime"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"syscall"
@@ -638,3 +639,39 @@ func TestConnectionServerClose(t *testing.T) {
 	//time.Sleep(time.Second)
 	wg.Wait()
 }
+
+func TestConnectionDailTimeoutAndClose(t *testing.T) {
+	ln, err := createTestListener("tcp", ":12345")
+	MustNil(t, err)
+	defer ln.Close()
+
+	el, err := NewEventLoop(
+		func(ctx context.Context, connection Connection) error {
+			_, err = connection.Reader().Next(connection.Reader().Len())
+			return err
+		},
+	)
+	defer el.Shutdown(context.Background())
+	go func() {
+		err := el.Serve(ln)
+		if err != nil {
+			t.Logf("servce end with error: %v", err)
+		}
+	}()
+
+	loops := 100
+	conns := 100
+	for l := 0; l < loops; l++ {
+		var wg sync.WaitGroup
+		wg.Add(conns)
+		for i := 0; i < conns; i++ {
+			go func() {
+				defer wg.Done()
+				conn, err := DialConnection("tcp", ":12345", time.Nanosecond)
+				Assert(t, err == nil || strings.Contains(err.Error(), "i/o timeout"))
+				_ = conn
+			}()
+		}
+		wg.Wait()
+	}
+}
diff --git a/net_polldesc.go b/net_polldesc.go
index dfd95de1..c197850e 100644
--- a/net_polldesc.go
+++ b/net_polldesc.go
@@ -53,14 +53,17 @@ func (pd *pollDesc) WaitWrite(ctx context.Context) (err error) {
 	}
 
 	select {
-	case <-pd.closeTrigger:
+	case <-pd.closeTrigger: // triggered by poller
 		// no need to detach, since poller has done it in OnHup.
 		return Exception(ErrConnClosed, "by peer")
-	case <-pd.writeTrigger:
+	case <-pd.writeTrigger: // triggered by poller
 		err = nil
-	case <-ctx.Done():
+	case <-ctx.Done(): // triggered by ctx
 		// deregister from poller, upper caller function will close fd
+		// detach first but there's a very small possibility that operator is doing in poller,
+		// so need call unused() to wait operator done
 		pd.detach()
+		pd.operator.unused()
 		err = mapErr(ctx.Err())
 	}
 	// double check close trigger