diff --git a/cluster.go b/cluster.go
index ec5a51f..f9a4fef 100644
--- a/cluster.go
+++ b/cluster.go
@@ -156,7 +156,7 @@ func (c *clusterNodes) All() ([]*clusterNode, error) {
 		return nil, pool.ErrClosed
 	}
 
-	var nodes []*clusterNode
+	nodes := make([]*clusterNode, 0, len(c.nodes))
 	for _, node := range c.nodes {
 		nodes = append(nodes, node)
 	}
@@ -208,7 +208,7 @@ func (c *clusterNodes) Random() (*clusterNode, error) {
 	}
 
 	var nodeErr error
-	for i := 0; i < 10; i++ {
+	for i := 0; i <= c.opt.MaxRedirects; i++ {
 		n := rand.Intn(len(addrs))
 		node, err := c.Get(addrs[n])
 		if err != nil {
@@ -446,6 +446,10 @@ func (c *ClusterClient) Process(cmd Cmder) error {
 		// On network errors try random node.
 		if internal.IsRetryableError(err) {
 			node, err = c.nodes.Random()
+			if err != nil {
+				cmd.setErr(err)
+				return err
+			}
 			continue
 		}
 
@@ -475,6 +479,39 @@ func (c *ClusterClient) Process(cmd Cmder) error {
 	return cmd.Err()
 }
 
+// ForEachNode concurrently calls the fn on each ever known node in the cluster.
+// It returns the first error if any.
+func (c *ClusterClient) ForEachNode(fn func(client *Client) error) error {
+	nodes, err := c.nodes.All()
+	if err != nil {
+		return err
+	}
+
+	var wg sync.WaitGroup
+	errCh := make(chan error, 1)
+	for _, node := range nodes {
+		wg.Add(1)
+		go func(node *clusterNode) {
+			defer wg.Done()
+			err := fn(node.Client)
+			if err != nil {
+				select {
+				case errCh <- err:
+				default:
+				}
+			}
+		}(node)
+	}
+	wg.Wait()
+
+	select {
+	case err := <-errCh:
+		return err
+	default:
+		return nil
+	}
+}
+
 // ForEachMaster concurrently calls the fn on each master node in the cluster.
 // It returns the first error if any.
 func (c *ClusterClient) ForEachMaster(fn func(client *Client) error) error {
@@ -649,10 +686,10 @@ func (c *ClusterClient) pipelineExec(cmds []Cmder) error {
 			}
 
 			failedCmds, err = c.execClusterCmds(cn, cmds, failedCmds)
+			node.Client.putConn(cn, err, false)
 			if err != nil {
 				setFirstErr(err)
 			}
-			node.Client.putConn(cn, err, false)
 		}
 
 		cmdsMap = failedCmds
@@ -686,9 +723,15 @@ func (c *ClusterClient) execClusterCmds(
 			continue
 		}
 
-		if i == 0 && internal.IsNetworkError(err) {
+		if i == 0 && internal.IsRetryableError(err) {
+			node, err := c.nodes.Random()
+			if err != nil {
+				setFirstErr(err)
+				continue
+			}
+
 			cmd.reset()
-			failedCmds[nil] = append(failedCmds[nil], cmds...)
+			failedCmds[node] = append(failedCmds[node], cmds...)
 			break
 		}
 
diff --git a/cluster_test.go b/cluster_test.go
index 2c49f99..3406768 100644
--- a/cluster_test.go
+++ b/cluster_test.go
@@ -602,6 +602,33 @@ var _ = Describe("ClusterClient timeout", func() {
 
 		testTimeout()
 	})
+
+	Context("network timeout", func() {
+		const pause = time.Second
+
+		BeforeEach(func() {
+			opt := redisClusterOptions()
+			opt.ReadTimeout = 100 * time.Millisecond
+			opt.WriteTimeout = 100 * time.Millisecond
+			opt.MaxRedirects = 1
+			client = cluster.clusterClient(opt)
+
+			err := client.ForEachNode(func(client *redis.Client) error {
+				return client.ClientPause(pause).Err()
+			})
+			Expect(err).NotTo(HaveOccurred())
+		})
+
+		AfterEach(func() {
+			Eventually(func() error {
+				return client.ForEachNode(func(client *redis.Client) error {
+					return client.Ping().Err()
+				})
+			}, pause).ShouldNot(HaveOccurred())
+		})
+
+		testTimeout()
+	})
 })
 
 //------------------------------------------------------------------------------
diff --git a/tx.go b/tx.go
index 772b3c9..30beda9 100644
--- a/tx.go
+++ b/tx.go
@@ -176,10 +176,8 @@ func (c *Tx) execCmds(cn *pool.Conn, cmds []Cmder) error {
 	// Loop starts from 1 to omit MULTI cmd.
 	for i := 1; i < cmdsLen; i++ {
 		cmd := cmds[i]
-		if err := cmd.readReply(cn); err != nil {
-			if firstErr == nil {
-				firstErr = err
-			}
+		if err := cmd.readReply(cn); err != nil && firstErr == nil {
+			firstErr = err
 		}
 	}