multi: fix recovering from bad connection.

Author: vmihailenco

cluster_pipeline.go

@@ -63,7 +63,7 @@ func (pipe *ClusterPipeline) Exec() (cmds []Cmder, retErr error) {
 				continue
 			}
 
-			cn, err := client.conn()
+			cn, _, err := client.conn()
 			if err != nil {
 				setCmdsErr(cmds, err)
 				retErr = err

error.go

@@ -26,10 +26,24 @@ func (err redisError) Error() string {
 }
 
 func isNetworkError(err error) bool {
-	if _, ok := err.(net.Error); ok || err == io.EOF {
+	if err == io.EOF {
 		return true
 	}
-	return false
+	_, ok := err.(net.Error)
+	return ok
+}
+
+func isBadConn(cn *conn, ei error) bool {
+	if cn.rd.Buffered() > 0 {
+		return true
+	}
+	if ei == nil {
+		return false
+	}
+	if _, ok := ei.(redisError); ok {
+		return false
+	}
+	return true
 }
 
 func isMovedError(err error) (moved bool, ask bool, addr string) {

multi.go

@@ -10,10 +10,10 @@ var errDiscard = errors.New("redis: Discard can be used only inside Exec")
 
 // Multi implements Redis transactions as described in
 // http://redis.io/topics/transactions. It's NOT safe for concurrent use
-// by multiple goroutines, because Exec resets connection state.
+// by multiple goroutines, because Exec resets list of watched keys.
 // If you don't need WATCH it is better to use Pipeline.
 //
-// TODO(vmihailenco): rename to Tx
+// TODO(vmihailenco): rename to Tx and rework API
 type Multi struct {
 	commandable
 
@@ -34,6 +34,18 @@ func (c *Client) Multi() *Multi {
 	return multi
 }
 
+func (c *Multi) putConn(cn *conn, ei error) {
+	var err error
+	if isBadConn(cn, ei) {
+		err = c.base.connPool.Remove(nil) // nil to force removal
+	} else {
+		err = c.base.connPool.Put(cn)
+	}
+	if err != nil {
+		log.Printf("redis: putConn failed: %s", err)
+	}
+}
+
 func (c *Multi) process(cmd Cmder) {
 	if c.cmds == nil {
 		c.base.process(cmd)
@@ -112,15 +124,18 @@ func (c *Multi) Exec(f func() error) ([]Cmder, error) {
 		return []Cmder{}, nil
 	}
 
-	cn, err := c.base.conn()
+	// Strip MULTI and EXEC commands.
+	retCmds := cmds[1 : len(cmds)-1]
+
+	cn, _, err := c.base.conn()
 	if err != nil {
-		setCmdsErr(cmds[1:len(cmds)-1], err)
-		return cmds[1 : len(cmds)-1], err
+		setCmdsErr(retCmds, err)
+		return retCmds, err
 	}
 
 	err = c.execCmds(cn, cmds)
-	c.base.putConn(cn, err)
-	return cmds[1 : len(cmds)-1], err
+	c.putConn(cn, err)
+	return retCmds, err
 }
 
 func (c *Multi) execCmds(cn *conn, cmds []Cmder) error {

multi_test.go

@@ -119,4 +119,30 @@ var _ = Describe("Multi", func() {
 		Expect(get.Val()).To(Equal("20000"))
 	})
 
+	It("should recover from bad connection", func() {
+		// Put bad connection in the pool.
+		cn, _, err := client.Pool().Get()
+		Expect(err).NotTo(HaveOccurred())
+
+		cn.SetNetConn(&badConn{})
+		err = client.Pool().Put(cn)
+		Expect(err).NotTo(HaveOccurred())
+
+		multi := client.Multi()
+		defer func() {
+			Expect(multi.Close()).NotTo(HaveOccurred())
+		}()
+
+		_, err = multi.Exec(func() error {
+			multi.Ping()
+			return nil
+		})
+		Expect(err).To(MatchError("bad connection"))
+
+		_, err = multi.Exec(func() error {
+			multi.Ping()
+			return nil
+		})
+		Expect(err).NotTo(HaveOccurred())
+	})
 })

pipeline.go

@@ -88,7 +88,7 @@ func (pipe *Pipeline) Exec() (cmds []Cmder, retErr error) {
 
 	failedCmds := cmds
 	for i := 0; i <= pipe.client.opt.MaxRetries; i++ {
-		cn, err := pipe.client.conn()
+		cn, _, err := pipe.client.conn()
 		if err != nil {
 			setCmdsErr(failedCmds, err)
 			return cmds, err

pool.go

@@ -18,7 +18,7 @@ var (
 
 type pool interface {
 	First() *conn
-	Get() (*conn, error)
+	Get() (*conn, bool, error)
 	Put(*conn) error
 	Remove(*conn) error
 	Len() int
@@ -212,33 +212,36 @@ func (p *connPool) new() (*conn, error) {
 }
 
 // Get returns existed connection from the pool or creates a new one.
-func (p *connPool) Get() (*conn, error) {
+func (p *connPool) Get() (cn *conn, isNew bool, err error) {
 	if p.closed() {
-		return nil, errClosed
+		err = errClosed
+		return
 	}
 
 	// Fetch first non-idle connection, if available.
-	if cn := p.First(); cn != nil {
-		return cn, nil
+	if cn = p.First(); cn != nil {
+		return
 	}
 
 	// Try to create a new one.
 	if p.conns.Reserve() {
-		cn, err := p.new()
+		cn, err = p.new()
 		if err != nil {
 			p.conns.Remove(nil)
-			return nil, err
+			return
 		}
 		p.conns.Add(cn)
-		return cn, nil
+		isNew = true
+		return
 	}
 
 	// Otherwise, wait for the available connection.
-	if cn := p.wait(); cn != nil {
-		return cn, nil
+	if cn = p.wait(); cn != nil {
+		return
 	}
 
-	return nil, errPoolTimeout
+	err = errPoolTimeout
+	return
 }
 
 func (p *connPool) Put(cn *conn) error {
@@ -327,8 +330,8 @@ func (p *singleConnPool) First() *conn {
 	return p.cn
 }
 
-func (p *singleConnPool) Get() (*conn, error) {
-	return p.cn, nil
+func (p *singleConnPool) Get() (*conn, bool, error) {
+	return p.cn, false, nil
 }
 
 func (p *singleConnPool) Put(cn *conn) error {
@@ -382,24 +385,25 @@ func (p *stickyConnPool) First() *conn {
 	return cn
 }
 
-func (p *stickyConnPool) Get() (*conn, error) {
+func (p *stickyConnPool) Get() (cn *conn, isNew bool, err error) {
 	defer p.mx.Unlock()
 	p.mx.Lock()
 
 	if p.closed {
-		return nil, errClosed
+		err = errClosed
+		return
 	}
 	if p.cn != nil {
-		return p.cn, nil
+		cn = p.cn
+		return
 	}
 
-	cn, err := p.pool.Get()
+	cn, isNew, err = p.pool.Get()
 	if err != nil {
-		return nil, err
+		return
 	}
 	p.cn = cn
-
-	return p.cn, nil
+	return
 }
 
 func (p *stickyConnPool) put() (err error) {

pool_test.go

@@ -107,7 +107,7 @@ var _ = Describe("pool", func() {
 	})
 
 	It("should remove broken connections", func() {
-		cn, err := client.Pool().Get()
+		cn, _, err := client.Pool().Get()
 		Expect(err).NotTo(HaveOccurred())
 		Expect(cn.Close()).NotTo(HaveOccurred())
 		Expect(client.Pool().Put(cn)).NotTo(HaveOccurred())
@@ -141,12 +141,12 @@ var _ = Describe("pool", func() {
 		pool := client.Pool()
 
 		// Reserve one connection.
-		cn, err := client.Pool().Get()
+		cn, _, err := client.Pool().Get()
 		Expect(err).NotTo(HaveOccurred())
 
 		// Reserve the rest of connections.
 		for i := 0; i < 9; i++ {
-			_, err := client.Pool().Get()
+			_, _, err := client.Pool().Get()
 			Expect(err).NotTo(HaveOccurred())
 		}
 
@@ -191,7 +191,7 @@ func BenchmarkPool(b *testing.B) {
 
 	b.RunParallel(func(pb *testing.PB) {
 		for pb.Next() {
-			conn, err := pool.Get()
+			conn, _, err := pool.Get()
 			if err != nil {
 				b.Fatalf("no error expected on pool.Get but received: %s", err.Error())
 			}

pubsub.go

@@ -47,7 +47,7 @@ func (c *Client) PSubscribe(channels ...string) (*PubSub, error) {
 }
 
 func (c *PubSub) subscribe(cmd string, channels ...string) error {
-	cn, err := c.conn()
+	cn, _, err := c.conn()
 	if err != nil {
 		return err
 	}
@@ -112,7 +112,7 @@ func (c *PubSub) PUnsubscribe(patterns ...string) error {
 }
 
 func (c *PubSub) Ping(payload string) error {
-	cn, err := c.conn()
+	cn, _, err := c.conn()
 	if err != nil {
 		return err
 	}
@@ -208,14 +208,16 @@ func newMessage(reply []interface{}) (interface{}, error) {
 // is not received in time. This is low-level API and most clients
 // should use ReceiveMessage.
 func (c *PubSub) ReceiveTimeout(timeout time.Duration) (interface{}, error) {
-	cn, err := c.conn()
+	cn, _, err := c.conn()
 	if err != nil {
 		return nil, err
 	}
 	cn.ReadTimeout = timeout
 
 	cmd := NewSliceCmd()
-	if err := cmd.readReply(cn); err != nil {
+	err = cmd.readReply(cn)
+	c.putConn(cn, err)
+	if err != nil {
 		return nil, err
 	}
 	return newMessage(cmd.Val())
@@ -229,7 +231,7 @@ func (c *PubSub) Receive() (interface{}, error) {
 }
 
 func (c *PubSub) reconnect() {
-	c.connPool.Remove(nil) // close current connection
+	c.connPool.Remove(nil) // nil to force removal
 	if len(c.channels) > 0 {
 		if err := c.Subscribe(c.channels...); err != nil {
 			log.Printf("redis: Subscribe failed: %s", err)

pubsub_test.go

@@ -254,7 +254,7 @@ var _ = Describe("PubSub", func() {
 		Expect(err).NotTo(HaveOccurred())
 		defer pubsub.Close()
 
-		cn, err := pubsub.Pool().Get()
+		cn, _, err := pubsub.Pool().Get()
 		Expect(err).NotTo(HaveOccurred())
 		cn.SetNetConn(&badConn{
 			readErr:  errTimeout,

redis.go

@@ -16,20 +16,16 @@ func (c *baseClient) String() string {
 	return fmt.Sprintf("Redis<%s db:%d>", c.opt.Addr, c.opt.DB)
 }
 
-func (c *baseClient) conn() (*conn, error) {
+func (c *baseClient) conn() (*conn, bool, error) {
 	return c.connPool.Get()
 }
 
 func (c *baseClient) putConn(cn *conn, ei error) {
 	var err error
-	if cn.rd.Buffered() > 0 {
+	if isBadConn(cn, ei) {
 		err = c.connPool.Remove(cn)
-	} else if ei == nil {
-		err = c.connPool.Put(cn)
-	} else if _, ok := ei.(redisError); ok {
-		err = c.connPool.Put(cn)
 	} else {
-		err = c.connPool.Remove(cn)
+		err = c.connPool.Put(cn)
 	}
 	if err != nil {
 		log.Printf("redis: putConn failed: %s", err)
@@ -42,7 +38,7 @@ func (c *baseClient) process(cmd Cmder) {
 			cmd.reset()
 		}
 
-		cn, err := c.conn()
+		cn, _, err := c.conn()
 		if err != nil {
 			cmd.setErr(err)
 			return

redis_test.go

@@ -157,7 +157,7 @@ var _ = Describe("Client", func() {
 		})
 
 		// Put bad connection in the pool.
-		cn, err := client.Pool().Get()
+		cn, _, err := client.Pool().Get()
 		Expect(err).NotTo(HaveOccurred())
 
 		cn.SetNetConn(&badConn{})

ring.go

@@ -313,7 +313,7 @@ func (pipe *RingPipeline) Exec() (cmds []Cmder, retErr error) {
 
 		for name, cmds := range cmdsMap {
 			client := pipe.ring.shards[name].Client
-			cn, err := client.conn()
+			cn, _, err := client.conn()
 			if err != nil {
 				setCmdsErr(cmds, err)
 				if retErr == nil {