Retry queries if a write to the stream fails.

Previously, we'd only retry when a write failed (in case the peer was using
single-use streams). However, in the new world of half-open streams, writes
still succeed on remote-closed streams so this isn't sufficient. Instead, we
need to retry on read failure as well.

Note: this was technically broken before because a peer could write on a stream
before receiving the close message causing the write to succeed but the
subsequent read to fail.

dht_net.go

@@ -161,73 +161,88 @@ const streamReuseTries = 3
 func (ms *messageSender) SendMessage(ctx context.Context, pmes *pb.Message) error {
 	ms.lk.Lock()
 	defer ms.lk.Unlock()
-	if err := ms.prep(); err != nil {
-		return err
-	}
-
-	if err := ms.writeMessage(pmes); err != nil {
-		return err
-	}
-
-	if ms.singleMes > streamReuseTries {
-		ms.s.Close()
-		ms.s = nil
-	}
-
-	return nil
-}
-
-func (ms *messageSender) writeMessage(pmes *pb.Message) error {
-	err := ms.w.WriteMsg(pmes)
-	if err != nil {
-		// If the other side isnt expecting us to be reusing streams, we're gonna
-		// end up erroring here. To make sure things work seamlessly, lets retry once
-		// before continuing
-
-		log.Infof("error writing message: ", err)
-		ms.s.Close()
-		ms.s = nil
+	retry := false
+	for {
 		if err := ms.prep(); err != nil {
 			return err
 		}
 
 		if err := ms.w.WriteMsg(pmes); err != nil {
-			return err
+			ms.s.Reset()
+			ms.s = nil
+
+			if retry {
+				log.Info("error writing message, bailing: ", err)
+				return err
+			} else {
+				log.Info("error writing message, trying again: ", err)
+				retry = true
+				continue
+			}
+		}
+
+		log.Event(ctx, "dhtSentMessage", ms.dht.self, ms.p, pmes)
+
+		if ms.singleMes > streamReuseTries {
+			ms.s.Close()
+			ms.s = nil
+		} else if retry {
+			ms.singleMes++
 		}
 
-		// keep track of this happening. If it happens a few times, its
-		// likely we can assume the otherside will never support stream reuse
-		ms.singleMes++
+		return nil
 	}
-	return nil
 }
 
 func (ms *messageSender) SendRequest(ctx context.Context, pmes *pb.Message) (*pb.Message, error) {
 	ms.lk.Lock()
 	defer ms.lk.Unlock()
-	if err := ms.prep(); err != nil {
-		return nil, err
-	}
+	retry := false
+	for {
+		if err := ms.prep(); err != nil {
+			return nil, err
+		}
 
-	if err := ms.writeMessage(pmes); err != nil {
-		return nil, err
-	}
+		if err := ms.w.WriteMsg(pmes); err != nil {
+			ms.s.Reset()
+			ms.s = nil
+
+			if retry {
+				log.Info("error writing message, bailing: ", err)
+				return nil, err
+			} else {
+				log.Info("error writing message, trying again: ", err)
+				retry = true
+				continue
+			}
+		}
 
-	log.Event(ctx, "dhtSentMessage", ms.dht.self, ms.p, pmes)
+		mes := new(pb.Message)
+		if err := ms.ctxReadMsg(ctx, mes); err != nil {
+			ms.s.Reset()
+			ms.s = nil
+
+			if retry {
+				log.Info("error reading message, bailing: ", err)
+				return nil, err
+			} else {
+				log.Info("error reading message, trying again: ", err)
+				retry = true
+				continue
+			}
+		}
 
-	mes := new(pb.Message)
-	if err := ms.ctxReadMsg(ctx, mes); err != nil {
-		ms.s.Reset()
-		ms.s = nil
-		return nil, err
-	}
+		log.Event(ctx, "dhtSentMessage", ms.dht.self, ms.p, pmes)
 
-	if ms.singleMes > streamReuseTries {
-		ms.s.Close()
-		ms.s = nil
-	}
+		if ms.singleMes > streamReuseTries {
+			ms.s.Close()
+			ms.s = nil
+		} else if retry {
+			ms.singleMes++
+		}
 
-	return mes, nil
+		return mes, nil
+	}
 }
 
 func (ms *messageSender) ctxReadMsg(ctx context.Context, mes *pb.Message) error {

ext_test.go

@@ -192,7 +192,6 @@ func TestNotFound(t *testing.T) {
 				if err := pbw.WriteMsg(resp); err != nil {
 					panic(err)
 				}
-
 			default:
 				panic("Shouldnt recieve this.")
 			}
@@ -288,3 +287,68 @@ func TestLessThanKResponses(t *testing.T) {
 	}
 	t.Fatal("Expected to recieve an error.")
 }
+
+// Test multiple queries against a node that closes its stream after every query.
+func TestMultipleQueries(t *testing.T) {
+	if testing.Short() {
+		t.SkipNow()
+	}
+
+	ctx := context.Background()
+	mn, err := mocknet.FullMeshConnected(ctx, 2)
+	if err != nil {
+		t.Fatal(err)
+	}
+	hosts := mn.Hosts()
+	tsds := dssync.MutexWrap(ds.NewMapDatastore())
+	d := NewDHT(ctx, hosts[0], tsds)
+
+	d.Update(ctx, hosts[1].ID())
+
+	// It would be nice to be able to just get a value and succeed but then
+	// we'd need to deal with selectors and validators...
+	hosts[1].SetStreamHandler(ProtocolDHT, func(s inet.Stream) {
+		defer s.Close()
+
+		pbr := ggio.NewDelimitedReader(s, inet.MessageSizeMax)
+		pbw := ggio.NewDelimitedWriter(s)
+
+		pmes := new(pb.Message)
+		if err := pbr.ReadMsg(pmes); err != nil {
+			panic(err)
+		}
+
+		switch pmes.GetType() {
+		case pb.Message_GET_VALUE:
+			pi := hosts[1].Peerstore().PeerInfo(hosts[0].ID())
+			resp := &pb.Message{
+				Type:        pmes.Type,
+				CloserPeers: pb.PeerInfosToPBPeers(d.host.Network(), []pstore.PeerInfo{pi}),
+			}
+
+			if err := pbw.WriteMsg(resp); err != nil {
+				panic(err)
+			}
+		default:
+			panic("Shouldnt recieve this.")
+		}
+	})
+
+	// long timeout to ensure timing is not at play.
+	ctx, cancel := context.WithTimeout(ctx, time.Second*20)
+	defer cancel()
+	for i := 0; i < 10; i++ {
+		if _, err := d.GetValue(ctx, "hello"); err != nil {
+			switch err {
+			case routing.ErrNotFound:
+				//Success!
+				continue
+			case u.ErrTimeout:
+				t.Fatal("Should not have gotten timeout!")
+			default:
+				t.Fatalf("Got unexpected error: %s", err)
+			}
+		}
+		t.Fatal("Expected to recieve an error.")
+	}
+}