grpclb: support pickfirst (#2652)

Use pickfirst as balancing policy for backends. It can only set at init time.
Does not support changing with service config now.

balancer/grpclb/grpclb.go

@@ -129,8 +129,19 @@ func newLBBuilderWithFallbackTimeout(fallbackTimeout time.Duration) balancer.Bui
 	}
 }
 
+// newLBBuilderWithPickFirst creates a grpclb builder with pick-first.
+func newLBBuilderWithPickFirst() balancer.Builder {
+	return &lbBuilder{
+		usePickFirst: true,
+	}
+}
+
 type lbBuilder struct {
 	fallbackTimeout time.Duration
+
+	// TODO: delete this when balancer can handle service config. This should be
+	//  updated by service config.
+	usePickFirst bool // Use roundrobin or pickfirst for backends.
 }
 
 func (b *lbBuilder) Name() string {
@@ -156,6 +167,7 @@ func (b *lbBuilder) Build(cc balancer.ClientConn, opt balancer.BuildOptions) bal
 		cc:              newLBCacheClientConn(cc),
 		target:          target,
 		opt:             opt,
+		usePickFirst:    b.usePickFirst,
 		fallbackTimeout: b.fallbackTimeout,
 		doneCh:          make(chan struct{}),
 
@@ -188,6 +200,8 @@ type lbBalancer struct {
 	target string
 	opt    balancer.BuildOptions
 
+	usePickFirst bool
+
 	// grpclbClientConnCreds is the creds bundle to be used to connect to grpclb
 	// servers. If it's nil, use the TransportCredentials from BuildOptions
 	// instead.
@@ -249,7 +263,16 @@ func (lb *lbBalancer) regeneratePicker(resetDrop bool) {
 		lb.picker = &errPicker{err: balancer.ErrTransientFailure}
 		return
 	}
+
 	var readySCs []balancer.SubConn
+	if lb.usePickFirst {
+		if lb.state == connectivity.Ready || lb.state == connectivity.Idle {
+			for _, sc := range lb.subConns {
+				readySCs = append(readySCs, sc)
+				break
+			}
+		}
+	} else {
 		for _, a := range lb.backendAddrs {
 			if sc, ok := lb.subConns[a]; ok {
 				if st, ok := lb.scStates[sc]; ok && st == connectivity.Ready {
@@ -257,6 +280,7 @@ func (lb *lbBalancer) regeneratePicker(resetDrop bool) {
 				}
 			}
 		}
+	}
 
 	if len(readySCs) <= 0 {
 		// If there's no ready SubConns, always re-pick. This is to avoid drops

balancer/grpclb/grpclb_remote_balancer.go

@@ -96,14 +96,36 @@ func (lb *lbBalancer) processServerList(l *lbpb.ServerList) {
 // indicating whether the backendAddrs are different from the cached
 // backendAddrs (whether any SubConn was newed/removed).
 // Caller must hold lb.mu.
-func (lb *lbBalancer) refreshSubConns(backendAddrs []resolver.Address, fromGRPCLBServer bool) bool {
+func (lb *lbBalancer) refreshSubConns(backendAddrs []resolver.Address, fromGRPCLBServer bool) {
 	opts := balancer.NewSubConnOptions{}
 	if fromGRPCLBServer {
 		opts.CredsBundle = lb.grpclbBackendCreds
 	}
 
 	lb.backendAddrs = nil
-	var backendsUpdated bool
+
+	if lb.usePickFirst {
+		var sc balancer.SubConn
+		for _, sc = range lb.subConns {
+			break
+		}
+		if sc != nil {
+			sc.UpdateAddresses(backendAddrs)
+			sc.Connect()
+			return
+		}
+		// This bypasses the cc wrapper with SubConn cache.
+		sc, err := lb.cc.cc.NewSubConn(backendAddrs, opts)
+		if err != nil {
+			grpclog.Warningf("grpclb: failed to create new SubConn: %v", err)
+			return
+		}
+		sc.Connect()
+		lb.subConns[backendAddrs[0]] = sc
+		lb.scStates[sc] = connectivity.Idle
+		return
+	}
+
 	// addrsSet is the set converted from backendAddrs, it's used to quick
 	// lookup for an address.
 	addrsSet := make(map[resolver.Address]struct{})
@@ -115,12 +137,10 @@ func (lb *lbBalancer) refreshSubConns(backendAddrs []resolver.Address, fromGRPCL
 		lb.backendAddrs = append(lb.backendAddrs, addrWithoutMD)
 
 		if _, ok := lb.subConns[addrWithoutMD]; !ok {
-			backendsUpdated = true
-
 			// Use addrWithMD to create the SubConn.
 			sc, err := lb.cc.NewSubConn([]resolver.Address{addr}, opts)
 			if err != nil {
-				grpclog.Warningf("roundrobinBalancer: failed to create new SubConn: %v", err)
+				grpclog.Warningf("grpclb: failed to create new SubConn: %v", err)
 				continue
 			}
 			lb.subConns[addrWithoutMD] = sc // Use the addr without MD as key for the map.
@@ -136,16 +156,12 @@ func (lb *lbBalancer) refreshSubConns(backendAddrs []resolver.Address, fromGRPCL
 	for a, sc := range lb.subConns {
 		// a was removed by resolver.
 		if _, ok := addrsSet[a]; !ok {
-			backendsUpdated = true
-
 			lb.cc.RemoveSubConn(sc)
 			delete(lb.subConns, a)
 			// Keep the state of this sc in b.scStates until sc's state becomes Shutdown.
 			// The entry will be deleted in HandleSubConnStateChange.
 		}
 	}
-
-	return backendsUpdated
 }
 
 func (lb *lbBalancer) readServerList(s *balanceLoadClientStream) error {

balancer/grpclb/grpclb_test.go

@@ -751,6 +751,107 @@ func TestFallback(t *testing.T) {
 	t.Fatalf("No RPC sent to backend behind remote balancer after 1 second")
 }
 
+// The remote balancer sends response with duplicates to grpclb client.
+func TestGRPCLBPickFirst(t *testing.T) {
+	balancer.Register(newLBBuilderWithPickFirst())
+	defer balancer.Register(newLBBuilder())
+
+	defer leakcheck.Check(t)
+
+	r, cleanup := manual.GenerateAndRegisterManualResolver()
+	defer cleanup()
+
+	tss, cleanup, err := newLoadBalancer(3)
+	if err != nil {
+		t.Fatalf("failed to create new load balancer: %v", err)
+	}
+	defer cleanup()
+
+	beServers := []*lbpb.Server{{
+		IpAddress:        tss.beIPs[0],
+		Port:             int32(tss.bePorts[0]),
+		LoadBalanceToken: lbToken,
+	}, {
+		IpAddress:        tss.beIPs[1],
+		Port:             int32(tss.bePorts[1]),
+		LoadBalanceToken: lbToken,
+	}, {
+		IpAddress:        tss.beIPs[2],
+		Port:             int32(tss.bePorts[2]),
+		LoadBalanceToken: lbToken,
+	}}
+	portsToIndex := make(map[int]int)
+	for i := range beServers {
+		portsToIndex[tss.bePorts[i]] = i
+	}
+
+	creds := serverNameCheckCreds{
+		expected: beServerName,
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	cc, err := grpc.DialContext(ctx, r.Scheme()+":///"+beServerName,
+		grpc.WithTransportCredentials(&creds), grpc.WithContextDialer(fakeNameDialer))
+	if err != nil {
+		t.Fatalf("Failed to dial to the backend %v", err)
+	}
+	defer cc.Close()
+	testC := testpb.NewTestServiceClient(cc)
+
+	r.NewAddress([]resolver.Address{{
+		Addr:       tss.lbAddr,
+		Type:       resolver.GRPCLB,
+		ServerName: lbServerName,
+	}})
+
+	var p peer.Peer
+
+	portPicked1 := 0
+	tss.ls.sls <- &lbpb.ServerList{Servers: beServers[1:2]}
+	for i := 0; i < 1000; i++ {
+		if _, err := testC.EmptyCall(context.Background(), &testpb.Empty{}, grpc.WaitForReady(true), grpc.Peer(&p)); err != nil {
+			t.Fatalf("_.EmptyCall(_, _) = _, %v, want _, <nil>", err)
+		}
+		if portPicked1 == 0 {
+			portPicked1 = p.Addr.(*net.TCPAddr).Port
+			continue
+		}
+		if portPicked1 != p.Addr.(*net.TCPAddr).Port {
+			t.Fatalf("Different backends are picked for RPCs: %v vs %v", portPicked1, p.Addr.(*net.TCPAddr).Port)
+		}
+	}
+
+	portPicked2 := portPicked1
+	tss.ls.sls <- &lbpb.ServerList{Servers: beServers[:1]}
+	for i := 0; i < 1000; i++ {
+		if _, err := testC.EmptyCall(context.Background(), &testpb.Empty{}, grpc.WaitForReady(true), grpc.Peer(&p)); err != nil {
+			t.Fatalf("_.EmptyCall(_, _) = _, %v, want _, <nil>", err)
+		}
+		if portPicked2 == portPicked1 {
+			portPicked2 = p.Addr.(*net.TCPAddr).Port
+			continue
+		}
+		if portPicked2 != p.Addr.(*net.TCPAddr).Port {
+			t.Fatalf("Different backends are picked for RPCs: %v vs %v", portPicked2, p.Addr.(*net.TCPAddr).Port)
+		}
+	}
+
+	portPicked := portPicked2
+	tss.ls.sls <- &lbpb.ServerList{Servers: beServers[1:]}
+	for i := 0; i < 1000; i++ {
+		if _, err := testC.EmptyCall(context.Background(), &testpb.Empty{}, grpc.WaitForReady(true), grpc.Peer(&p)); err != nil {
+			t.Fatalf("_.EmptyCall(_, _) = _, %v, want _, <nil>", err)
+		}
+		if portPicked == portPicked2 {
+			portPicked = p.Addr.(*net.TCPAddr).Port
+			continue
+		}
+		if portPicked != p.Addr.(*net.TCPAddr).Port {
+			t.Fatalf("Different backends are picked for RPCs: %v vs %v", portPicked, p.Addr.(*net.TCPAddr).Port)
+		}
+	}
+}
+
 type failPreRPCCred struct{}
 
 func (failPreRPCCred) GetRequestMetadata(ctx context.Context, uri ...string) (map[string]string, error) {