Vendoring swarmkit for TCP hack

Also Cherry-pick moby/swarmkit#1651
to identify the issue in allocator.

Signed-off-by: Madhu Venugopal <madhu@docker.com>

Author: mavenugo

daemon/cluster/cluster.go

@@ -282,7 +282,7 @@ func (c *Cluster) startNewNode(forceNewCluster bool, localAddr, remoteAddr, list
 	n, err := swarmagent.NewNode(&swarmagent.NodeConfig{
 		Hostname:           c.config.Name,
 		ForceNewCluster:    forceNewCluster,
-		ListenControlAPI:   filepath.Join(c.runtimeRoot, controlSocket),
+		ListenControlAPI:   "127.0.0.1:1234",
 		ListenRemoteAPI:    listenAddr,
 		AdvertiseRemoteAPI: advertiseAddr,
 		JoinAddr:           joinAddr,

hack/vendor.sh

@@ -143,7 +143,7 @@ clone git github.com/docker/docker-credential-helpers v0.3.0
 clone git github.com/docker/containerd 837e8c5e1cad013ed57f5c2090c8591c10cbbdae
 
 # cluster
-clone git github.com/docker/swarmkit 7e63bdefb94e5bea2641e8bdebae2cfa61a0ed44
+clone git github.com/docker/swarmkit 5fc74cad8d204ef9818d3ef3b78d8ee962f2f852 https://github.com/mavenugo/swarmkit
 clone git github.com/golang/mock bd3c8e81be01eef76d4b503f5e687d2d1354d2d9
 clone git github.com/gogo/protobuf v0.3
 clone git github.com/cloudflare/cfssl 7fb22c8cba7ecaf98e4082d22d65800cf45e042a

vendor/src/github.com/docker/swarmkit/agent/node.go

@@ -4,7 +4,6 @@ import (
 	"crypto/tls"
 	"encoding/json"
 	"io/ioutil"
-	"net"
 	"os"
 	"path/filepath"
 	"reflect"
@@ -527,10 +526,6 @@ func (n *Node) initManagerConnection(ctx context.Context, ready chan<- struct{})
 	// Using listen address instead of advertised address because this is a
 	// local connection.
 	addr := n.config.ListenControlAPI
-	opts = append(opts, grpc.WithDialer(
-		func(addr string, timeout time.Duration) (net.Conn, error) {
-			return net.DialTimeout("unix", addr, timeout)
-		}))
 	conn, err := grpc.Dial(addr, opts...)
 	if err != nil {
 		return err
@@ -595,7 +590,7 @@ func (n *Node) runManager(ctx context.Context, securityConfig *ca.SecurityConfig
 			ForceNewCluster: n.config.ForceNewCluster,
 			ProtoAddr: map[string]string{
 				"tcp":  n.config.ListenRemoteAPI,
-				"unix": n.config.ListenControlAPI,
+				"tcp2": n.config.ListenControlAPI,
 			},
 			AdvertiseAddr:  n.config.AdvertiseRemoteAPI,
 			SecurityConfig: securityConfig,

vendor/src/github.com/docker/swarmkit/manager/allocator/allocator.go

@@ -125,6 +125,8 @@ func (a *Allocator) Run(ctx context.Context) error {
 		aaCopy := aa
 		actor := func() error {
 			wg.Add(1)
+			defer wg.Done()
+
 			// init might return an allocator specific context
 			// which is a child of the passed in context to hold
 			// allocator specific state
@@ -133,10 +135,10 @@ func (a *Allocator) Run(ctx context.Context) error {
 				// if we are failing in the init of
 				// this allocator.
 				aa.cancel()
-				wg.Done()
 				return err
 			}
 
+			wg.Add(1)
 			go func() {
 				defer wg.Done()
 				a.run(ctx, aaCopy)

vendor/src/github.com/docker/swarmkit/manager/allocator/network.go

@@ -68,7 +68,7 @@ type networkContext struct {
 	unallocatedNetworks map[string]*api.Network
 }
 
-func (a *Allocator) doNetworkInit(ctx context.Context) error {
+func (a *Allocator) doNetworkInit(ctx context.Context) (err error) {
 	na, err := networkallocator.New()
 	if err != nil {
 		return err
@@ -81,6 +81,13 @@ func (a *Allocator) doNetworkInit(ctx context.Context) error {
 		unallocatedNetworks: make(map[string]*api.Network),
 		ingressNetwork:      newIngressNetwork(),
 	}
+	a.netCtx = nc
+	defer func() {
+		// Clear a.netCtx if initialization was unsuccessful.
+		if err != nil {
+			a.netCtx = nil
+		}
+	}()
 
 	// Check if we have the ingress network. If not found create
 	// it before reading all network objects for allocation.
@@ -125,7 +132,7 @@ func (a *Allocator) doNetworkInit(ctx context.Context) error {
 	// that the we can get the preferred subnet for ingress
 	// network.
 	if !na.IsAllocated(nc.ingressNetwork) {
-		if err := a.allocateNetwork(ctx, nc, nc.ingressNetwork); err != nil {
+		if err := a.allocateNetwork(ctx, nc.ingressNetwork); err != nil {
 			log.G(ctx).WithError(err).Error("failed allocating ingress network during init")
 		}
 
@@ -155,7 +162,7 @@ func (a *Allocator) doNetworkInit(ctx context.Context) error {
 			continue
 		}
 
-		if err := a.allocateNetwork(ctx, nc, n); err != nil {
+		if err := a.allocateNetwork(ctx, n); err != nil {
 			log.G(ctx).WithError(err).Errorf("failed allocating network %s during init", n.ID)
 		}
 	}
@@ -179,7 +186,7 @@ func (a *Allocator) doNetworkInit(ctx context.Context) error {
 		}
 
 		node.Attachment.Network = nc.ingressNetwork.Copy()
-		if err := a.allocateNode(ctx, nc, node); err != nil {
+		if err := a.allocateNode(ctx, node); err != nil {
 			log.G(ctx).WithError(err).Errorf("Failed to allocate network resources for node %s during init", node.ID)
 		}
 	}
@@ -198,7 +205,7 @@ func (a *Allocator) doNetworkInit(ctx context.Context) error {
 			continue
 		}
 
-		if err := a.allocateService(ctx, nc, s); err != nil {
+		if err := a.allocateService(ctx, s); err != nil {
 			log.G(ctx).WithError(err).Errorf("failed allocating service %s during init", s.ID)
 		}
 	}
@@ -260,7 +267,7 @@ func (a *Allocator) doNetworkInit(ctx context.Context) error {
 			}
 
 			err := batch.Update(func(tx store.Tx) error {
-				_, err := a.allocateTask(ctx, nc, tx, t)
+				_, err := a.allocateTask(ctx, tx, t)
 				return err
 			})
 			if err != nil {
@@ -274,7 +281,6 @@ func (a *Allocator) doNetworkInit(ctx context.Context) error {
 		return err
 	}
 
-	a.netCtx = nc
 	return nil
 }
 
@@ -288,7 +294,7 @@ func (a *Allocator) doNetworkAlloc(ctx context.Context, ev events.Event) {
 			break
 		}
 
-		if err := a.allocateNetwork(ctx, nc, n); err != nil {
+		if err := a.allocateNetwork(ctx, n); err != nil {
 			log.G(ctx).WithError(err).Errorf("Failed allocation for network %s", n.ID)
 			break
 		}
@@ -309,7 +315,7 @@ func (a *Allocator) doNetworkAlloc(ctx context.Context, ev events.Event) {
 			break
 		}
 
-		if err := a.allocateService(ctx, nc, s); err != nil {
+		if err := a.allocateService(ctx, s); err != nil {
 			log.G(ctx).WithError(err).Errorf("Failed allocation for service %s", s.ID)
 			break
 		}
@@ -320,7 +326,7 @@ func (a *Allocator) doNetworkAlloc(ctx context.Context, ev events.Event) {
 			break
 		}
 
-		if err := a.allocateService(ctx, nc, s); err != nil {
+		if err := a.allocateService(ctx, s); err != nil {
 			log.G(ctx).WithError(err).Errorf("Failed allocation during update of service %s", s.ID)
 			break
 		}
@@ -335,18 +341,18 @@ func (a *Allocator) doNetworkAlloc(ctx context.Context, ev events.Event) {
 		// it's still there.
 		delete(nc.unallocatedServices, s.ID)
 	case state.EventCreateNode, state.EventUpdateNode, state.EventDeleteNode:
-		a.doNodeAlloc(ctx, nc, ev)
+		a.doNodeAlloc(ctx, ev)
 	case state.EventCreateTask, state.EventUpdateTask, state.EventDeleteTask:
-		a.doTaskAlloc(ctx, nc, ev)
+		a.doTaskAlloc(ctx, ev)
 	case state.EventCommit:
-		a.procUnallocatedNetworks(ctx, nc)
-		a.procUnallocatedServices(ctx, nc)
-		a.procUnallocatedTasksNetwork(ctx, nc)
+		a.procUnallocatedNetworks(ctx)
+		a.procUnallocatedServices(ctx)
+		a.procUnallocatedTasksNetwork(ctx)
 		return
 	}
 }
 
-func (a *Allocator) doNodeAlloc(ctx context.Context, nc *networkContext, ev events.Event) {
+func (a *Allocator) doNodeAlloc(ctx context.Context, ev events.Event) {
 	var (
 		isDelete bool
 		node     *api.Node
@@ -362,6 +368,8 @@ func (a *Allocator) doNodeAlloc(ctx context.Context, nc *networkContext, ev even
 		node = v.Node.Copy()
 	}
 
+	nc := a.netCtx
+
 	if isDelete {
 		if nc.nwkAllocator.IsNodeAllocated(node) {
 			if err := nc.nwkAllocator.DeallocateNode(node); err != nil {
@@ -377,7 +385,7 @@ func (a *Allocator) doNodeAlloc(ctx context.Context, nc *networkContext, ev even
 		}
 
 		node.Attachment.Network = nc.ingressNetwork.Copy()
-		if err := a.allocateNode(ctx, nc, node); err != nil {
+		if err := a.allocateNode(ctx, node); err != nil {
 			log.G(ctx).WithError(err).Errorf("Failed to allocate network resources for node %s", node.ID)
 		}
 	}
@@ -464,7 +472,7 @@ func (a *Allocator) taskCreateNetworkAttachments(t *api.Task, s *api.Service) {
 	taskUpdateNetworks(t, networks)
 }
 
-func (a *Allocator) doTaskAlloc(ctx context.Context, nc *networkContext, ev events.Event) {
+func (a *Allocator) doTaskAlloc(ctx context.Context, ev events.Event) {
 	var (
 		isDelete bool
 		t        *api.Task
@@ -480,6 +488,8 @@ func (a *Allocator) doTaskAlloc(ctx context.Context, nc *networkContext, ev even
 		t = v.Task.Copy()
 	}
 
+	nc := a.netCtx
+
 	// If the task has stopped running or it's being deleted then
 	// we should free the network resources associated with the
 	// task right away.
@@ -530,7 +540,9 @@ func (a *Allocator) doTaskAlloc(ctx context.Context, nc *networkContext, ev even
 	nc.unallocatedTasks[t.ID] = t
 }
 
-func (a *Allocator) allocateNode(ctx context.Context, nc *networkContext, node *api.Node) error {
+func (a *Allocator) allocateNode(ctx context.Context, node *api.Node) error {
+	nc := a.netCtx
+
 	if err := nc.nwkAllocator.AllocateNode(node); err != nil {
 		return err
 	}
@@ -563,7 +575,9 @@ func (a *Allocator) allocateNode(ctx context.Context, nc *networkContext, node *
 	return nil
 }
 
-func (a *Allocator) allocateService(ctx context.Context, nc *networkContext, s *api.Service) error {
+func (a *Allocator) allocateService(ctx context.Context, s *api.Service) error {
+	nc := a.netCtx
+
 	if s.Spec.Endpoint != nil {
 		// service has user-defined endpoint
 		if s.Endpoint == nil {
@@ -648,7 +662,9 @@ func (a *Allocator) allocateService(ctx context.Context, nc *networkContext, s *
 	return nil
 }
 
-func (a *Allocator) allocateNetwork(ctx context.Context, nc *networkContext, n *api.Network) error {
+func (a *Allocator) allocateNetwork(ctx context.Context, n *api.Network) error {
+	nc := a.netCtx
+
 	if err := nc.nwkAllocator.Allocate(n); err != nil {
 		nc.unallocatedNetworks[n.ID] = n
 		return errors.Wrapf(err, "failed during network allocation for network %s", n.ID)
@@ -670,7 +686,7 @@ func (a *Allocator) allocateNetwork(ctx context.Context, nc *networkContext, n *
 	return nil
 }
 
-func (a *Allocator) allocateTask(ctx context.Context, nc *networkContext, tx store.Tx, t *api.Task) (*api.Task, error) {
+func (a *Allocator) allocateTask(ctx context.Context, tx store.Tx, t *api.Task) (*api.Task, error) {
 	taskUpdated := false
 
 	// Get the latest task state from the store before updating.
@@ -679,6 +695,8 @@ func (a *Allocator) allocateTask(ctx context.Context, nc *networkContext, tx sto
 		return nil, fmt.Errorf("could not find task %s while trying to update network allocation", t.ID)
 	}
 
+	nc := a.netCtx
+
 	// We might be here even if a task allocation has already
 	// happened but wasn't successfully committed to store. In such
 	// cases skip allocation and go straight ahead to updating the
@@ -738,10 +756,11 @@ func (a *Allocator) allocateTask(ctx context.Context, nc *networkContext, tx sto
 	return storeT, nil
 }
 
-func (a *Allocator) procUnallocatedNetworks(ctx context.Context, nc *networkContext) {
+func (a *Allocator) procUnallocatedNetworks(ctx context.Context) {
+	nc := a.netCtx
 	for _, n := range nc.unallocatedNetworks {
 		if !nc.nwkAllocator.IsAllocated(n) {
-			if err := a.allocateNetwork(ctx, nc, n); err != nil {
+			if err := a.allocateNetwork(ctx, n); err != nil {
 				log.G(ctx).Debugf("Failed allocation of unallocated network %s: %v", n.ID, err)
 				continue
 			}
@@ -751,10 +770,11 @@ func (a *Allocator) procUnallocatedNetworks(ctx context.Context, nc *networkCont
 	}
 }
 
-func (a *Allocator) procUnallocatedServices(ctx context.Context, nc *networkContext) {
+func (a *Allocator) procUnallocatedServices(ctx context.Context) {
+	nc := a.netCtx
 	for _, s := range nc.unallocatedServices {
 		if !nc.nwkAllocator.IsServiceAllocated(s) {
-			if err := a.allocateService(ctx, nc, s); err != nil {
+			if err := a.allocateService(ctx, s); err != nil {
 				log.G(ctx).Debugf("Failed allocation of unallocated service %s: %v", s.ID, err)
 				continue
 			}
@@ -764,15 +784,16 @@ func (a *Allocator) procUnallocatedServices(ctx context.Context, nc *networkCont
 	}
 }
 
-func (a *Allocator) procUnallocatedTasksNetwork(ctx context.Context, nc *networkContext) {
+func (a *Allocator) procUnallocatedTasksNetwork(ctx context.Context) {
+	nc := a.netCtx
 	tasks := make([]*api.Task, 0, len(nc.unallocatedTasks))
 
 	committed, err := a.store.Batch(func(batch *store.Batch) error {
 		for _, t := range nc.unallocatedTasks {
 			var allocatedT *api.Task
 			err := batch.Update(func(tx store.Tx) error {
 				var err error
-				allocatedT, err = a.allocateTask(ctx, nc, tx, t)
+				allocatedT, err = a.allocateTask(ctx, tx, t)
 				return err
 			})
 

vendor/src/github.com/docker/swarmkit/manager/allocator/networkallocator/networkallocator.go

@@ -738,7 +738,7 @@ func (na *NetworkAllocator) allocatePools(n *api.Network) (map[string]string, er
 			ic.Subnet = poolIP.String()
 		}
 
-		if ic.Gateway == "" {
+		if ic.Gateway == "" && gwIP != nil {
 			ic.Gateway = gwIP.IP.String()
 		}
 

vendor/src/github.com/docker/swarmkit/manager/manager.go

@@ -141,12 +141,7 @@ func New(config *Config) (*Manager, error) {
 		tcpAddr = net.JoinHostPort("0.0.0.0", tcpAddrPort)
 	}
 
-	err := os.MkdirAll(filepath.Dir(config.ProtoAddr["unix"]), 0700)
-	if err != nil {
-		return nil, errors.Wrap(err, "failed to create socket directory")
-	}
-
-	err = os.MkdirAll(config.StateDir, 0700)
+	err := os.MkdirAll(config.StateDir, 0700)
 	if err != nil {
 		return nil, errors.Wrap(err, "failed to create state directory")
 	}
@@ -164,7 +159,11 @@ func New(config *Config) (*Manager, error) {
 		listeners = make(map[string]net.Listener)
 
 		for proto, addr := range config.ProtoAddr {
-			l, err := net.Listen(proto, addr)
+			p := proto
+			if proto == "tcp2" {
+				p = "tcp"
+			}
+			l, err := net.Listen(p, addr)
 
 			// A unix socket may fail to bind if the file already
 			// exists. Try replacing the file.
@@ -549,12 +548,12 @@ func (m *Manager) serveListener(ctx context.Context, errServe chan error, proto
 		logrus.Fields{
 			"proto": lis.Addr().Network(),
 			"addr":  lis.Addr().String()}))
-	if proto == "unix" {
+	if proto == "tcp2" {
 		log.G(ctx).Info("Listening for local connections")
 		// we need to disallow double closes because UnixListener.Close
 		// can delete unix-socket file of newer listener. grpc calls
 		// Close twice indeed: in Serve and in Stop.
-		errServe <- m.localserver.Serve(&closeOnceListener{Listener: lis})
+		errServe <- m.localserver.Serve(lis)
 	} else {
 		log.G(ctx).Info("Listening for connections")
 		errServe <- m.server.Serve(lis)