[networkinfo] retry for bootnode connection

We experienced a few cases that node failed to connect to bootnode and
the node program will not be useable. So, we add a retry mechanism to
increase the connection times to bootnode.

Signed-off-by: Leo Chen <leo@harmony.one>

api/service/networkinfo/service.go

@@ -30,12 +30,23 @@ type Service struct {
 	peerInfo    <-chan peerstore.PeerInfo
 	discovery   *libp2pdis.RoutingDiscovery
 	messageChan chan *msg_pb.Message
+	started     bool
 }
 
+// ConnectionRetry set the number of retry of connection to bootnode in case the initial connection is failed
+var (
+	// retry for 10 minutes and give up then
+	ConnectionRetry = 300
+)
+
+const (
+	waitInRetry       = 2 * time.Second
+	connectionTimeout = 3 * time.Minute
+)
+
 // New returns role conversion service.
 func New(h p2p.Host, rendezvous p2p.GroupID, peerChan chan p2p.Peer, bootnodes utils.AddrList) *Service {
-	timeout := 30 * time.Minute
+	ctx, cancel := context.WithTimeout(context.Background(), connectionTimeout)
-	ctx, cancel := context.WithTimeout(context.Background(), timeout)
 	dht, err := libp2pdht.New(ctx, h.GetP2PHost())
 	if err != nil {
 		panic(err)
@@ -51,13 +62,20 @@ func New(h p2p.Host, rendezvous p2p.GroupID, peerChan chan p2p.Peer, bootnodes u
 		stoppedChan: make(chan struct{}),
 		peerChan:    peerChan,
 		bootnodes:   bootnodes,
+		discovery:   nil,
+		started:     false,
 	}
 }
 
 // StartService starts network info service.
 func (s *Service) StartService() {
-	s.Init()
+	err := s.Init()
+	if err != nil {
+		utils.GetLogInstance().Error("Service Init Failed", "error", err)
+		return
+	}
 	s.Run()
+	s.started = true
 }
 
 // Init initializes role conversion service.
@@ -76,20 +94,31 @@ func (s *Service) Init() error {
 		s.bootnodes = utils.BootNodes
 	}
 
+	connected := false
 	for _, peerAddr := range s.bootnodes {
 		peerinfo, _ := peerstore.InfoFromP2pAddr(peerAddr)
 		wg.Add(1)
 		go func() {
 			defer wg.Done()
+			for i := 0; i < ConnectionRetry; i++ {
 				if err := s.Host.GetP2PHost().Connect(s.ctx, *peerinfo); err != nil {
-				utils.GetLogInstance().Warn("can't connect to bootnode", "error", err)
+					utils.GetLogInstance().Warn("can't connect to bootnode", "error", err, "try", i)
+					time.Sleep(waitInRetry)
 				} else {
-				utils.GetLogInstance().Info("connected to bootnode", "node", *peerinfo)
+					utils.GetLogInstance().Info("connected to bootnode", "node", *peerinfo, "try", i)
+					// it is okay if any bootnode is connected
+					connected = true
+					break
+				}
 			}
 		}()
 	}
 	wg.Wait()
 
+	if !connected {
+		return fmt.Errorf("[FATAL] error connecting to bootnodes")
+	}
+
 	// We use a rendezvous point "shardID" to announce our location.
 	utils.GetLogInstance().Info("Announcing ourselves...")
 	s.discovery = libp2pdis.NewRoutingDiscovery(s.dht)
@@ -102,10 +131,16 @@ func (s *Service) Init() error {
 // Run runs network info.
 func (s *Service) Run() {
 	defer close(s.stoppedChan)
+	if s.discovery == nil {
+		utils.GetLogInstance().Error("discovery is not initialized")
+		return
+	}
+
 	var err error
 	s.peerInfo, err = s.discovery.FindPeers(s.ctx, string(s.Rendezvous))
 	if err != nil {
 		utils.GetLogInstance().Error("FindPeers", "error", err)
+		return
 	}
 
 	go s.DoService()
@@ -160,6 +195,11 @@ func (s *Service) StopService() {
 	utils.GetLogInstance().Info("Stopping network info service.")
 	defer s.cancel()
 
+	if !s.started {
+		utils.GetLogInstance().Info("Service didn't started. Exit.")
+		return
+	}
+
 	s.stopChan <- struct{}{}
 	<-s.stoppedChan
 	utils.GetLogInstance().Info("Network info service stopped.")

api/service/networkinfo/service_test.go

@@ -25,9 +25,11 @@ func TestService(t *testing.T) {
 		t.Fatal("unable to new host in harmony")
 	}
 
-	s := New(host, "teststring", nil, nil)
+	s := New(host, p2p.GroupIDBeaconClient, nil, nil)
+
 	s.StartService()
 
 	time.Sleep(2 * time.Second)
+
 	s.StopService()
 }

internal/wallet/wallet/lib_test.go

@@ -7,6 +7,7 @@ import (
 	"github.com/golang/mock/gomock"
 	"github.com/harmony-one/harmony/api/client"
 	proto_node "github.com/harmony-one/harmony/api/proto/node"
+	"github.com/harmony-one/harmony/api/service/networkinfo"
 	"github.com/harmony-one/harmony/core/types"
 	"github.com/harmony-one/harmony/node"
 	"github.com/harmony-one/harmony/p2p"
@@ -15,6 +16,9 @@ import (
 )
 
 func TestCreateWalletNode(test *testing.T) {
+	// shorten the retry time
+	networkinfo.ConnectionRetry = 3
+
 	walletNode := CreateWalletNode()
 
 	if walletNode.Client == nil {