bugfix-node-not-leaving

- Add lock when perform synchronisation on concurrent access

pkg/api/apiserver.go

@@ -235,9 +235,19 @@ func NewSmegServer(conf ApiServerConf) (ApiServer, error) {
 		words:  words,
 	}
 
-	router.GET("/meshes", smegServer.GetMeshes)
-	router.GET("/mesh/:meshid", smegServer.GetMesh)
-	router.POST("/mesh/create", smegServer.CreateMesh)
-	router.POST("/mesh/join", smegServer.JoinMesh)
+	v1 := router.Group("/api/v1")
+	{
+		meshes := v1.Group("/meshes")
+		{
+			meshes.GET("/", smegServer.GetMeshes)
+		}
+		mesh := v1.Group("/mesh")
+		{
+			mesh.GET("/:meshid", smegServer.GetMesh)
+			mesh.POST("/create", smegServer.CreateMesh)
+			mesh.POST("/join", smegServer.JoinMesh)
+		}
+	}
+
 	return smegServer, nil
 }

pkg/conn/cluster.go

@@ -23,9 +23,10 @@ func binarySearch(global []string, selfId string, groupSize int) (int, int) {
 
 	lower := 0
 	higher := len(global) - 1
-	mid := (lower + higher) / 2
 
 	for (higher+1)-lower > groupSize {
+		mid := (lower + higher) / 2
+
 		if global[mid] < selfId {
 			lower = mid + 1
 		} else if global[mid] > selfId {
@@ -33,8 +34,6 @@ func binarySearch(global []string, selfId string, groupSize int) (int, int) {
 		} else {
 			break
 		}
-
-		mid = (lower + higher) / 2
 	}
 
 	return lower, int(math.Min(float64(lower+groupSize), float64(len(global))))

pkg/mesh/manager.go

@@ -10,6 +10,7 @@ import (
 	"github.com/tim-beatham/smegmesh/pkg/conf"
 	"github.com/tim-beatham/smegmesh/pkg/ip"
 	"github.com/tim-beatham/smegmesh/pkg/lib"
+	logging "github.com/tim-beatham/smegmesh/pkg/log"
 	"github.com/tim-beatham/smegmesh/pkg/wg"
 	"golang.zx2c4.com/wireguard/wgctrl"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
@@ -355,7 +356,7 @@ func (s *MeshManagerImpl) LeaveMesh(meshId string) error {
 	err := mesh.RemoveNode(s.HostParameters.GetPublicKey())
 
 	if err != nil {
-		return err
+		logging.Log.WriteErrorf(err.Error())
 	}
 
 	if s.OnDelete != nil {

pkg/sync/syncer.go

@@ -28,6 +28,7 @@ type SyncerImpl struct {
 	cluster        conn.ConnCluster
 	conf           *conf.DaemonConfiguration
 	lastSync       map[string]int64
+	lock           sync.RWMutex
 }
 
 // Sync: Sync with random nodes
@@ -38,11 +39,10 @@ func (s *SyncerImpl) Sync(correspondingMesh mesh.MeshProvider) error {
 
 	// Self can be nil if the node is removed
 	selfID := s.manager.GetPublicKey()
-	self, _ := correspondingMesh.GetNode(selfID.String())
+	self, err := correspondingMesh.GetNode(selfID.String())
 
-	// Mesh has been removed
-	if self == nil {
-		return fmt.Errorf("mesh %s does not exist", correspondingMesh.GetMeshId())
+	if err != nil {
+		logging.Log.WriteErrorf(err.Error())
 	}
 
 	correspondingMesh.Prune()
@@ -51,7 +51,8 @@ func (s *SyncerImpl) Sync(correspondingMesh mesh.MeshProvider) error {
 		logging.Log.WriteInfof("meshes %s has changes", correspondingMesh.GetMeshId())
 	}
 
-	if self.GetType() == conf.PEER_ROLE && !correspondingMesh.HasChanges() && s.infectionCount == 0 {
+	// If removed sync with other nodes to gossip the node is removed
+	if self != nil && self.GetType() == conf.PEER_ROLE && !correspondingMesh.HasChanges() && s.infectionCount == 0 {
 		logging.Log.WriteInfof("no changes for %s", correspondingMesh.GetMeshId())
 
 		// If not synchronised in certain time pull from random neighbour
@@ -63,16 +64,19 @@ func (s *SyncerImpl) Sync(correspondingMesh mesh.MeshProvider) error {
 	}
 
 	before := time.Now()
-	s.manager.GetRouteManager().UpdateRoutes()
+	err = s.manager.GetRouteManager().UpdateRoutes()
+
+	if err != nil {
+		logging.Log.WriteErrorf(err.Error())
+	}
 
 	publicKey := s.manager.GetPublicKey()
 	nodeNames := correspondingMesh.GetPeers()
 
-	if self != nil {
 	nodeNames = lib.Filter(nodeNames, func(s string) bool {
-			return s != mesh.NodeID(self)
+		// Filter our only public key out so we dont sync with ourself
+		return s != publicKey.String()
 	})
-	}
 
 	var gossipNodes []string
 
@@ -101,14 +105,14 @@ func (s *SyncerImpl) Sync(correspondingMesh mesh.MeshProvider) error {
 
 	// Do this synchronously to conserve bandwidth
 	for _, node := range gossipNodes {
-		correspondingPeer := s.manager.GetNode(correspondingMesh.GetMeshId(), node)
+		correspondingPeer, err := correspondingMesh.GetNode(node)
 
-		if correspondingPeer == nil {
+		if correspondingPeer == nil || err != nil {
 			logging.Log.WriteErrorf("node %s does not exist", node)
 			continue
 		}
 
-		err := s.requester.SyncMesh(correspondingMesh.GetMeshId(), correspondingPeer)
+		err = s.requester.SyncMesh(correspondingMesh.GetMeshId(), correspondingPeer)
 
 		if err == nil || err == io.EOF {
 			succeeded = true
@@ -131,7 +135,9 @@ func (s *SyncerImpl) Sync(correspondingMesh mesh.MeshProvider) error {
 
 	correspondingMesh.SaveChanges()
 
+	s.lock.Lock()
 	s.lastSync[correspondingMesh.GetMeshId()] = time.Now().Unix()
+	s.lock.Unlock()
 	return nil
 }
 
@@ -188,6 +194,7 @@ func (s *SyncerImpl) SyncMeshes() error {
 
 		go sync()
 	}
+	wg.Wait()
 
 	logging.Log.WriteInfof("updating the WireGuard configuration")
 	err := s.manager.ApplyConfig()