Merge pull request #515 from laijs/watchHyperstart-cancellable

make watchHyperstart() cancellable

Author: laijs

hypervisor/context.go

@@ -33,8 +33,6 @@ type VmContext struct {
 	PauseState int
 	Boot       *BootConfig
 
-	vmHyperstartAPIVersion uint32
-
 	// Communication Context
 	Hub    chan VmEvent
 	client chan *types.VmResponse
@@ -68,6 +66,8 @@ type VmContext struct {
 	current string
 	timer   *time.Timer
 
+	cancelWatchHyperstart chan struct{}
+
 	logPrefix string
 
 	lock      sync.RWMutex //protect update of context
@@ -147,6 +147,8 @@ func InitContext(id string, hub chan VmEvent, client chan *types.VmResponse, dc
 		networks:        NewNetworkContext(),
 		vmExec:          make(map[string]*hyperstartapi.ExecCommand),
 		logPrefix:       fmt.Sprintf("SB[%s] ", id),
+
+		cancelWatchHyperstart: make(chan struct{}),
 	}
 	ctx.networks.sandbox = ctx
 
@@ -241,10 +243,14 @@ func (ctx *VmContext) Close() {
 		ctx.Log(INFO, "VmContext Close()")
 		ctx.lock.Lock()
 		defer ctx.lock.Unlock()
+		select {
+		case ctx.cancelWatchHyperstart <- struct{}{}:
+		default:
+		}
+		ctx.hyperstart.Close()
 		ctx.unsetTimeout()
 		ctx.networks.close()
 		ctx.DCtx.Close()
-		ctx.hyperstart.Close()
 		close(ctx.client)
 		os.Remove(ctx.ShareDir)
 		ctx.handler = nil

hypervisor/hypervisor.go

@@ -32,14 +32,14 @@ func (ctx *VmContext) loop() {
 }
 
 func (ctx *VmContext) watchHyperstart() {
+	next := time.NewTimer(10 * time.Second)
 	timeout := time.AfterFunc(60*time.Second, func() {
-		if ctx.PauseState == PauseStateUnpaused {
-			ctx.Log(ERROR, "watch hyperstart timeout")
-			ctx.Hub <- &InitFailedEvent{Reason: "watch hyperstart timeout"}
-			ctx.hyperstart.Close()
-		}
+		ctx.Log(ERROR, "watch hyperstart timeout")
+		ctx.Hub <- &InitFailedEvent{Reason: "watch hyperstart timeout"}
+		ctx.hyperstart.Close()
 	})
 	ctx.Log(DEBUG, "watch hyperstart")
+loop:
 	for {
 		ctx.Log(TRACE, "issue VERSION request for keep-alive test")
 		_, err := ctx.hyperstart.APIVersion()
@@ -52,9 +52,15 @@ func (ctx *VmContext) watchHyperstart() {
 		if !timeout.Stop() {
 			<-timeout.C
 		}
-		time.Sleep(10 * time.Second)
+		select {
+		case <-ctx.cancelWatchHyperstart:
+			break loop
+		case <-next.C:
+		}
+		next.Reset(10 * time.Second)
 		timeout.Reset(60 * time.Second)
 	}
+	next.Stop()
 	timeout.Stop()
 }
 
@@ -107,7 +113,9 @@ func VmAssociate(vmId string, hub chan VmEvent, client chan *types.VmResponse, p
 
 	context.Become(stateRunning, StateRunning)
 
-	go context.watchHyperstart()
+	if !paused {
+		go context.watchHyperstart()
+	}
 	go context.loop()
 	return context, nil
 }

hypervisor/vm.go

@@ -652,6 +652,7 @@ func (vm *Vm) Pause(pause bool) error {
 	if ctx.PauseState != pauseState {
 		/* FIXME: only support pause whole vm now */
 		if pause {
+			ctx.cancelWatchHyperstart <- struct{}{}
 			err = ctx.hyperstart.PauseSync()
 		}
 		if err != nil {
@@ -668,6 +669,7 @@ func (vm *Vm) Pause(pause bool) error {
 
 		if !pause {
 			err = ctx.hyperstart.Unpause()
+			go ctx.watchHyperstart()
 		}
 		if err != nil {
 			vm.Log(ERROR, "%s sandbox failed: %v", command, err)