Cadre de communication cloud–edge de KubeEdge

//Context is object for Context channel
type Context struct{
	//ConfigFactory goarchaius.ConfigurationFactory
	channels map[string]chan model.Message
	chsLock  sync.RwMutex
	typeChannels map[string]map[string]chan model.Message
	typeChsLock  sync.RWMutex
	anonChannels map[string]chan model.Message
	anonChsLock  sync.RWMutex
}

//Module Interface
type Moudule interface{
	Name() string
	Group() string
	Start()
	Enable() bool
}

	type ModuleContext interface {
	AddModule(info *common.ModuleInfo)
	AddModuleGroup(module.group string)
	Cleanup(module string)
}

//MessageContext is interface for messaging syncing
type MessageContext interface {
	//async mode
	Send(module string, message model.Message)
	Receive(module string)(model.Message error)
	//sync mode
	SendSync(module string,message model.Message,timeout time.Duration)(model.Message error)
	SendResp(message model.Message)
	//group broadcast
	SendToGroup(group string,message model.Message)
	SendToGroupSync(group string,message model.Message,timeout time.Duration) error
}

//registerModules register all the modules started in cloudcore
func registerModules(c *v1alpha1.CloudCoreConfig){
	cloudhub.Register(c.Modules.Cloudhub)
	edgecontroller.Register(c.Modules.EdgeController)
	devicecontroller.Register(c.Modules.DeviceController)
	nodeupgradejobcontroller.Register(c.Modules.NodeUpgradeJobController)
	synccontroller.Register(c.Modules.SyncController)
	cloudstream.Register(c.Modules.CloudStream,c.CommonConfig)
	router.Register(c.Modules.Router)
	dynamiccontroller.Register(c.Modules.DynamicController)
	policycontroller.Register(client.CrdConfig)
}

//StartModules starts modules that are registered
func StartModules(){
	//only register channel mode,if want to use socket mode,we should also pass in common.MsgCtxTypeUSparameter
	beehiveContext.InitContext([]string{common.MsgCtxTypeChannel})

	modules := GetModules()

	for name, module := range modules{
		var m common.ModuleInfo
		switch module.contextType{
		case common.MsgCtxTypeChannel:
			m = common.ModuleInfo{
				ModuleName: name,
				ModuleType: module.contextType,
			}
		......

		default:
			klog.Exitf("unsupported context type: %s",module.contextType)
		}

		beehiveContext.AddModule(&m)
		beehiveContext.AddModuleGroup(name,module.module.Group())

		go moduleKeeper(name, module, m)
		klog.Infof("starting module %s",name)
	}
}

//the operation set of connection
type Connection interface{
	//process message from the connection
	ServeConn() //服务端从通道中持续的读取消息

	//SetReadDeadline sets the deadline for future Read calls
	//and any currently-blocked Read call.
	//A zero value for t means Read will not time out.
	SetReadDeadline(t time.Time) error

	//SetWriteDeadline sets the deadline for future write calls
	//and any currently-blocked write call.
	//Even if write times out.It may return n > 0, indicating that
	//some of the data was successfully written.
	//A zero value for t means write will not time out.
	SetWriteDeadline(t time.Time) error

	//write write raw data to the connection
	//it will open a stream for raw data
	write(raw []byte)(int,error)

	//writeMessageAsync writes data to the connection and don't care about the response
	WriteMessageAsync(msg *model.Message) error //异步

	//writeMessageSync writes data to the connection and care about the response
	WriteMessageSync(msg *model.Message)(*model.Message,error) //同步

	//ReadMessage reads message from the connection
	//it will be blocked when no message received
	//if you want to use this api for message reading
	//make sure AutoRoute be false
	ReadMessage(msg *model.Message ) error

	//RemoteAddr returns the remote network address
	RemoteAddr() net.Addr

	//LocalAddr returns the local network address
	LocalAddr() net.Addr

	//connectState return the current connection state
	ConnectionState() connectionState

	//Close closes the connection
	//Any blocked Read or Write operations will be unblocked and return errors
	Close() error
}

//protocol server
type ProtocolServer interface {
	ListenAndServerTLS() error
	close() error
}

func (srv *WSServer) ListenAndServeTLS() error{
		return srv.server.ListenAndServeTLS("","")
}
func (srv *WSServer)Close() error{
	if srv.server != nil{
		return srv.server.Close()
	}
	return nil
}

//each protocol(websocket/quic) provides Connect
type Protocolclient interface{
	Connect() (conn.Connection,error)
}

//Connect try to connect remote server
func(c *WSClient)Connect()(conn.Connection,error){
	header := c.exOpts.Header
	header.Add("ConnectionUse",string(c.options.ConnUse))
	wsConn,resp,err := c.dialer.Dial(c.options.Addr,header)
	if err ==nil{
		klog.Infof("dial %s successfully",c.options.Addr)

		//do user's processing on connection or response
		if c.exOpts.callback != nil{
			c.exOpts.Callback(wsConn,resp)
		}
		return conn.NewConnection(&conn.ConnectionOptions{
			ConnType: api.ProtocolTypeWS,
			ConnUse:  c.options.ConnUse,
			Base:     wsConn,
			Consumer: c.options.Consumer,
			Handler:  c.options.Handler,
			CtrlLane: lane.NewLane(api.ProtocolTypeWS,wsConn),
			State:    &conn.ConnectionState{
				State:  api.StatConnected,
				Headers:c.exOpts.Header.Clone(),
			},
			AutoRoute: c.options.AutoRoute,
		}),nil
	}

	//something wrong!
	var respMsg string
	if resp != nil{
		body, errRead := io.ReadAll(io.LimitReader(resp.Body,comm.MaxReadLength))
		if errRead ==nil{
			respMsg = fmt.Sprintf("response code: %d,response body: %s",resp.StatusCode,string(body))
		}else{
			respMsg = fmt.Sprintf("response code: %d",resp.StatusCode)
		}
		resp.Body.Close()
	}
	klog.Errorf("dial websocket error(%+v), response message: %s",err, respMsg)

	return nil,err
}

cloudhub.Register(c.modules.Cloudhub)

func (ch *cloudHub) Start(){
	if !cache.WaitForCacheSync(beehiveContext.Done(),ch.informersSyncedFuncs...
	{
		klog.Errorf("unable to sync caches for objectSyncController")
		os.Exit(1)
	}

	//start dispatch message from the cloud to edge node
	go ch.dispatcher.DispatchDownstream()

	//check whether the certificates exists in the local directory.
	//and then check whether certificates exists in the secret.
	//generate if they don't exist
	if err := httpserver.PrepareAllCerts(): err!=nil{
		klog.Exit(err)
	}

	DoneTLSTunnelCerts <- true
	close(DoneTLSTunnelCerts)

	//generate Token
	if err:=httpserver.GenerateToken():err!=nil{
		klog.Exit(err)
	}

	//HttpServer mainly used to issue certificates for the edge
	go httpserver.StartHTTPServer()

	servers.StartCloudHub(ch.messageHandler)

	if hubconfig.Config.UnixSocket.Enable{
	//The uds server is only used to communicate with csi driver from kubeedge on cloud
	//It is not used to communicate  between cloud and edge
	go udsserver.StartServer(hubconfig.Config.unixSocket.Address)
	}
}

type Handler interface{
	//HandleConnection is invoked when a new connection arrives
	HandleConnection(connection conn.Connection)

	//HandleMessage is invoked when a new message arrives
	HandleMessage(container *mux.MessageContainer,writer mux.ResponseWriter)

	//OnEdgeNodeConnect is invoked when a new connection is established
	OnEdgeNodeConnect(info *model.HubInfo,connection conn.Connection) error
	//OnEdgeNodeDisconnect is invoked when a connection is lost
	OnEdgeNodeDisconnect(info *model.HubInfo,connection conn.Connection)

	//OnReadTransportErr is invoked when the connection read message err

}

nodeID := connection.ConnectionState().Headers.Get("node_id")
projectID := connection.ConnectionState().Headers.Get("project_id")

if mh.SessionManager.ReachLimit(){
	klog.Errorf("Fail to serve node %s,reach node limit",nodeID)
	return
}

//init node message pool and add to the dispatcher
nodeMessagePool := common.InitNodeMessagePool(nodeID)
mh.MessageDispatcher.AddNodeMessagePool(nodeID,nodeMessagePool)

//NodeMessagePool is a collection of all downstream message sent to an
//edge node.There are two types of messages,one that requires an ack
//and another that does not.For each type of message.we use the 'queue'
//to mark the order of sending, and use the 'store' to store specific messages
type NodeMessagePool struct{
	//AckMessageStore store message that will send to edge node
	//and require acknowledgement from edge node
	AckMessageStore cache.Store
	//AckMessageQueue store message key that will send to edge node
	//and require acknowledgement from edge node
	AckMessageQueue workqueue.RateLimitingInterface
	//NoAckMessageStore store message that will send to edge node
	//and do not require acknowledgement from edge node
	NoAckMessageStore cache.Store
	//NoAckMessageQueue store message key that will send to edge node
	//and do not require acknowledgement from edge node
	NoAckMessageQueue workqueue.RateLimitingInterface
}

//create a node session for each edge node
nodeSession := session.NewNodeSession(nodeID,projectID,connection,
	keepaliveInterval,nodeMessagePool,mh.reliableClient)
//add node session  to the session manager
mh.SessionManager.AddSession(nodeSession)

//Start session for each edge node and it will keep running until
//it encounters some Transport Error from underlying connection.
nodeSession.Start()

//Start the main goroutine responsible for serving node session
func (ns *NodeSession)Start(){
	klog.Infof("Start session for edge node %s",ns.nodeID)

	go ns.KeepAliveCheck()
	go ns.SendAckMessage()
	go ns.SendNoAckMessage()

	<-ns.ctx.Done()
}

//HandleMessage handle all the request from node
func (mh *messageHandler)HandleMessage(coantainer *mux.MessageContainer,writer mux.ResponseWriter){
	nodeID := container.Header.Get("node_id")
	projectID := container.Header.Get("project_id")

	//validate message
	if container.Message == nil{
		klog.Errorf("The message is nil for node: %s",nodeID)
		return
	}

	klog.v(4).Infof("[messageHandler]get msg from node(%s): %+v",nodeID,container.Message)

	//dispatch upstream message
	mh.MessageDispatcher.DispatchUpstream(container.Message,&model.HubInfo{ProjectID: projectID,NodeID:nodeID})
}

func (ns *NodeSession)SendMessageWithRetry(copyMsg, msg *beehivemodel.Message)error{
	ackChan := make(chan struct{})
	ns.ackMessageCache.Store(copyMsg.GetID(),ackChan)

	//initialize retry count and timer for sending message
	retryCount := 0
	ticker := time.NewTimer(sendRetryInterval)

	err := ns.connection.WriteMessageAsync(copyMsg)
	if err !=nil{
		return err
	}

	for{
		select{
		case <- ackChan:
			ns.saveSuccessPoint(msg)
			return nil

		case <- ticker.C:
			if retryCount == 4{
				return ErrwaitTimeout
			}
			err := ns.connection.WriteMessageAsync(copyMsg)
			if err !=nil{
				return err
			}
			retryCount++
			ticker.Reset(sendRetryInterval)
		}
	}
}

synccontroller.Register(c.Modules.syncController)

func (sctl *SyncController) Start(){
	if !cache.WaitForCacheSync(beehiveContext.Done(),sctl.informersSyncedFuncs...){
		klog.Errorf("unable to sync caches for sync controller")
		return
	}

	sctl.deleteObjectSyncs() //check outdate sync before start to reconcile
	go wait.Until(sctl.reconcile, 5*time.Second,beehiveContext.Done())
}

func Register(eh *v1alpha2.EdgeHub,nodeName string){
	config.InitConfigure(eh,nodeName)
	core.Register(newEdgeHub(eh.Enable))
}

func (eh *EdgeHub)Start(){
	eh.certManager = certificate.NewCertManager(config.config.EdgeHub,config.config.NodeName)
	eh.certManager.Start()
	for _, v := range GetCertSyncChannel(){
		v <- true
		close(v)
	}

	go eh.ifRotationDone()

	for{
		select{
			case <- beehiveContext.Done():
			klog.warning("EdgeHub stop")
			return
		default:
		}
		err := eh.initial()
		if err != nil{
			klog.Exitf("failed to init controller:%v",err)
			return
		}
		waitTime := time.Duration(config.Config.Heartbeat)*time.Second*2

		err = eh.chClient.Init()
		if err!= nil{
			klog.Errorf("connection failed: %v,will reconnect after %s",err,waitTime.String())
			time.Sleep(waitTime)
			continue
		}
		//execute hook func after connect
		eh.pubConnectInfo(true)
		go eh.routeToEdge()
		go eh.routeToCloud()
		go eh.keepalive()

		//wait the stop signal
		//stop authinfo manager/websocket connection
		<-eh.reconnectChan
		eh.chClient.UnInit()

		//execute hook function after disconnect
		eh.pubConnectInfo(false)

		//sleep one period of heartbeat, then try to connect cloud hub again
		klog.warningf("connection is broken, will reconnect after %s",waitTime.String())
		time.Sleep(waitTime)

	//clean channel
	clean:
		for{
			select{
			case <- eh.reconnectChan:
			default:
				break clean
			}
		}
	}
}

func (eh *EdgeHub)routeToEdge(){
	for{
		select{
		case<-beehiveContext.Done():
			klog.Warning("EdgeHub RouteToEdge stop")
			return
		default:
		}
		message, err := eh.chClient.Receive()
		if err!=nil{
			klog.Errorf("failed to dispatch message,discard: %v",err)
		}
	}
}

func (eh *EdgeHub)routeToCloud(){
	for{
		select{
		case<-beehiveContext.Done():
			klog.warning("EdgeHub RouteToCloud stop")
			return
		default:
		}
		message,err:= beehiveContext.Receive(modules.EdgeHubModuleName)
		if err !=nil{
			klog.Errorf("failed to receive message from edge: %v",err)
			time.Sleep(time.Second)
			continue
		}

		err = eh.tryThrottle(message.GetID())
		if err !=nil{
			klog.Errorf("msgID: %s,client rate limiter returned an error: %v",message.GetID(),err)
			continue
		}

		//post message to cloud hub
		err = eh.sendToCloud(message)
		if err !=nil{
			klog.Errorf("failed to send message to cloud: %v",err)
			eh.reconnectChan <- struct{}{}
			return
		}
	}
}

func (eh *EdgeHub)keepalive(){
	for{
		select{
		case <-beehiveContext.Done():
			klog.warning("EdgeHub KeepAlive stop")
			return
		default:
		}
		msg := model.NewMessage("").
			BuildRouter(modules.EdgeHubModuleName,"resource","node",messagepkg.OperationKeepalive)
			FillBody("ping")

		//post message to cloud hub
		err := eh.sendToCloud(*msg)
		if err != nil{
			klog.Errorf("websocket write error: %v",err)
			eh.reconnectChan <- struct{}{}
			return
		}
		time.Sleep(time.Duration(config.Config.Heartbeat)*time.Second)
	}
}