前言
1)akka框架是一个并发的、分布式的、可伸缩性的、高性能的RPC通信框架,大数据开发框架Spark、flink底层原理中或多或少都用到了
2)scala语言真的很强大、好用、方便,结合了面向对象语言和函数式语言的特点
akka的原理图
大多数分布式框架或工具 都遵循着主从节点的架构设计,在这里我们暂不考虑高可用的模式(高可用可参考文章Zookeeper之HDFS-HA高可用模式)
每个机器上的一个进程中只存在着1个通信角色对象 ActorSystem ,也就是说 ActorSystem 对象的示例只有一个,但由它创建的Master和Worker可以有多个,是多例
1)启动master 内部定时器定期检查有无超时连接(就是在一定时间内没有向我发送心跳的worker),并将失效的进行移除
2)启动worker,跟master建立网络连接,将自己的信息(workerid,内存,内核数cpu等信息)发给master进行注册
3)master收到注册信息,将注册的信息进行保存到内存(高效),也可以持久化到磁盘或zookeeper当中(数据安全),之后向worker发送注册成功的信息
4)worker收到master发来的注册成功的信息,很高兴,并启动定时器,定期发送心跳,向master报活 代码实现
Worker类代码:- import java.util.UUID
- import java.util.concurrent.TimeUnit
-
- import akka.actor.{Actor, ActorSelection, ActorSystem, Props}
- import com.typesafe.config.ConfigFactory
-
- import scala.concurrent.duration._
-
- /**
- * @author:tom
- * @Date:Created in 16:49 2020/12/18
- */
- class Worker extends Actor {
-
-
- var masterRef: ActorSelection = _
-
- var workerId = UUID.randomUUID().toString
-
- //在执行构造函数(实例化对象)之后、receive方法执行之前一定会执行一次
- override def preStart(): Unit = {
-
- //向master 进行注册信息
- //可以与master建立连接
- masterRef = context.actorSelection("akka.tcp://MasterActorSystem@localhost:8888/user/MasterActor")
- //发送消息
- masterRef ! RegisterWorker(workerId, "2048", 4)
- }
-
- override def receive: Receive = {
- //自己给自己发送的周期消息
- case SendHeartbeat => {
- // if () {
- //
- // } 向Master发送心跳
- masterRef ! HeartBeat(workerId)
- }
-
- case RegisteredWorker => {
- // println("a response from master")
-
- //启动一个定时器
- import context.dispatcher
- context.system.scheduler.schedule(Duration(0, TimeUnit.MILLISECONDS), 10000.millisecond, self, SendHeartbeat)
-
- }
- }
- }
-
- object Worker {
-
- def main(args: Array[String]): Unit = {
- val host = "localhost"
- val port = 9999
- val configStr =
- s"""
- |akka.actor.provider = "akka.remote.RemoteActorRefProvider"
- |akka.remote.netty.tcp.hostname = $host
- |akka.remote.netty.tcp.port = $port
- |""".stripMargin
- val config = ConfigFactory.parseString(configStr)
- //创建workerActorSystem
- val workerActorSystem = ActorSystem.apply("workerActorSystem", config)
- //创建workerActor
- val workerActor = workerActorSystem.actorOf(Props(new Worker), "WorkerActor")
- }
-
-
- }
复制代码 Master代码:- import akka.actor.{Actor, ActorSystem, Props}
- import com.typesafe.config.ConfigFactory
-
- import scala.collection.mutable
- import scala.concurrent.duration._
-
- /**
- * @author:tom
- * @Date:Created in 16:08 2020/12/18
- */
- class Master extends Actor {
-
- //定义一个可变的HashMap集合用来存储worker的信息
- val idToWorker = new mutable.HashMap[String, WorkerInfo]()
-
-
- //master定期检查自己 是否有新的节点(worker出现)
- override def preStart(): Unit = {
- import context.dispatcher
- context.system.scheduler.schedule(0 millisecond, 15000.millisecond, self, CheckTimeOutWorker)
- }
-
- //用来接收消息
- override def receive: Receive = {
-
- //模式匹配
- case "hello" => {
- println("hello~")
- }
- case "hi" => {
- println("hi~")
- }
-
- //定时检查
- case CheckTimeOutWorker => {
- val deadWorkers = idToWorker.values.filter(w => System.currentTimeMillis() - w.lastHeartbeatTime > 30000)
- deadWorkers.foreach(dw => {
- idToWorker -= dw.workerId
- })
- println(s"current alive worker size:${idToWorker.size}")
- }
-
- //有worker来进行注册信息需要执行的逻辑
- case RegisterWorker(workerId, memory, cores) => {
- // println(s"workerId:$workerId,memory:$memory,cores:$cores")
-
- //worker 注册成功应该执行的逻辑
-
- //将信息存入到内存集合当中
- val workerInfo: WorkerInfo = new WorkerInfo(workerId, memory, cores)
- idToWorker.put(workerId, workerInfo)
- //返回一个注册成功的信息
- sender() ! RegisteredWorker
- }
-
- //worker端发送过来的心跳信息
- case HeartBeat(workerId) => {
- //根据workerId到Map中查找对应的WorkerInfo
- if (idToWorker.contains(workerId)) {
- //如果存在 则取出信息
- val workerInfo = idToWorker(workerId)
- //更新上一次的心跳时间
- workerInfo.lastHeartbeatTime = System.currentTimeMillis()
- }
- }
-
- }
- }
-
- object Master {
-
- def main(args: Array[String]): Unit = {
- val host = "localhost"
- val port = 8888
- val configStr =
- s"""
- |akka.actor.provider = "akka.remote.RemoteActorRefProvider"
- |akka.remote.netty.tcp.hostname = $host
- |akka.remote.netty.tcp.port = $port
- |""".stripMargin
- val config = ConfigFactory.parseString(configStr)
- //创建一个ActorSystem实例(单例)
- val masterActorSystem = ActorSystem("MasterActorSystem", config)
- //创建一个Actor
- val actor = masterActorSystem.actorOf(Props[Master], "MasterActor")
- //自己给自己发消息
- actor ! "hello"
- }
-
-
- }
复制代码 更多学习、面试资料尽在微信公众号:Hadoop大数据开发 |