java actor akka_Akka的Actor模型及使用实例

本文的绝大部分内容转载自rerun.me这一blog，老外写的东西就是好啊。

ACTORS介绍

Anyone who has done multithreading in the past won't deny how hard and painful it is to manage multithreaded applications. I said manage because it starts out simple and it became a whole lot of fun once you start seeing performance improvements. However, it aches when you see that you don't have a easier way to recover from errors in your sub-tasks OR those zombie bugs that you find hard to reproduce OR when your profiler shows that your threads are spending a lot of time blocking wastefully before writing to a shared state.

I prefer not to talk about how Java concurrency API and their collections made it better and easier because I am sure if you are here, you probably needed more control over the sub-tasks or simply because you don't like to write locks and synchronized blocks and would prefer a higher level of abstraction.

In this series of Akka Notes, we would go through simple Akka examples to explore the various features that we have in the toolkit.

WHAT ARE ACTORS?

Treat Actors like People. People who don't talk to each other in person. They just talk through mails.

Let's expand on that a bit.

1.     MESSAGING

Consider two persons - A wise Teacher and Student. The Student sends a mail every morning to the Teacher and the wise Teacher sends a wise quote back.

Points to note :

The student sends a mail. Once sent, the mail couldn't be edited. Talk about natural immutability.

The Teacher checks his mailbox when he wishes to do so.

The Teacher also sends a mail back (immutable again).

The student checks the mailbox at his own time.

The student doesn't wait for the reply. (no blocking)

That pretty much sums up the basic block of the Actor Model - passing messages.

2. CONCURRENCY

Now, imagine there are 3 wise teachers and 3 students - every student sends notes to every other teacher. What happens then? Nothing changes actually. Everybody has their own mailbox. One subtle point to note here is this :

By default, Mails in the mailbox are read/processed in the order they arrived.

Internally, by default it is a ConcurrentLinkedQueue. And since nobody waits for the mail to be picked up, it is simply a non-blocking message. (There are a variety of built-in mailboxes including bounded and priority based. In fact, we could build one ourself too)

3. FAILOVER

Imagine these 3 teachers are from three different departments - History, Geography and Philosophy.

History teachers replies with a note on an Event in the past, Geography teachers sends an Interesting Place and Philosophy teachers, a quote. Each student sends message to each teacher and gets responses. The student doesnt care which teacher in the department sends the reply back. What if one day, a teacher falls sick? There has to be at least one teacher handling the mails from the department. In this case, another teacher in the department steps up and does the job.

Points to note :

There could be a pool of Actors who does different things.

An Actor could do something that causes an exception. It wouldn't be able to recover by itself. In which case a new Actor could be created in place of the old one. Alternatively, the Actor could just ignore that one particular message and proceed with the rest of the messages. These are called Directives and we'll discuss them later.

4. MULTITASKING

For a twist, let's assume that each of these teachers also send the exam score through mail too, if the student asks for it. Similarly, an the Actor could handle more than one type of message comfortably.

5. CHAINING

What if the student would like to get only one final consolidated trivia mail instead of three?

We could do that too with Actors too. We could chain the teachers as a hierarchy. We'll come back to that later when we talk about Supervisors and revisit the same thought when we talk about Futures.

As requested by Mohan, let's just try to map the analogy components with the the components in the Actor Model.

Students and the Teachers becomes our Actors. The Email Inbox becomes the Mailbox component. The request and the response can't be modified. They are immutable objects. Finally, the MessageDispatcher component in Actor manages the mailbox and routes the messages to the respective Mailbox.

ACTOR 消息机制

From the introductory first part of the Akka Notes, we saw a bird's eye view of Actors in the Akka Toolkit. In this second part of the Akka Notes, we'll look at the messaging part of Actors. As for the example, we would use the same Student-Teacher example that we discussed earlier.

In this first part of Actor Messaging, we'll create the Teacher Actor and instead of the Student Actor, we'll use a main program called StudentSimulatorApp.

REVISITING STUDENT-TEACHER IN DETAIL

Let's for now consider the message sent by the StudentSimulatorApp to the TeacherActor alone. When I say StudentSimulatorApp, I just mean a normal main program.

The picture conveys this :

(if the terms are overwhelming, don't worry, we'll go through them in detail)

Student creates something called an ActorSystem

It uses the ActorSystem to create something called as ActorRef. The QuoteRequest message is sent to the ActorRef (a proxy to TeacherActor)

Actor ref passes the message along to a Dispatcher

The Dispatcher enqueues the message in the target Actor's MailBox.

The Dispatcher then puts the Mailbox on a Thread (more on that in the next section).

The MailBox dequeues a message and eventually delegates that to the actual Teacher Actor's receive method.

Like I said, don't worry about it. Let's look at each step in detail now. You can come back and revisit these five steps once we are done.

THE STUDENTSIMULATORAPP PROGRAM

We would use this StudentSimulatorApp to bring up the JVM and initialize the ActorSystem.

As we understand from the picture, the StudentSimulatorApp

Creates an ActorSystem

Uses the ActorSystem to create a proxy to the Teacher Actor (ActorRef)

Sends the QuoteRequest message to the proxy.

Let's focus on these three points alone now.

1.Creating an ActorSystem

ActorSystem is the entry point into the ActorWorld. ActorSystems are through which you could create and stop Actors. Or even shutdown the entire Actor environment.

On the other end of the spectrum, Actors are hierarchical and the ActorSystem is also similar to the java.lang.Object or scala.Any for all Actors - meaning, it is the root for all Actors. When you create an Actor using the ActorSystem's actorOf method, you create an Actor just below the ActorSystem.

The code for initializing the ActorSystem looks like

val system=ActorSystem("UniversityMessageSystem")

The UniversityMessageSystem is simply a cute name you give to your ActorSystem.

2.Creating a Proxy for TeacherActor?

Let's consider the following snippet :

val teacherActorRef:ActorRef=actorSystem.actorOf(Props[TeacherActor])

The actorOf is the Actor creation method in ActorSystem. But, as you can see, it doesn't return a TeacherActor which we need. It returns something of type ActorRef.

The ActorRef acts as a Proxy for the actual Actors. The clients do not talk directly with the Actor. This is Actor Model's way of avoiding direct access to any custom/private methods or variables in the TeacherActor or any Actor for that sake.

To repeat, you send messages only to the ActorRef and it eventually reaches your actual Actor. You can NEVER talk to your Actor directly. People will hate you to death if you find some mean ways to do that.

3.Send aQuoteRequestto the Proxy

It's an one liner again. You just tell the QuoteRequest message to the ActorRef. The tell method in Actor is actually !. (there's also a tell method in ActorRef which just delegates the call back to !)

//send a message to the Teacher Actor

teacherActorRef!QuoteRequest

That's it !!!

If you think I am lying, check the entire code of the StudentSimulatorApp below :

STUDENTSIMULATORAPP.SCALA

package me.rerun.akkanotes.messaging.actormsg1

import akka.actor.ActorSystem

import akka.actor.Props

import akka.actor.actorRef2Scala

import me.rerun.akkanotes.messaging.protocols.TeacherProtocol._

object StudentSimulatorApp extends App{

//Initialize the ActorSystem

val actorSystem=ActorSystem("UniversityMessageSystem")

//construct the Teacher Actor Ref

val teacherActorRef=actorSystem.actorOf(Props[TeacherActor])

//send a message to the Teacher Actor

teacherActorRef!QuoteRequest

//Let's wait for a couple of seconds before we shut down the system

Thread.sleep (2000)

//Shut down the ActorSystem.

actorSystem.shutdown()

}

Well, I cheated a little. You'll have to shutdown the ActorSystem or otherwise, the JVM keeps running forever. And I am making the main thread sleep for a little while just to give the TeacherActor to finish off its task. I know this sounds stupid. Don't worry about it. We'll write some neat testcases in the next part in order to avoid this hack.

THE MESSAGE

We just told a QuoteRequest to the ActorRef but we didn't see the message class at all !!

Here it comes :

(It is a recommended practice to wrap your messages in a nice object for easier organization)

TeacherProtocol

package me.rerun.akkanotes.messaging.protocols

object TeacherProtocol{

case class QuoteRequest()

case class QuoteResponse(quoteString:String)

}

As you know, the QuoteRequest is for the requests that come to the TeacherActor. The Actor would respond back with a QuoteResponse.

DISPATCHER AND A MAILBOX

The ActorRef delegates the message handling functionality to the Dispatcher. Under the hood, while we created the ActorSystem and the ActorRef, aDispatcher and a MailBox was created. Let's see what they are about.

MailBox

Ever Actor has one MailBox (we'll see one special case later). Per our analogy, every Teacher has one mailbox too. The Teacher has to check the mailbox and process the message. In Actor world, it's the other way round - the mailbox, when it gets a chance uses the Actor to accomplish its work.

Also the mailbox has a queue to store and process the messages in a FIFO fashion - a little different from our regular inbox where the most latest is the one at the top.

Now, the dispatcher

Dispatcher does some really cool stuff. From the looks of it, the Dispatcher just gets the message from the ActorRef and passes it on to the MailBox. But there's one amazing thing happening behind the scenes :

Dispatcher does some really cool stuff. From the looks of it, the Dispatcher just gets the message from the ActorRef and passes it on to the MailBox. But there's one amazing thing happening behind the scenes :

The Dispatcher wraps an ExecutorService (ForkJoinPool or ThreadPoolExecutor). It executes the MailBox against this ExecutorService.

Check out this snippet from the Dispatcher

protected[akka] override def registerForExecution(mbox: Mailbox, ...): Boolean = {

...

try {

executorService execute mbox

...

}

示例代码

这是一套用于计算熵增益的代码，用在相似度检测和内容推荐领域，主要思想是通过Master Actor派发计算任务给Worker Actor完成计算任务后，发消息通知Master完成计算结果的综合工作，最后结果的展示由listener Actor完成。

EntropyMain

packagecom.emcc.recommand.entorpy;importorg.apache.log4j.Logger;importakka.actor.ActorRef;importakka.actor.ActorSystem;importakka.actor.Props;importcom.emcc.recommand.entorpy.actor.Listener;importcom.emcc.recommand.entorpy.actor.Master;importcom.emcc.recommand.entorpy.message.Calculate;public classEntropyMain

{private static Logger logger = Logger.getLogger(EntropyMain.class);/*** 启动的工作线程数*/

private static final int WORKER_COUNT = 10;public static voidmain(String[] args)

{

logger.info("begin to execute entorpy for recommand");

calculate(WORKER_COUNT);

logger.info("execute entorpy for recommand finished.");

}/*** 工作线程*/

private static void calculate(intnumOfWorkers)

{//创建Akka系统

ActorSystem system = ActorSystem.create("Entropy-calc");//结果监听器，用于结果输出

ActorRef listener = system.actorOf(Props.create(Listener.class),"listener");//启动actor，驱动计算开始

ActorRef master =system.actorOf(

Props.create(Master.class, numOfWorkers, listener), "master");//发送启动消息

master.tell(newCalculate(), master);

}

}

Master

packagecom.emcc.recommand.entorpy.actor;importjava.util.ArrayList;importjava.util.Collections;importjava.util.List;importjava.util.concurrent.TimeUnit;importorg.apache.commons.collections.CollectionUtils;importorg.apache.log4j.Logger;importscala.concurrent.duration.Duration;importakka.actor.ActorRef;importakka.actor.Props;importakka.actor.UntypedActor;importakka.routing.RoundRobinRouter;importcom.emcc.recommand.entorpy.message.Calculate;importcom.emcc.recommand.entorpy.message.Result;importcom.emcc.recommand.entorpy.message.SortedEntropy;importcom.emcc.recommand.entorpy.message.Work;importcom.emcc.recommand.entorpy.pojo.Entropy;importcom.emcc.recommand.entorpy.service.ProbabilityCalcService;importcom.emcc.recommand.entorpy.service.impl.ProbabilityCalcServiceImpl;

@SuppressWarnings("deprecation")public class Master extendsUntypedActor

{private static Logger logger = Logger.getLogger(Master.class);/*** 返回的结果数量，用于判断计算是否完成*/

private intnumOfResults;/*** 用于将结果传递到外部*/

privateActorRef listener;/*** 执行器*/

privateActorRef workerRouter;/*** 内容key*/

private List contentKeys = null;/*** 特征key*/

private List itemKeys = null;/*** 中间结果集合*/

private List middleResult = new ArrayList();/*** 计算开始时间*/

private final long start =System.currentTimeMillis();/*** 计算服务*/

private ProbabilityCalcService pcService = newProbabilityCalcServiceImpl();/***

*@paramnrOfWorkers

* 表示要启动多少worker

*@paramnrOfElements

* 分配给每个worker的分块数量

*@paramlistener*/

public Master(intnrOfWorkers, ActorRef listener)

{this.contentKeys =pcService.getCntCodes();this.itemKeys =pcService.getItems();this.listener =listener;

workerRouter= this.getContext().actorOf(

Props.create(Worker.class).withRouter(newRoundRobinRouter(nrOfWorkers)));

}

@Overridepublic void onReceive(Object message) throwsException

{if (message instanceofCalculate)

{

logger.info("receive calculate messsage.");if(CollectionUtils.isNotEmpty(itemKeys)&&CollectionUtils.isNotEmpty(contentKeys))

{long count = 0L;for(String item : itemKeys)

{

count++;

logger.info("send message to worker, item=" + item + ", "

+ count + "/" +itemKeys.size());

workerRouter.tell(newWork(item, contentKeys), getSelf());

}

}else{

logger.error("stop process, no data to calculate.");//停止整个系统

getContext().system().shutdown();

}

}else if (message instanceofResult)

{

Result result=(Result) message;

numOfResults+= 1;

logger.info("receive result messsage from worker, item="

+ result.getItem() + ", " + numOfResults + "/"

+itemKeys.size());//内存存入缓存

middleResult.add(newEntropy(result.getItem(), result.getValue()));//计算完毕

if (numOfResults ==itemKeys.size())

{

logger.info("calculate finished, begin to sort the result.");//发送消息到listener

Duration duration =Duration.create(System.currentTimeMillis()-start, TimeUnit.MILLISECONDS);//对输出结果进行派讯

Collections.sort(middleResult);

logger.info("send SortedEntropy message to listener.");

listener.tell(newSortedEntropy(middleResult, duration),

getSelf());//停止master以及其控制的actor

getContext().stop(getSelf());

}

}else{

unhandled(message);

}

}

}

Worker

packagecom.emcc.recommand.entorpy.actor;importjava.math.BigDecimal;importjava.util.List;importorg.apache.log4j.Logger;importakka.actor.UntypedActor;importcom.emcc.recommand.entorpy.message.Result;importcom.emcc.recommand.entorpy.message.Work;importcom.emcc.recommand.entorpy.service.ProbabilityCalcService;importcom.emcc.recommand.entorpy.service.impl.ProbabilityCalcServiceImpl;importcom.emcc.recommand.entorpy.util.Logarithm;/*** 计算某个item的熵增益*/

public class Worker extendsUntypedActor

{private static Logger logger = Logger.getLogger(Worker.class);/*** 计算结果保留位数*/

private static final int DECIMAL_SIZE = 4;private ProbabilityCalcService pcService = newProbabilityCalcServiceImpl();

@Overridepublic void onReceive(Object message) throwsException

{if (message instanceofWork)

{

Work work=(Work) message;

logger.info("receive work message from master, item="

+work.getItem());double result =calculateEntropyForItem(work.getItem(),

work.getCntKeyList());

logger.info("send result message to master, item=" +work.getItem()+ ", value=" +result);

getSender().tell(newResult(result, work.getItem()), getSelf());

}else{

unhandled(message);

}

}/*** 计算对某个item的熵增益

*

*@paramitem

* 要计算的短语名称

*@paramcntKeys

* 系统内全部的cnt

*@return熵增益*/

private double calculateEntropyForItem(String item, ListcntCodes)

{

Double entropyIncr= 0.0;//中间计算逻辑，计算出对某个item的熵增益并返回

if (cntCodes != null && !cntCodes.isEmpty())

{

Double x=pcService.calcCntProbability();

Double y=pcService.calcDocProbability(item);for(String cntCode : cntCodes)

{

Double z=pcService.calcCndProbability(cntCode, item);if(isParamLegal(x, y, z))

{

Double temp=calculateByFormular(x, y, z);if (temp != null && !temp.isInfinite() && !temp.isNaN())

{

entropyIncr= entropyIncr +temp;

}else{

logger.error("calculate ectropy error, item=" +item+ ", cntCode=" + cntCode + ", x=" + x + ",y="

+ y + ", z=" +z);

}

}

}

}returnformat(entropyIncr, DECIMAL_SIZE);

}/*** 计算值*/

private doublecalculateByFormular(Double x, Double y, Double z)

{

Double a1= -x *Logarithm.log2(x);

Double a2= z * x * Logarithm.log2((z * x) /y);

Double a3= (1 - z) * x * Logarithm.log2(((1 - z) * x) / (1 -y));return a1 + a2 +a3;

}private booleanisParamLegal(Double x, Double y, Double z)

{boolean res = false;if (x != null && y != null && z != null)

{

res= true;

}else{

logger.error("parameter error, cancel this round, x=" + x + ", y="

+ y + ", z=" +z);

}returnres;

}/*** 计算结果完整的返回用于调整算法*/

private Double format(Double org, intdecimalSize)

{try{

BigDecimal b= newBigDecimal(org);

Double f=b.setScale(decimalSize, BigDecimal.ROUND_HALF_UP)

.doubleValue();returnf;

}catch(Exception e)

{//TODO Auto-generated catch block

e.printStackTrace();

}return 0.0;

}

}

Listener

packagecom.emcc.recommand.entorpy.actor;importjava.text.DateFormat;importjava.text.SimpleDateFormat;importjava.util.Date;importorg.apache.log4j.Logger;importakka.actor.UntypedActor;importcom.emcc.recommand.entorpy.message.SortedEntropy;importcom.emcc.recommand.entorpy.util.FileUtil;/*** 结果输出actor*/

public class Listener extendsUntypedActor

{private static Logger logger = Logger.getLogger(Listener.class);

@Overridepublic void onReceive(Object message) throwsException

{if (message instanceofSortedEntropy)

{

SortedEntropy sortedEntropy=(SortedEntropy) message;//结果输出至文件系统

FileUtil.list2File(getOutputPath(), sortedEntropy.getEntorpyList());

logger.info("output resultfile finished, takes "

+ sortedEntropy.getDuration().toSeconds() + "s.");

getContext().system().shutdown();

}else{

unhandled(message);

}

}privateString getOutputPath()

{

DateFormat df= new SimpleDateFormat("yyyyMMddHHmmss");

String date= df.format(newDate());return "./entropy_" + date + ".out";

}

}

使用总结

Akka确实可以简化多线程编程，开发人员可以从繁复的多线程开发中解脱出来，更加专注于业务实现，这种抽象适合绝大多数应用。