Thread(執行緒)

Program,Process,Thread

在介紹Thread之前,我們必須先把Program和Process這兩個觀念作一個釐清。

Program:一群程式碼的集合,用以解決特定的問題。以物件導向的觀念來類比,相當於Class。
Process:由Program所產生的執行個體,一個Program可以同時執行多次,產生多個Process。以物件導向的觀念來類比,相當於Object。每一個Process又由以下兩個東西組成
- 一個Memory Space。相當於Object的variable,不同Process的Memory Space也不同,彼此看不到對方的Memory Space。
- 一個以上的Thread。Thread代表從某個起始點開始(例如main),到目前為止所有函數的呼叫路徑,以及這些呼叫路徑上所用到的區域變數。當然程式的執行狀態,除了紀錄在主記憶體外,CPU內部的暫存器(如Program Counter, Stack Pointer, Program Status Word等)也需要一起紀錄。所以Thread又由下面兩項組成
  - Stack:紀錄函數呼叫路徑,以及這些函數所用到的區域變數
  - 目前CPU的狀態

由上面的描述中,我們在歸納Thread的重點如下

一個Process可以有多個Thread。
同一個Process內的Thread使用相同的Memory Space,但這些Thread各自擁有其Stack。換句話說,Thread能透過reference存取到相同的Object,但是local variable卻是各自獨立的。
作業系統會根據Thread的優先權以及已經用掉的CPU時間,在不同的Thread作切換,以讓各個Thread都有機會執行。

如何產生Thread

Java以java.lang.Thread這個類別來表示Thread。Class Thread有兩個Constructor:

Thread()
Thread(Runnable)

第一個Constrctor沒有參數,第二個需要一個Runnable物件當參數。Runnable是一個interface,定義於java.lang內,其宣告為

public interface Runnable {
    public void run();
}

使用Thread()產生的Thread,其進入點為Thread裡的run();使用Thread(Runnable)產生的Thread,其進入點為Runnable物件裡的run()。當run()結束時,這個Thread也就結束了;這和main()結束有相同的效果。其用法以下面範例說明:

public class ThreadExample1 extends Thread {
    public void run() { // override Thread's run()
        System.out.println("Here is the starting point of Thread.");
        for (;;) { // infinite loop to print message
            System.out.println("User Created Thread");
        }
    }
    public static void main(String[] argv) {
        Thread t = new ThreadExample1(); // 產生Thread物件
        t.start(); // 開始執行t.run()
        for (;;) {
            System.out.println("Main Thread");
        }
    }
}

以上程式執行後,螢幕上會持續印出"User Created Thread"或"Main Thread"的字樣。利用Runnable的寫法如下

public class ThreadExample2 implements Runnable {
    public void run() { // implements Runnable run()
        System.out.println("Here is the starting point of Thread.");
        for (;;) { // infinite loop to print message
            System.out.println("User Created Thread");
        }
    }
    public static void main(String[] argv) {
        Thread t = new Thread(new ThreadExample2()); // 產生Thread物件
        t.start(); // 開始執行Runnable.run();
        for (;;) {
            System.out.println("Main Thread");
        }
    }
}

Thread的優先權與影響資源的相關方法

Thread.setPriority(int)可以設定Thread的優先權,數字越大優先權越高。Thread定義了3個相關的static final variable

public static final int MAX_PRIORITY 10
public static final int MIN_PRIORITY 1
public static final int NORM_PRIORITY 5

要提醒讀者的是,優先權高的Thread其佔有CPU的機會比較高,但優先權低的也都會有機會執行到。其他有關Thread執行的方法有:

yield():先讓給別的Thread執行
sleep(int time):休息time mini second(1/1000秒)
join():呼叫ThreadA.join()的執行緒會等到ThreadA結束後,才能繼續執行

你可以執行下面的程式,看看yield()的效果

public class ThreadExample1 extends Thread {
    public void run() { // overwrite Thread's run()
        System.out.println("Here is the starting point of Thread.");
        for (;;) { // infinite loop to print message
            System.out.println("User Created Thread");
            yield();
        }
    }
    public static void main(String[] argv) {
        Thread t = new ThreadExample1(); // 產生Thread物件
        t.start(); // 開始執行t.run()
        for (;;) {
            System.out.println("Main Thread");
            yield();
        }
    }
}

觀看join的效果

public class JoinExample extends Thread {
    String myId;
    public JoinExample(String id) {
        myId = id;
    }
    public void run() { // overwrite Thread's run()
	for (int i=0; i < 500; i++) {
            System.out.println(myId+" Thread");
        }
    }
    public static void main(String[] argv) {
        Thread t1 = new JoinExample("T1"); // 產生Thread物件
        Thread t2 = new JoinExample("T2"); // 產生Thread物件
        t1.start(); // 開始執行t1.run()
        t2.start();
        try {
            t1.join(); // 等待t1結束
            t2.join(); // 等待t2結束
        } catch (InterruptedException e) {}
        for (int i=0;i < 5; i++) {
            System.out.println("Main Thread");
        }
    }
}

觀看sleep的效果

public class SleepExample extends Thread {
    String myId;
    public SleepExample(String id) {
        myId = id;
    }
    public void run() { // overwrite Thread's run()
        for (int i=0; i < 500; i++) {
            System.out.println(myId+" Thread");
            try {
                sleep(100);
            } catch (InterruptedException e) {}
        }
    }
    public static void main(String[] argv) {
        Thread t1 = new SleepExample("T1"); // 產生Thread物件
        Thread t2 = new SleepExample("T2"); // 產生Thread物件
        t1.start(); // 開始執行t1.run()
        t2.start();
    }
}

Critical Section(關鍵時刻)的保護措施

如果設計者沒有提供保護機制的話,Thread取得和失去CPU控制權的時機是由作業系統來決定。也就是說Thread可能在執行任何一個機器指令時,被作業系統取走CPU控制權,並交給另一個Thread。由於某些真實世界的動作是不可分割的,例如跨行轉帳X圓由A帳戶到B帳戶,轉帳前後這兩個帳戶的總金額必須相同,但以程式來實作時,卻無法用一個指令就完成,如轉帳可能要寫成下面的這一段程式碼

if (A >= X) {
    A = A - X; // 翻譯成3個機器指令LOAD A, SUB X, STORE A
    B = B +X;
}

如果兩個Thread同時要存取A,B兩帳戶進行轉帳,假設當Thread one執行到SUBX後被中斷,Threadtwo接手執行完成另一個轉帳要求,然後Threadone繼續執行未完成的動作,請問這兩個轉帳動作正確嗎?我們以A=1000,B=0,分別轉帳100,200圓來說明此結果

    LOAD A // Thread 1, 現在A還是1000
    SUB 100 // Thread 1
    LOAD A // 假設此時Thread 1被中斷,Thread 2接手, 因為Thread 1 還沒有執行STORE A, 所以變數A還是1000
    SUB 200 // Thread 2
    STORE A // Thread 2, A = 800
    LOAD B // Thread 2, B現在是0
    ADD 200 // Thread 2
    STORE B // B=200
    STORE A // Thread 1拿回控制權, A = 900
    LOAD B // Thread 1, B = 200
    ADD 100 // Thread 1
    STORE B // B = 300

你會發現執行完成後A=900,B=300,也就是說銀行平白損失了200圓。當然另外的執行順序可能造成其他不正確的結果。我們把這問題再整理一下:

寫程式時假設指令會循序執行
某些不可分割的動作,需要以多個機器指令來完成
Thread執行時可能在某個機器指令被中斷
兩個Thread可能執行同一段程式碼,存取同一個資料結構
這樣就破壞了第1點的假設

因此在撰寫多執行緒的程式時,必須特別考慮這種狀況(又稱為race condition)。Java的解決辦法是,JVM會在每個物件上擺一把鎖(lock),然後程式設計者可以宣告執行某一段程式(通常是用來存取共同資料結構的程式碼, 又稱為Critical Section)時,必須拿到某物件的鎖才行,這個鎖同時間最多只有一個執行緒可以擁有它。

public class Transfer extends Thread {
    public static Object lock = new Object();
    public static int A = 1000;
    public static int B = 0;
    private int amount;
    public Transfer(int x) {
        amount = x;
    }
    public void run() {
        synchronized(lock) { // 取得lock,如果別的thread A已取得,則目前這個thread會等到thread A釋放該lock
            if (A >= amount) {
                A = A - amount;
                B = B + amount;
            }
        } // 離開synchronized區塊後,此thread會自動釋放lock
    }
    public static void main(String[] argv) {
        Thread t1 = new Transfer(100);
        Thread t2 = new Transfer(200);
        t1.start();
        t2.start();
    }
}

除了synchronized(ref)的語法可以鎖定ref指到的物件外,synchronized也可以用在object method前面,表示要鎖定this物件才能執行該方法。以下是Queue結構的範例

public class Queue {
    private Object[] data;
    private int size;
    private int head;
    private int tail;
    public Queue(int maxLen) {
        data = new Object[maxLen];
    }
    public synchronized Object deQueue() {
        Object tmp = data[head];
        data[head] = null;
        head = (head+1)%data.length;
        size--;
        return tmp;
    }
    public synchronized void enQueue(Object c) {
        data[tail++] = c;
        tail %= data.length;
        size++;
    }
}

雖然上面的程式正確無誤,但並未考慮資源不足時該如何處理。例如Queue已經沒有資料了,卻還想拿出來;或是Queue裡已經塞滿了資料,使用者卻還要放進去?我們當然可以使用Exception Handling的機制:

public class Queue {
    private Object[] data;
    private int size;
    private int head;
    private int tail;
    public Queue(int maxLen) {
        data = new Object[maxLen];
    }
    public synchronized Object deQueue() throws Exception {
        if (size == 0) {
            throw new Exception();
        }
        Object tmp = data[head];
        data[head] = null;
        head = (head+1)%data.length;
        size--;
        return tmp;
    }
    public synchronized void enQueue(Object c) throws Exception {
        if (size >= maxLen) {
            throw new Exception();
        }
        data[tail++] = c;
        tail %= data.length;
        size++;
    }
}

但假設我們的執行環境是,某些Thread專門負責讀取使用者的需求,並把工作放到Queue裡面,某些Thread則專門由Queue裡抓取工作需求做進一步處理。這種架構的好處是,可以把慢速或不定速的輸入(如透過網路讀資料,連線速度可能差很多),和快速的處理分開,可使系統的反應速度更快,更節省資源。那麼以Exceptoin來處理Queue空掉或爆掉的情況並不合適,因為使用Queue的人必須處理例外狀況,並不斷的消耗CPU資源:

public class Getter extends Thread {
    Queue q;
    public Getter(Queue q) {
        this.q = q;
    }
    public void run() {
        for (;;) {
            try {
                Object data = q.deQueue();
                // processing
            } catch(Exception e) {
                // if we try to sleep here, user may feel slow response
                // if we do not sleep, CPU will be wasted
            }
        }
    }
}
public class Putter extends Thread {
    Queue q;
    public Putter(Queue q) {
        this.q = q;
    }
    public void run() {
        for (;;) {
            try {
                Object data = null;
                // get user request
                 q.enQueue(data);
            } catch(Exception e) {
                // if we try to sleep here, user may feel slow response
                // if we do not sleep, CPU will be wasted
            }
        }
    }
}
public class Main {
    public static void main(String[] argv) {
        Queue q = new Queue(10);
        Getter r1 = new Getter(q);
        Getter r2 = new Getter(q);
        Putter w1 = new Putter(q);
        Putter w2 = new Putter(q);
        r1.start();
        r2.start();
        w1.start();
        w2.start();
    }
}

為了解決這類資源分配的問題,Java Object提供了下面三個method:

wait():使呼叫此方法的Thread進入Blocking Mode,並設為等待該Object, 呼叫wait()時, 該Thread必須擁有該物件的lock。Blocking Mode下的Thread必須釋放所有手中的lock,並且無法使用CPU。
notifyAll():讓等待該Object的所有Thread進入Runnable Mode。
notify():讓等待該Object的某一個Thread進入Runnable Mode。

所謂Runnable Mode是指該Thread隨時可由作業系統分配CPU資源。Blocking Mode表示該Thread正在等待某個事件發生,作業系統不會讓這種Thread取得CPU資源。前一個Queue的範例就可以寫成:

public class Queue {
    private Object[] data;
    private int size;
    private int head;
    private int tail;
    public Queue(int maxLen) {
        data = new Object[maxLen];
    }
    public synchronized Object deQueue() {
        while (size==0) { // When executing here, Thread must have got lock and be in running mode
            // Let current Thread wait this object(to sleeping mode)
            try {
                wait(); // to sleeping mode, and release all lock
            } catch(Exception ex) {};
        }
        Object tmp = data[head];
        data[head] = null;
        head = (head+1)%data.length;
        if (size==data.length) {
            // wake up all Threads waiting this object
            notifyAll();
        }
        size--;
        return tmp;
    } // release lock
    public synchronized void enQueue(Object c) {
        while (size==data.length) {  // When executing here, Thread must have got lock and be in running mode
            // Let current thread wait this object(to sleeping mode)
            try {
                wait(); // to sleeping mode, and release all lock
            } catch(Exception ex) {};
        }
        data[tail++] = c;
        tail %= data.length;
        size++;
        if (size==1) {
            // wake up all Threads waiting this object
            notifyAll();
        }
    }
}


public class ReaderWriter extends Thread {
    public static final int READER = 1;
    public static final int WRITER = 2;
    private Queue q;
    private int mode;
    public void run() {
        for (int i=0; i < 1000; i++) {
            if (mode==READER) {
                q.deQueue();
            } else if (mode==WRITER) {
                q.enQueue(new Integer(i));
            }
        }
    }
    public ReaderWriter(Queue q, int mode) {
        this.q = q;
        this.mode = mode;
    }
    public static void main(String[] args) {
        Queue q = new Queue(5);
        ReaderWriter r1, r2, w1, w2;
        (w1 = new ReaderWriter(q, WRITER)).start();
        (w2 = new ReaderWriter(q, WRITER)).start();
        (r1 = new ReaderWriter(q, READER)).start();
        (r2 = new ReaderWriter(q, READER)).start();
        try {
            w1.join(); // wait until w1 complete
            w2.join(); // wait until w2 complete
            r1.join(); // wait until r1 complete
            r2.join(); // wait until r2 complete
        } catch(InterruptedException epp) {
        }
    }
}

Multiple Reader-Writer Monitors

上一節的Queue資料結構,不論是enQueue()或deQueue()都會更動到Queue的內容。而在許多應用裡,資料結構可以允許同時多個讀一個寫。本節舉出幾個不同的例子,說明多個Reader-Writer時的可能排程法。

Single Reader-Writer, 只同時允許一個執行緒存取

public class SingleReaderWriter {
    int n; // number of reader and write, 0 or 1
    public synchronized void startReading() throws InterruptedException {
        while (n != 0) {
            wait();
        }
        n = 1;
    }
    public synchronized void stopReading() {
        n = 0;
        notify();
    }
    public synchronized void startWriting() throws InterruptedException {
        while (n != 0) {
            wait();
        }
        n = 1;
    }
    public synchronized void stopWriting() {
        n = 0;
        notify();
    }
}
// 這是一個使用範例, 程式能否正確執行要靠呼叫正確的start和stop
public class WriterThread extends Thread {
    SingleReaderWriter srw;
    public WriterThread(SingleReaderWriter srw) {
        this.srw = srw;
    }
    public void run() {
        startWring();
        // insert real job here
        stopWriting();
    }
}
public class ReaderThread extends Thread {
    SingleReaderWriter srw;
    public ReaderThread(SingleReaderWriter srw) {
        this.srw = srw;
    }
    public void run() {
        startReading();
        // insert real job here
        stopReading();
    }
}
public class Test {
    public static void main(String[] argv) {
        SingleReaderWriter srw = new SingleReaderWriter;
        // create four threads
        (new WriterThread(srw)).start();
        (new WriterThread(srw)).start();
        (new ReaderThread(srw)).start();
        (new ReaderThread(srw)).start();
    }
}

其他可能的策略實作如下:

Reader優先:

public class ReadersPreferredMonitor {
    int nr; // The number of threads currently reading, nr > = 0
    int nw; // The number of threads currently writing, 0 or 1
    int nrtotal; // The number of threads either reading or waiting to read, nrtotal > = nr
    int nwtotal; // The number of threads either writing or waiting to write
    public synchronized void startReading() throws InterruptedException {
        nrtotal++; // 想要read的thread又多了一個
        while (nw != 0) { // 還有write thread正在write
            wait();
        }
        nr++; // 正在讀的thread多了一個
    }
    public synchronized void startWriting() throws InterruptedException {
        nwtotal++; // 想要寫的thread又多了一個
        while (nrtotal+nw != 0) { // 只要有thread想要讀,或是有thread正在寫,禮讓
            wait();
        }
        nw = 1;
    }
    public synchronized void stopReading() {
        nr--; // 正在讀的少一個
        nrtotal--; // 想要讀的少一個
        if (nrtotal == 0) { // 如果沒有要讀的,叫醒想寫的
            notify();
        }
    }
    public synchronized void stopWriting() {
        nw = 0; // 沒有thread正在寫
        nwtotal--; // 想寫的少一個
        notifyAll(); // 叫醒所有想讀和想寫的
    }
}

Writer優先:

public class WritersPreferredMonitor {
    int nr; // The number of threads currently reading, nr > = 0
    int nw; // The number of threads currently writing, 0 or 1
    int nrtotal; // The number of threads either reading or waiting to read, nrtotal > = nr
    int nwtotal; // The number of threads either writing or waiting to write
    public synchronized void startReading() throws InterruptedException {
        nrtotal++; // 想要read的thread又多了一個
        while (nwtotal != 0) { // 還有thread想要write
            wait();
        }
        nr++; // 正在讀的thread多了一個
    }
    public synchronized void startWriting() throws InterruptedException {
        nwtotal++; // 想要寫的thread又多了一個
        while (nr+nw != 0) { // 有thread正在讀,或是有thread正在寫
            wait();
        }
        nw = 1;
    }
    public synchronized void stopReading() {
        nr--; // 正在讀的少一個
        nrtotal--; // 想要讀的少一個
        if (nr == 0) { // 如果沒有正在讀的,叫醒所有的(包括想寫的)
            notifyAll();
        }
    }
    public synchronized void stopWriting() {
        nw = 0; // 沒有thread正在寫
        nwtotal--; // 想寫的少一個
        notifyAll(); // 叫醒所有想讀和想寫的
    }
}

Reader和Writer交互執行:

public class AlternatingReadersWritersMonitor {
    int[] nr = new int[2]; // The number of threads currently reading
    int thisBatch; // Index in nr of the batch of readers currently reading(0 or 1)
    int nextBatch = 1; // Index in nr of the batch of readers waitin to read(always 1-thisBatch)
    int nw; // The number of threads currently writing(0 or 1)
    int nwtotal; // The number of threads either writing or waiting to write
    public synchronized void startReading() throws InterruptedException {
        if (nwtotal == 0) { // 沒有thread要write, 將reader都放到目前要處理的這一批
            nr[thisBatch]++;
        } else {
            nr[nextBatch]++;
            int myBatch = nextBatch;
            while (thisBatch != myBatch) {
                wait();
            }
        }
    }
    public synchronized void stopReading() {
        nr[thisBatch]--;
        if (nr[thisBatch] == 0) { // 目前這批的reader都讀完了,找下一個writer
            notifyAll();
        }
    }
    public synchronized void startWriting() throws InterruptedException {
        nwtotal++;
        while (nr[thisBatch]+nw != 0) { // 目前這批還沒完,或有thread正在寫
            wait();
        }
        nw = 1;
    }
    public synchronized void stopWriting() {
        nw = 0;
        nwtotal--;
        int tmp = thisBatch; // 交換下一批要讀的
        thisBatch = nextBatch;
        nextBatch = tmp;
        notifyAll();
    }
}

給號依序執行

public class TakeANumberMonitor {
    int nr; // The number of threads currently reading
    int nextNumber; // The number to be taken by the next thread to arrive
    int nowServing; // The number of the thread to be served next
    public synchronized void startReading() throws InterruptedException {
        int myNumber = nextNumber++;
        while (nowServing != myNumber) { // 還沒輪到我
            wait();
        }
        nr++; // 多了一個Reader
        nowServing++; // 準備檢查下一個
        notifyAll();
    }
    public synchronized void startWriting() throws InterruptedException {
        int myNumber = nextNumber++;
        while (nowServing != myNumber) { // 還沒輪到我
            wait();
        }
        while (nr >  0) { // 要等所有的Reader結束
            wait();
        }
    }
    public synchronized void stopReading() {
        nr--; // 少了一個Reader
        if (nr == 0) {
            notifyAll();
        }
    }
    public synchronized void stopWriting() {
        nowServing++; // 準備檢查下一個
        notifyAll();
    }
}