RabbitMQ (二)工作隊列
轉載請標明出處:http://blog.csdn.net/lmj623565791/article/details/37620057
本系列教程主要來自于官網入門教程的翻譯,然后自己進行了部分的修改與實驗,內容僅供參考。
上一篇博客中我們寫了通過一個命名的隊列發送和接收消息,如果你還不了解請點擊:RabbitMQ 入門 Helloworld。這篇中我們將會創建一個工作隊列用來在工作者(consumer)間分發耗時任務。
工作隊列的主要任務是:避免立刻執行資源密集型任務,然后必須等待其完成。相反地,我們進行任務調度:我們把任務封裝為消息發送給隊列。工作進行在后臺運行并不斷的從隊列中取出任務然后執行。當你運行了多個工作進程時,任務隊列中的任務將會被工作進程共享執行。這樣的概念在web應用中極其有用,當在很短的HTTP請求間需要執行復雜的任務。
1、 準備
我們使用Thread.sleep來模擬耗時的任務。我們在發送到隊列的消息的末尾添加一定數量的點,每個點代表在工作線程中需要耗時1秒,例如hello…將會需要等待3秒。
發送端:
NewTask.java
package com.zhy.rabbit._02_workqueue; import java.io.IOException; import com.rabbitmq.client.Channel; import com.rabbitmq.client.Connection; import com.rabbitmq.client.ConnectionFactory; public class NewTask { //隊列名稱 private final static String QUEUE_NAME = "workqueue"; public static void main(String[] args) throws IOException { //創建連接和頻道 ConnectionFactory factory = new ConnectionFactory(); factory.setHost("localhost"); Connection connection = factory.newConnection(); Channel channel = connection.createChannel(); //聲明隊列 channel.queueDeclare(QUEUE_NAME, false, false, false, null); //發送10條消息,依次在消息后面附加1-10個點 for (int i = 0; i < 10; i++) { String dots = ""; for (int j = 0; j <= i; j++) { dots += "."; } String message = "helloworld" + dots+dots.length(); channel.basicPublish("", QUEUE_NAME, null, message.getBytes()); System.out.println(" [x] Sent '" + message + "'"); } //關閉頻道和資源 channel.close(); connection.close(); } }
接收端:
Work.java
package com.zhy.rabbit._02_workqueue; import com.rabbitmq.client.Channel; import com.rabbitmq.client.Connection; import com.rabbitmq.client.ConnectionFactory; import com.rabbitmq.client.QueueingConsumer; public class Work { //隊列名稱 private final static String QUEUE_NAME = "workqueue"; public static void main(String[] argv) throws java.io.IOException, java.lang.InterruptedException { //區分不同工作進程的輸出 int hashCode = Work.class.hashCode(); //創建連接和頻道 ConnectionFactory factory = new ConnectionFactory(); factory.setHost("localhost"); Connection connection = factory.newConnection(); Channel channel = connection.createChannel(); //聲明隊列 channel.queueDeclare(QUEUE_NAME, false, false, false, null); System.out.println(hashCode + " [*] Waiting for messages. To exit press CTRL+C"); QueueingConsumer consumer = new QueueingConsumer(channel); // 指定消費隊列 channel.basicConsume(QUEUE_NAME, true, consumer); while (true) { QueueingConsumer.Delivery delivery = consumer.nextDelivery(); String message = new String(delivery.getBody()); System.out.println(hashCode + " [x] Received '" + message + "'"); doWork(message); System.out.println(hashCode + " [x] Done"); } } /** * 每個點耗時1s * @param task * @throws InterruptedException */ private static void doWork(String task) throws InterruptedException { for (char ch : task.toCharArray()) { if (ch == '.') Thread.sleep(1000); } } }
Round-robin 轉發
使用任務隊列的好處是能夠很容易的并行工作。如果我們積壓了很多工作,我們僅僅通過增加更多的工作者就可以解決問題,使系統的伸縮性更加容易。
下面我們先運行3個工作者(Work.java)實例,然后運行NewTask.java,3個工作者實例都會得到信息。但是如何分配呢?讓我們來看輸出結果:[x] Sent 'helloworld.1'
[x] Sent 'helloworld..2'
[x] Sent 'helloworld...3'
[x] Sent 'helloworld....4'
[x] Sent 'helloworld.....5'
[x] Sent 'helloworld......6'
[x] Sent 'helloworld.......7'
[x] Sent 'helloworld........8'
[x] Sent 'helloworld.........9'
[x] Sent 'helloworld..........10'
工作者1:
605645 [*] Waiting for messages. To exit press CTRL+C
605645 [x] Received 'helloworld.1'
605645 [x] Done
605645 [x] Received 'helloworld....4'
605645 [x] Done
605645 [x] Received 'helloworld.......7'
605645 [x] Done
605645 [x] Received 'helloworld..........10'
605645 [x] Done
工作者2:
18019860 [*] Waiting for messages. To exit press CTRL+C
18019860 [x] Received 'helloworld..2'
18019860 [x] Done
18019860 [x] Received 'helloworld.....5'
18019860 [x] Done
18019860 [x] Received 'helloworld........8'
18019860 [x] Done
工作者3:
18019860 [*] Waiting for messages. To exit press CTRL+C
18019860 [x] Received 'helloworld...3'
18019860 [x] Done
18019860 [x] Received 'helloworld......6'
18019860 [x] Done
18019860 [x] Received 'helloworld.........9'
18019860 [x] Done
可以看到,默認的,RabbitMQ會一個一個的發送信息給下一個消費者(consumer),而不考慮每個任務的時長等等,且是一次性分配,并非一個一個分配。平均的每個消費者將會獲得相等數量的消息。這樣分發消息的方式叫做round-robin。
2、 消息應答(message acknowledgments)
執行一個任務需要花費幾秒鐘。你可能會擔心當一個工作者在執行任務時發生中斷。我們上面的代碼,一旦RabbItMQ交付了一個信息給消費者,會馬上從內存中移除這個信息。在這種情況下,如果殺死正在執行任務的某個工作者,我們會丟失它正在處理的信息。我們也會丟失已經轉發給這個工作者且它還未執行的消息。上面的例子,我們首先開啟兩個任務,然后執行發送任務的代碼(NewTask.java),然后立即關閉第二個任務,結果為:
工作者2:
31054905 [*] Waiting for messages. To exit press CTRL+C
31054905 [x] Received 'helloworld..2'
31054905 [x] Done
31054905 [x] Received 'helloworld....4'
工作者1:
18019860 [*] Waiting for messages. To exit press CTRL+C
18019860 [x] Received 'helloworld.1'
18019860 [x] Done
18019860 [x] Received 'helloworld...3'
18019860 [x] Done
18019860 [x] Received 'helloworld.....5'
18019860 [x] Done
18019860 [x] Received 'helloworld.......7'
18019860 [x] Done
18019860 [x] Received 'helloworld.........9'
18019860 [x] Done
可以看到,第二個工作者至少丟失了6,8,10號任務,且4號任務未完成。
但是,我們不希望丟失任何任務(信息)。當某個工作者(接收者)被殺死時,我們希望將任務傳遞給另一個工作者。
為了保證消息永遠不會丟失,RabbitMQ支持消息應答(message acknowledgments)。消費者發送應答給RabbitMQ,告訴它信息已經被接收和處理,然后RabbitMQ可以自由的進行信息刪除。
如果消費者被殺死而沒有發送應答,RabbitMQ會認為該信息沒有被完全的處理,然后將會重新轉發給別的消費者。通過這種方式,你可以確認信息不會被丟失,即使消者偶爾被殺死。
這種機制并沒有超時時間這么一說,RabbitMQ只有在消費者連接斷開是重新轉發此信息。如果消費者處理一個信息需要耗費特別特別長的時間是允許的。
消息應答默認是打開的。上面的代碼中我們通過顯示的設置autoAsk=true關閉了這種機制。下面我們修改代碼(Work.java):
boolean ack = false ; //打開應答機制 channel.basicConsume(QUEUE_NAME, ack, consumer); //另外需要在每次處理完成一個消息后,手動發送一次應答。 channel.basicAck(delivery.getEnvelope().getDeliveryTag(), false);
完整修改后的Work.java
package com.zhy.rabbit._02_workqueue.ack; import com.rabbitmq.client.Channel; import com.rabbitmq.client.Connection; import com.rabbitmq.client.ConnectionFactory; import com.rabbitmq.client.QueueingConsumer; public class Work { //隊列名稱 private final static String QUEUE_NAME = "workqueue"; public static void main(String[] argv) throws java.io.IOException, java.lang.InterruptedException { //區分不同工作進程的輸出 int hashCode = Work.class.hashCode(); //創建連接和頻道 ConnectionFactory factory = new ConnectionFactory(); factory.setHost("localhost"); Connection connection = factory.newConnection(); Channel channel = connection.createChannel(); //聲明隊列 channel.queueDeclare(QUEUE_NAME, false, false, false, null); System.out.println(hashCode + " [*] Waiting for messages. To exit press CTRL+C"); QueueingConsumer consumer = new QueueingConsumer(channel); // 指定消費隊列 boolean ack = false ; //打開應答機制 channel.basicConsume(QUEUE_NAME, ack, consumer); while (true) { QueueingConsumer.Delivery delivery = consumer.nextDelivery(); String message = new String(delivery.getBody()); System.out.println(hashCode + " [x] Received '" + message + "'"); doWork(message); System.out.println(hashCode + " [x] Done"); //發送應答 channel.basicAck(delivery.getEnvelope().getDeliveryTag(), false); } } }測試:
我們把消息數量改為5,然后先打開兩個消費者(Work.java),然后發送任務(NewTask.java),立即關閉一個消費者,觀察輸出:
[x] Sent 'helloworld.1'
[x] Sent 'helloworld..2'
[x] Sent 'helloworld...3'
[x] Sent 'helloworld....4'
[x] Sent 'helloworld.....5'
工作者2
18019860 [*] Waiting for messages. To exit press CTRL+C
18019860 [x] Received 'helloworld..2'
18019860 [x] Done
18019860 [x] Received 'helloworld....4'
工作者1
31054905 [*] Waiting for messages. To exit press CTRL+C
31054905 [x] Received 'helloworld.1'
31054905 [x] Done
31054905 [x] Received 'helloworld...3'
31054905 [x] Done
31054905 [x] Received 'helloworld.....5'
31054905 [x] Done
31054905 [x] Received 'helloworld....4'
31054905 [x] Done
可以看到工作者2沒有完成的任務4,重新轉發給工作者1進行完成了。
3、 消息持久化(Message durability)
我們已經學習了即使消費者被殺死,消息也不會被丟失。但是如果此時RabbitMQ服務被停止,我們的消息仍然會丟失。
當RabbitMQ退出或者異常退出,將會丟失所有的隊列和信息,除非你告訴它不要丟失。我們需要做兩件事來確保信息不會被丟失:我們需要給所有的隊列和消息設置持久化的標志。
第一, 我們需要確認RabbitMQ永遠不會丟失我們的隊列。為了這樣,我們需要聲明它為持久化的。
boolean durable = true;
channel.queueDeclare("task_queue", durable, false, false, null);
注:RabbitMQ不允許使用不同的參數重新定義一個隊列,所以已經存在的隊列,我們無法修改其屬性。
第二, 我們需要標識我們的信息為持久化的。通過設置MessageProperties(implements BasicProperties)值為PERSISTENT_TEXT_PLAIN。
channel.basicPublish("", "task_queue",MessageProperties.PERSISTENT_TEXT_PLAIN,message.getBytes());
現在你可以執行一個發送消息的程序,然后關閉服務,再重新啟動服務,運行消費者程序做下實驗。
4、公平轉發(Fair dispatch)
或許會發現,目前的消息轉發機制(Round-robin)并非是我們想要的。例如,這樣一種情況,對于兩個消費者,有一系列的任務,奇數任務特別耗時,而偶數任務卻很輕松,這樣造成一個消費者一直繁忙,另一個消費者卻很快執行完任務后等待。造成這樣的原因是因為RabbitMQ僅僅是當消息到達隊列進行轉發消息。并不在乎有多少任務消費者并未傳遞一個應答給RabbitMQ。僅僅盲目轉發所有的奇數給一個消費者,偶數給另一個消費者。
為了解決這樣的問題,我們可以使用basicQos方法,傳遞參數為prefetchCount = 1。這樣告訴RabbitMQ不要在同一時間給一個消費者超過一條消息。換句話說,只有在消費者空閑的時候會發送下一條信息。
int prefetchCount = 1; channel.basicQos(prefetchCount);

測試:改變發送消息的代碼,將消息末尾點數改為6-2個,然后首先開啟兩個工作者,接著發送消息:
[x] Sent 'helloworld.....5'
[x] Sent 'helloworld....4'
[x] Sent 'helloworld...3'
[x] Sent 'helloworld..2'
工作者1:
18019860 [*] Waiting for messages. To exit press CTRL+C
18019860 [x] Received 'helloworld......6'
18019860 [x] Done
18019860 [x] Received 'helloworld...3'
18019860 [x] Done
工作者2:
31054905 [*] Waiting for messages. To exit press CTRL+C
31054905 [x] Received 'helloworld.....5'
31054905 [x] Done
31054905 [x] Received 'helloworld....4'
31054905 [x] Done
31054905 [x] Received 'helloworld..2'
31054905 [x] Done
可以看出此時并沒有按照之前的Round-robin機制進行轉發消息,而是當消費者不忙時進行轉發。且這種模式下支持動態增加消費者,因為消息并沒有發送出去,動態增加了消費者馬上投入工作。而默認的轉發機制會造成,即使動態增加了消費者,此時的消息已經分配完畢,無法立即加入工作,即使有很多未完成的任務。
5、完整的代碼
NewTask.java
package com.zhy.rabbit._02_workqueue.ackandpersistence; import java.io.IOException; import com.rabbitmq.client.Channel; import com.rabbitmq.client.Connection; import com.rabbitmq.client.ConnectionFactory; import com.rabbitmq.client.MessageProperties; public class NewTask { // 隊列名稱 private final static String QUEUE_NAME = "workqueue_persistence"; public static void main(String[] args) throws IOException { // 創建連接和頻道 ConnectionFactory factory = new ConnectionFactory(); factory.setHost("localhost"); Connection connection = factory.newConnection(); Channel channel = connection.createChannel(); // 聲明隊列 boolean durable = true;// 1、設置隊列持久化 channel.queueDeclare(QUEUE_NAME, durable, false, false, null); // 發送10條消息,依次在消息后面附加1-10個點 for (int i = 5; i > 0; i--) { String dots = ""; for (int j = 0; j <= i; j++) { dots += "."; } String message = "helloworld" + dots + dots.length(); // MessageProperties 2、設置消息持久化 channel.basicPublish("", QUEUE_NAME, MessageProperties.PERSISTENT_TEXT_PLAIN, message.getBytes()); System.out.println(" [x] Sent '" + message + "'"); } // 關閉頻道和資源 channel.close(); connection.close(); } }
Work.java
package com.zhy.rabbit._02_workqueue.ackandpersistence; import com.rabbitmq.client.Channel; import com.rabbitmq.client.Connection; import com.rabbitmq.client.ConnectionFactory; import com.rabbitmq.client.QueueingConsumer; public class Work { // 隊列名稱 private final static String QUEUE_NAME = "workqueue_persistence"; public static void main(String[] argv) throws java.io.IOException, java.lang.InterruptedException { // 區分不同工作進程的輸出 int hashCode = Work.class.hashCode(); // 創建連接和頻道 ConnectionFactory factory = new ConnectionFactory(); factory.setHost("localhost"); Connection connection = factory.newConnection(); Channel channel = connection.createChannel(); // 聲明隊列 boolean durable = true; channel.queueDeclare(QUEUE_NAME, durable, false, false, null); System.out.println(hashCode + " [*] Waiting for messages. To exit press CTRL+C"); //設置最大服務轉發消息數量 int prefetchCount = 1; channel.basicQos(prefetchCount); QueueingConsumer consumer = new QueueingConsumer(channel); // 指定消費隊列 boolean ack = false; // 打開應答機制 channel.basicConsume(QUEUE_NAME, ack, consumer); while (true) { QueueingConsumer.Delivery delivery = consumer.nextDelivery(); String message = new String(delivery.getBody()); System.out.println(hashCode + " [x] Received '" + message + "'"); doWork(message); System.out.println(hashCode + " [x] Done"); //channel.basicAck(delivery.getEnvelope().getDeliveryTag(), false); channel.basicAck(delivery.getEnvelope().getDeliveryTag(), false); } } /** * 每個點耗時1s * * @param task * @throws InterruptedException */ private static void doWork(String task) throws InterruptedException { for (char ch : task.toCharArray()) { if (ch == '.') Thread.sleep(1000); } } }
來自: http://blog.csdn.net//lmj623565791/article/details/37620057