Question

我在使用spring-rabbit-1.3.9.RELEASE库进行Rabbitmq 3.3.5的POC时观察到奇怪的行为

当我开始一个生产线程时，事情顺利进行。但是如果同时启动多个线程，只有其中一个线程完成，所有其他线程无限期地被阻塞，即使队列变空也是如此。

从rabbitmqctl list_connections进行监控时，被阻止线程的连接状态仍然在运行。应该注意的是，当生产者阻止时，或者在整个运行期间的任何其他时间，都没有警报。

我还观察到，如果我在每次发送后进行1毫秒的睡眠，问题就会消失。

所以，我有这些问题

rabbitmq不支持并发生产者，高价出版吗？
即使连接确实被阻止，为什么它不会显示在rabbitmqctl list_connections中？
为什么它们会无限期阻塞而不能恢复乳清队列变空？

代码

    public static void main(String[] argv) throws java.io.IOException, InterruptedException {
        init();
        PocConfig config = new PocConfig();
        int threadCount = config.getThreadCount();
        final int eventsPerThread = config.getEvents() / threadCount;
        final long sleep = config.getSleep();

        System.out.println("Start producer with configuration [threadCount=" + threadCount + ", events=" + eventsPerThread + ", sleep="
            + sleep + "]");

        ExecutorService executorService = Executors.newFixedThreadPool(threadCount);
        for (int i = 0; i < threadCount; i++) {
            final int threadId = i;
            executorService.submit(new Runnable() {
                public void run() {
                    produce(eventsPerThread, sleep, threadId);
                }
            });
        }
        waitAndTearDown(executorService);
    }

    private static void produce(int events, long sleep, int threadId)     {
        long start = System.currentTimeMillis();
        for (int index = 1; index <= events; index++) {
            try {
                byte[] message = messageFactory.createTestMessage(index);
                amqpTemplate.convertAndSend(QUEUE_NAME, message);
                if (sleep > 0) {
                    Thread.sleep(sleep);
                }
            } catch (Exception e) {
                LOG.error("Error", e);
            }
        }
        long time = System.currentTimeMillis() - start;
        System.out.println("Producer:" + threadId + " finished, events: " + events + ", Time(s): " + time / 1000 + ", tps: " + (events * 1000) / time);
    }

Spring配置

<bean id="connectionFactory" class="org.springframework.amqp.rabbit.connection.CachingConnectionFactory">
    <property name="addresses" value="${addresses}" />
    <property name="username" value="${user}" />
    <property name="password" value="${passwd}" />
    <property name="cacheMode" value="CONNECTION" />
    <property name="connectionCacheSize" value="${threads}" />
    <property name="channelCacheSize" value="10" />
</bean>

<rabbit:template id="template" connection-factory="connectionFactory"
    exchange="testExchange" routing-key="testQueue"/>

Answer 1

没有什么我能想到会阻挡，所以我只是跑了你的考试;没有问题：

Start producer with configuration [threadCount=5, events=10, sleep=0]
Producer:2 finished, events: 1000, Time(s): 0, tps: 4405
Producer:3 finished, events: 1000, Time(s): 0, tps: 4132
Producer:1 finished, events: 1000, Time(s): 0, tps: 4048
Producer:0 finished, events: 1000, Time(s): 0, tps: 3968
Producer:4 finished, events: 1000, Time(s): 0, tps: 3952

是什么让你觉得他们被封锁了？

进行线程转储（例如使用jstack）以查看线程正在做什么。

修改：

即使有1M条消息和CacheMode CONNECTION ...
，我仍然无法重现它
Start producer with configuration [threadCount=5, events=200000, sleep=0] Producer:0 finished, events: 200000, Time(s): 50, tps: 3959 Producer:3 finished, events: 200000, Time(s): 53, tps: 3746 Producer:1 finished, events: 200000, Time(s): 55, tps: 3635 Producer:2 finished, events: 200000, Time(s): 55, tps: 3634 Producer:4 finished, events: 200000, Time(s): 55, tps: 3629

我执行看到队列进入flow模式（通过管理界面），但一切恢复正常。

我确实看到你的工人受到流量控制......

"pool-2-thread-3" prio=10 tid=0x00007f4af4849800 nid=0x65d5 runnable [0x00007f4ae082f000] java.lang.Thread.State: RUNNABLE at java.net.SocketOutputStream.socketWrite0(Native Method) at java.net.SocketOutputStream.socketWrite(SocketOutputStream.java:113)

你在兔子日志中看到了什么吗？您在管理界面上看到的关于消息率，状态等的内容是什么？

无论如何，这似乎与Spring AMQP没有任何关系;你需要联系rabbitmq-users谷歌小组的兔子家伙。

（我正在用rabbitmq 3.4.2测试）。

<强> EDIT2：

完全安装3.5.2 ......

Start producer with configuration [threadCount=5, events=200000, sleep=0] Producer:0 finished, events: 200000, Time(s): 39, tps: 5091 Producer:1 finished, events: 200000, Time(s): 39, tps: 5002 Producer:2 finished, events: 200000, Time(s): 40, tps: 4954 Producer:3 finished, events: 200000, Time(s): 40, tps: 4951 Producer:4 finished, events: 200000, Time(s): 40, tps: 4939

我在管理界面中看到没有flow状态（在队列中，但是频道/连接显示它们在流中，但又一次恢复）。

并发生产者在Rabbitmq中无限期地阻止

1 个答案: