mqttclient 性能 &MQTT(32202): 正在发布过多的消息
org.eclipse.paho.client.mqttv3
2.2 GHz Intel Core i7 mac 系统
publish 性能,注意请使用单线程的 mqttclinet
1 万条 341 毫秒
4 万条 1163 毫秒
5 万 1450 毫秒
10 万条 2700 毫秒
多线程的 mqttclinet MQTT(32202): 正在发布过多的消息 问题
异常信息
[15:07:21]: publish failed, message: aaaa
正在进行过多的发布 (32202)
at org.eclipse.paho.client.mqttv3.internal.ClientState.send(ClientState.java:496)
at org.eclipse.paho.client.mqttv3.internal.ClientComms.internalSend(ClientComms.java:132)
at org.eclipse.paho.client.mqttv3.internal.ClientComms.sendNoWait(ClientComms.java:156)
at org.eclipse.paho.client.mqttv3.MqttAsyncClient.publish(MqttAsyncClient.java:1027)
at org.eclipse.paho.client.mqttv3.MqttClient.publish(MqttClient.java:399)
at io.communet.ichater.emq.util.MqttUtil.publishMsg(MqttUtil.java:171)
at io.communet.ichater.emq.util.MqttUtil.publishMsg(MqttUtil.java:161)
at io.communet.ichater.emq.sub.MqttSendMsgEventSubscribe.onEvent(MqttSendMsgEventSubscribe.java:28)
at java.lang.reflect.Method.invoke(Native Method)
at java.lang.reflect.Method.invoke(Method.java:372)
at org.greenrobot.eventbus.EventBus.invokeSubscriber(EventBus.java:507)
at org.greenrobot.eventbus.EventBus.invokeSubscriber(EventBus.java:501)
at org.greenrobot.eventbus.AsyncPoster.run(AsyncPoster.java:46)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1112)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:587)
at java.lang.Thread.run(Thread.java:818)
解决办法
- 消息发送发送限流
- 用单独的一个线程来完成 MQ 消息的推送 (不用这个 MqttAsyncClient,使用 MqttClient 就没有事)
-
options.setMaxInflight(1000)
增加actualInFlight
的值;
反思
笔者出现这个错误是因为使用 EventBus
, 之前使用单独线程的 Handler
是没有问题的, 调查发现, 使用 EventBus
是新建线程运行的, 而 Handler
是单独一个线程.
所以当发送大量消息的时候, EventBus
几乎是同一个点发出去, 就会造成这个错误
原因
根据堆栈信息找到报错地方
if (actualInFlight >= this.maxInflight) {//@TRACE 613= sending {0} msgs at max inflight window
log.fine(CLASS_NAME, methodName, "613", new Object[]{new Integer(actualInFlight)});
throw new MqttException(MqttException.REASON_CODE_MAX_INFLIGHT);
}
其中 actualInFlight
如下
// processed until the inflight window has space.
if (actualInFlight < this.maxInflight) {
// The in flight window is not full so process the
// first message in the queue
result = (MqttWireMessage)pendingMessages.elementAt(0);
pendingMessages.removeElementAt(0);
actualInFlight++;
//@TRACE 623=+1 actualInFlight={0}
log.fine(CLASS_NAME,methodName,"623",new Object[]{new Integer(actualInFlight)});
}
从 pendingMessages
中取出消息时, actualInFlight
加 1, maxInflight
可以自己设定, 默认值为 10.
public class ClientState {
...
volatile private Vector pendingMessages;
...
}
在 ClientState
中:
public void send(MqttWireMessage message, MqttToken token) throws MqttException {
...
if (message instanceof MqttPublish) {synchronized (queueLock) {if (actualInFlight >= this.maxInflight) {//@TRACE 613= sending {0} msgs at max inflight window
log.fine(CLASS_NAME, methodName, "613", new Object[]{new Integer(actualInFlight)});
throw new MqttException(MqttException.REASON_CODE_MAX_INFLIGHT);
}
MqttMessage innerMessage = ((MqttPublish) message).getMessage();
//@TRACE 628=pending publish key={0} qos={1} message={2}
log.fine(CLASS_NAME,methodName,"628", new Object[]{new Integer(message.getMessageId()), new Integer(innerMessage.getQos()), message});
switch(innerMessage.getQos()) {
case 2:
outboundQoS2.put(new Integer(message.getMessageId()), message);
persistence.put(getSendPersistenceKey(message), (MqttPublish) message);
break;
case 1:
outboundQoS1.put(new Integer(message.getMessageId()), message);
persistence.put(getSendPersistenceKey(message), (MqttPublish) message);
break;
}
tokenStore.saveToken(token, message);
pendingMessages.addElement(message);
queueLock.notifyAll();}
} else {...}
}
可以看到 pendingMessages
中添加元素的时候并没有做 qos
类型的判断
private void decrementInFlight() {
final String methodName = "decrementInFlight";
synchronized (queueLock) {
actualInFlight--;
//@TRACE 646=-1 actualInFlight={0}
log.fine(CLASS_NAME,methodName,"646",new Object[]{new Integer(actualInFlight)});
if (!checkQuiesceLock()) {queueLock.notifyAll();
}
}
}
当收到消息反馈时 actualInFlight
减 1.