ActiveMQ的客户端支持失效重连。昨天在配置ActiveMQ的主从结构后，客户端使用失效重连协议连接到代理，如下所示：
failover:(tcp://master:61616,tcp://slave:61616)
测试的时候客户端有时候连接到master上，有时候连接到slave上。这就尴尬了，正常理解的话，配在前面的broker url应该先尝试连接，连接不上才连接后面的url，即从前到后的顺序，但是测试结果来看，是不一定的。
使用logback打开了ActiveMQ的日志
<logger name="org.apache.activemq" level="debug" additivity="false">
<appender-ref ref="STDOUT"/>
</logger>
日志中输出的确有的时候只直接连了slave，而不是每次都先连master，而且master是正常的，那就说明的确客户端是做过处理了，并不是按从前到后的顺序依次连接。
日志中会输出“Successfully connected to slave...”，logback配置中输出了代码行数，因此很快定位到程序，org.apache.activemq.transport.failover.FailoverTransport类的876行(客户端使用的是activemq-all-5.4.3.jar)，代码如下
final boolean doReconnect() {
Exception failure = null;
synchronized (reconnectMutex) {
// If updateURIsURL is specified, read the file and add any new
// transport URI's to this FailOverTransport.
// Note: Could track file timestamp to avoid unnecessary reading.
String fileURL = getUpdateURIsURL();
if (fileURL != null) {
BufferedReader in = null;
String newUris = null;
StringBuffer buffer = new StringBuffer();
try {
in = new BufferedReader(getURLStream(fileURL));
while (true) {
String line = in.readLine();
if (line == null) {
break;
}
buffer.append(line);
}
newUris = buffer.toString();
} catch (IOException ioe) {
LOG.error("Failed to read updateURIsURL: " + fileURL, ioe);
} finally {
if (in != null) {
try {
in.close();
} catch (IOException ioe) {
// ignore
}
}
}
processNewTransports(isRebalanceUpdateURIs(), newUris);
}
if (disposed || connectionFailure != null) {
reconnectMutex.notifyAll();
}
if (connectedTransport.get() != null || disposed || connectionFailure != null) {
return false;
} else {
List<URI> connectList = getConnectList();
if (connectList.isEmpty()) {
failure = new IOException("No uris available to connect to.");
} else {
if (!useExponentialBackOff || reconnectDelay == -1) {
reconnectDelay = initialReconnectDelay;
}
synchronized (backupMutex) {
if (backup && !backups.isEmpty()) {
BackupTransport bt = backups.remove(0);
Transport t = bt.getTransport();
URI uri = bt.getUri();
t.setTransportListener(myTransportListener);
try {
if (started) {
restoreTransport(t);
}
reconnectDelay = initialReconnectDelay;
failedConnectTransportURI = null;
connectedTransportURI = uri;
connectedTransport.set(t);
reconnectMutex.notifyAll();
connectFailures = 0;
LOG.info("Successfully reconnected to backup " + uri);
return false;
} catch (Exception e) {
LOG.debug("Backup transport failed", e);
}
}
}
Iterator<URI> iter = connectList.iterator();
while (iter.hasNext() && connectedTransport.get() == null && !disposed) {
URI uri = iter.next();
Transport t = null;
try {
LOG.debug("Attempting connect to: " + uri);
SslContext.setCurrentSslContext(brokerSslContext);
t = TransportFactory.compositeConnect(uri);
t.setTransportListener(myTransportListener);
t.start();
if (started) {
restoreTransport(t);
}
LOG.debug("Connection established");
reconnectDelay = initialReconnectDelay;
connectedTransportURI = uri;
connectedTransport.set(t);
reconnectMutex.notifyAll();
connectFailures = 0;
// Make sure on initial startup, that the
// transportListener
// has been initialized for this instance.
synchronized (listenerMutex) {
if (transportListener == null) {
try {
// if it isn't set after 2secs - it
// probably never will be
listenerMutex.wait(2000);
} catch (InterruptedException ex) {
}
}
}
if (transportListener != null) {
transportListener.transportResumed();
} else {
LOG.debug("transport resumed by transport listener not set");
}
if (firstConnection) {
firstConnection = false;
LOG.info("Successfully connected to " + uri);
} else {
LOG.info("Successfully reconnected to " + uri);
}
connected = true;
return false;
} catch (Exception e) {
failure = e;
LOG.debug("Connect fail to: " + uri + ", reason: " + e);
if (t != null) {
try {
t.stop();
} catch (Exception ee) {
LOG.debug("Stop of failed transport: " + t + " failed with reason: " + ee);
}
}
} finally {
SslContext.setCurrentSslContext(null);
}
}
}
}
int reconnectAttempts = 0;
if (firstConnection) {
if (this.startupMaxReconnectAttempts != 0) {
reconnectAttempts = this.startupMaxReconnectAttempts;
}
}
if (reconnectAttempts == 0) {
reconnectAttempts = this.maxReconnectAttempts;
}
if (reconnectAttempts > 0 && ++connectFailures >= reconnectAttempts) {
LOG.error("Failed to connect to transport after: " + connectFailures + " attempt(s)");
connectionFailure = failure;
// Make sure on initial startup, that the transportListener has
// been initialized
// for this instance.
synchronized (listenerMutex) {
if (transportListener == null) {
try {
listenerMutex.wait(2000);
} catch (InterruptedException ex) {
}
}
}
if (transportListener != null) {
if (connectionFailure instanceof IOException) {
transportListener.onException((IOException) connectionFailure);
} else {
transportListener.onException(IOExceptionSupport.create(connectionFailure));
}
}
reconnectMutex.notifyAll();
return false;
}
}
if (!disposed) {
LOG.debug("Waiting " + reconnectDelay + " ms before attempting connection. ");
synchronized (sleepMutex) {
try {
sleepMutex.wait(reconnectDelay);
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
}
}
if (useExponentialBackOff) {
// Exponential increment of reconnect delay.
reconnectDelay *= backOffMultiplier;
if (reconnectDelay > maxReconnectDelay) {
reconnectDelay = maxReconnectDelay;
}
}
}
return !disposed;
}
代码List<URI> connectList = getConnectList();获取配置的broker url，打了断点后debug发现，这边获取的List中的URI有时候已经不是配置中的顺序了，进入getConnectList方法
private List<URI> getConnectList() {
ArrayList<URI> l = new ArrayList<URI>(uris);
boolean removed = false;
if (failedConnectTransportURI != null) {
removed = l.remove(failedConnectTransportURI);
}
if (randomize) {
// Randomly, reorder the list by random swapping
for (int i = 0; i < l.size(); i++) {
int p = (int) (Math.random() * 100 % l.size());
URI t = l.get(p);
l.set(p, l.get(i));
l.set(i, t);
}
}
if (removed) {
l.add(failedConnectTransportURI);
}
LOG.debug("urlList connectionList:" + l);
return l;
}
这边有个randomize，默认值为true，即默认这边会对配置的url重新排序，所以导致连接初始化是可能不是按照想象的从前到后的顺序连接，解决也简单，在连接后价格参数，将randomize设为false
failover:(tcp://master:61616,tcp://slave:61616)?randomize=false
这样每次都会先连接master，master连不上才会连slave。