问题背景
对项目中的关键应用调用链日志需要结构化得统一吐出到mongodb中，同时项目中日志输出使用log4j，故准备使用log4j的Appender直接集成mongodb的输出，同时mongodb采用集群模式。因为对于日志有特殊结构化要求，所以没有使用log4mongo库。
mongodb驱动版本：3.5.0
log4j版本：1.2.17
问题现象
error和warn级别的日志输出没有任何问题，但是当输出info级别日志的时候，保存失败，进程被阻塞，过一阵子后控制台报出如下：
Exception in thread "main" com.mongodb.MongoTimeoutException: Timed out after 30000 ms while waiting for a server that matches WritableServerSelector. Client view of cluster state is {type=UNKNOWN, servers=[{address=127.0.0.1:27017, type=UNKNOWN, state=CONNECTING}]
排查过程
-
首先查看mongodb进程和端口全部正常，并且切回warn和error级别再次输出后，一切正常，mongodb数据成功插入，排除了mongodb进程的原因。此步直接确定是log4j与mongodb集成驱动的原因
-
再次执行info级别输出，在进程无响应的时候，通过jstack拿取线程栈，发现尽然后线程阻塞引发的死锁。如下：main线程锁住了0x0000000797b8f4d0，而cluster-ClusterId再等待该锁的释放。
"cluster-ClusterId{value='59a3c74a936270211a375fc6', description='null'}-127.0.0.1:27017" #12 daemon prio=5 os_prio=31 tid=0x00007f97c2263800 nid=0x570f waiting for monitor entry [0x0000700001452000]
java.lang.Thread.State: BLOCKED (on object monitor)
at org.apache.log4j.Category.callAppenders(Category.java:202)
- waiting to lock <0x0000000797b8f4d0> (a org.apache.log4j.spi.RootLogger)
at org.apache.log4j.Category.forcedLog(Category.java:388)
at org.apache.log4j.Category.log(Category.java:853)
at org.slf4j.impl.Log4jLoggerAdapter.info(Log4jLoggerAdapter.java:300)
at com.mongodb.diagnostics.logging.SLF4JLogger.info(SLF4JLogger.java:71)
at com.mongodb.connection.InternalStreamConnection.open(InternalStreamConnection.java:110)
at com.mongodb.connection.DefaultServerMonitor$ServerMonitorRunnable.run(DefaultServerMonitor.java:111)
- locked <0x0000000797b01148> (a com.mongodb.connection.DefaultServerMonitor$ServerMonitorRunnable)
"main" #1 prio=5 os_prio=31 tid=0x00007f97c202f000 nid=0x1003 waiting on condition [0x0000700000182000]
java.lang.Thread.State: TIMED_WAITING (parking)
at sun.misc.Unsafe.park(Native Method)
- parking to wait for <0x0000000797b03588> (a java.util.concurrent.CountDownLatch$Sync)
at java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:215)
at java.util.concurrent.locks.AbstractQueuedSynchronizer.doAcquireSharedNanos(AbstractQueuedSynchronizer.java:1037)
at java.util.concurrent.locks.AbstractQueuedSynchronizer.tryAcquireSharedNanos(AbstractQueuedSynchronizer.java:1328)
at java.util.concurrent.CountDownLatch.await(CountDownLatch.java:277)
at com.mongodb.connection.BaseCluster.selectServer(BaseCluster.java:114)
at com.mongodb.binding.ClusterBinding$ClusterBindingConnectionSource.<init>(ClusterBinding.java:75)
at com.mongodb.binding.ClusterBinding$ClusterBindingConnectionSource.<init>(ClusterBinding.java:71)
at com.mongodb.binding.ClusterBinding.getWriteConnectionSource(ClusterBinding.java:68)
at com.mongodb.operation.OperationHelper.withConnection(OperationHelper.java:413)
at com.mongodb.operation.MixedBulkWriteOperation.execute(MixedBulkWriteOperation.java:168)
at com.mongodb.operation.MixedBulkWriteOperation.execute(MixedBulkWriteOperation.java:74)
at com.mongodb.Mongo.execute(Mongo.java:819)
at com.mongodb.Mongo$2.execute(Mongo.java:802)
at com.mongodb.MongoCollectionImpl.executeSingleWriteRequest(MongoCollectionImpl.java:550)
at com.mongodb.MongoCollectionImpl.insertOne(MongoCollectionImpl.java:317)
at com.mongodb.MongoCollectionImpl.insertOne(MongoCollectionImpl.java:307)
at com.test.mongo.core.MongoLogAgent.saveLog(MongoLogAgent.java:52)
at com.test.log.log4j.MongoAppender.append(MongoAppender.java:86)
at org.apache.log4j.AppenderSkeleton.doAppend(AppenderSkeleton.java:230)
- locked <0x0000000797b8f3e8> (a com.test.log.log4j.MongoAppender)
at org.apache.log4j.helpers.AppenderAttachableImpl.appendLoopOnAppenders(AppenderAttachableImpl.java:65)
at org.apache.log4j.Category.callAppenders(Category.java:203)
- locked <0x0000000797b8f4d0> (a org.apache.log4j.spi.RootLogger)
at org.apache.log4j.Category.forcedLog(Category.java:388)
at org.apache.log4j.Category.log(Category.java:853)
3.测试代码及log4j.properties配置
private final static Logger logger = LoggerFactory.getLogger(Test.class);
public static void main(String[] args) throws Exception {
logger.info("测试",new Exception("出错了"));
}
log4j.rootLogger=info,console,file,mongo
log4j.appender.mongo=com.test.log.log4j.MongoAppender
PS:mongo就是自己实现的Appender，因为没有实质性内容，就不再贴出，主要就是拿到日志后开始写入mongodb
4.既然找到了问题原因，直接看源码分析
log4j的Category对象中，会迭代循环Category及父节点，拿到之后对该对象进行加锁。synchronized(c)就是造成了上述线程栈中main线程（- locked <0x0000000797b8f4d0> (a org.apache.log4j.spi.RootLogger)的原因）
AppenderAttachableImpl aai;
//中间代码省略
public void callAppenders(LoggingEvent event) {
int writes = 0;
for(Category c = this; c != null; c = c.parent) {
synchronized(c) {
if(c.aai != null) {
writes += c.aai.appendLoopOnAppenders(event);
}
if(!c.additive) {
break;
}
}
}
if(writes == 0) {
this.repository.emitNoAppenderWarning(this);
}
}
根据log4j.properties的配置，MongoAppender对象，会在Log4j读取配置文件的时候加入到Category的aai对象中，此处appendLoopOnAppenders方法内就是循环执行console,file,mongo三个Appender的写方法，用来记录日志，其中MongoAppender中通过获取MongoClient进行insertDocument操作。
那cluster-ClusterId线程又是干什么的呢？
这个线程就是mongodb客户端驱动用来启动并且监听Socket链接并且建立连接池的一个独立线程，mongodb的客户端驱动在传入多个mongodb服务端ip并创建MongoClient对象时会构造MultiServerCluster对象及DefaultServer对象，该对象会绑定一个缺省的监听对象DefaultServerMonitor，DefaultServerMonitor对象在构造函数中，会初始化一个线程，这个线程就是真正建立mongodb数据库链接的地方。而DefaultServer对象在构造函数中负责启动该线程。当建立好链接后会将链接放入池中
DefaultServer(final ServerId serverId, final ClusterConnectionMode clusterConnectionMode, final ConnectionPool connectionPool,
final ConnectionFactory connectionFactory, final ServerMonitorFactory serverMonitorFactory,
final ServerListener serverListener, final CommandListener commandListener) {
this.serverListener = notNull("serverListener", serverListener);
this.commandListener = commandListener;
notNull("serverAddress", serverId);
notNull("serverMonitorFactory", serverMonitorFactory);
this.clusterConnectionMode = notNull("clusterConnectionMode", clusterConnectionMode);
this.connectionFactory = notNull("connectionFactory", connectionFactory);
this.connectionPool = notNull("connectionPool", connectionPool);
this.serverStateListener = new DefaultServerStateListener();
this.serverId = serverId;
serverListener.serverOpening(new ServerOpeningEvent(this.serverId));
description = ServerDescription.builder().state(CONNECTING).address(serverId.getAddress()).build();
//构造并启动监控连接线程
serverMonitor = serverMonitorFactory.create(serverStateListener);
serverMonitor.start();
}
//无关代码已注释
class DefaultServerMonitor implements ServerMonitor {
DefaultServerMonitor(final ServerId serverId, final ServerSettings serverSettings,
final ChangeListener<ServerDescription> serverStateListener,
final InternalConnectionFactory internalConnectionFactory, final ConnectionPool connectionPool) {
this.serverSettings = serverSettings;
this.serverId = serverId;
this.serverMonitorListener = getServerMonitorListener(serverSettings);
this.serverStateListener = serverStateListener;
this.internalConnectionFactory = internalConnectionFactory;
this.connectionPool = connectionPool;
monitor = new ServerMonitorRunnable();
monitorThread = new Thread(monitor, "cluster-" + this.serverId.getClusterId() + "-" + this.serverId.getAddress());
monitorThread.setDaemon(true);
isClosed = false;
}
//一个内部类的线程，用于建立连接
class ServerMonitorRunnable implements Runnable {
private final ExponentiallyWeightedMovingAverage averageRoundTripTime = new ExponentiallyWeightedMovingAverage(0.2);
@Override
@SuppressWarnings("unchecked")
public synchronized void run() {
InternalConnection connection = null;
try {
ServerDescription currentServerDescription = getConnectingServerDescription(null);
while (!isClosed) {
ServerDescription previousServerDescription = currentServerDescription;
try {
if (connection == null) {
connection = internalConnectionFactory.create(serverId);
try {
connection.open();
} catch (Throwable t) {
connection = null;
throw t;
}
}
try {
currentServerDescription = lookupServerDescription(connection);
} catch (MongoSocketException e) {
connectionPool.invalidate();
connection.close();
connection = null;
connection = internalConnectionFactory.create(serverId);
try {
connection.open();
} catch (Throwable t) {
connection = null;
throw t;
}
try {
currentServerDescription = lookupServerDescription(connection);
} catch (MongoSocketException e1) {
connection.close();
connection = null;
throw e1;
}
}
} catch (Throwable t) {
averageRoundTripTime.reset();
currentServerDescription = getConnectingServerDescription(t);
}
if (!isClosed) {
try {
logStateChange(previousServerDescription, currentServerDescription);
serverStateListener.stateChanged(new ChangeEvent<ServerDescription>(previousServerDescription,
currentServerDescription));
} catch (Throwable t) {
LOGGER.warn("Exception in monitor thread during notification of server description state change", t);
}
waitForNext();
}
}
} finally {
if (connection != null) {
connection.close();
}
}
}
}
}
可以看到mongodb驱动客户端是通过独立线程来建立连接的，而main线程中是操作的MongoClient对象，当主线程运行较快，MongoClient在执行insert获取不到连接时，会在BaseCluster类的selectServer方法中循环等待是否可以建立连接,若循环多次后时间超过了maxWaitTimeNanos，则直接抛出MongoTimeoutException，这就是为什么造成了超时现象，而mongodb驱动将真实原因屏蔽的地方，这样有个好处就是不会造成线程僵死，影响应用系统，但同时对人造成了误导。
public Server selectServer(final ServerSelector serverSelector) {
isTrue("open", !isClosed());
try {
CountDownLatch currentPhase = phase.get();
ClusterDescription curDescription = description;
ServerSelector compositeServerSelector = getCompositeServerSelector(serverSelector);
Server server = selectRandomServer(compositeServerSelector, curDescription);
boolean selectionFailureLogged = false;
long startTimeNanos = System.nanoTime();
long curTimeNanos = startTimeNanos;
long maxWaitTimeNanos = getMaxWaitTimeNanos();
while (true) {
throwIfIncompatible(curDescription);
if (server != null) {
return server;
}
if (curTimeNanos - startTimeNanos > maxWaitTimeNanos) {
throw createTimeoutException(serverSelector, curDescription);
}
if (!selectionFailureLogged) {
logServerSelectionFailure(serverSelector, curDescription);
selectionFailureLogged = true;
}
connect();
currentPhase.await(Math.min(maxWaitTimeNanos - (curTimeNanos - startTimeNanos), getMinWaitTimeNanos()), NANOSECONDS);
curTimeNanos = System.nanoTime();
currentPhase = phase.get();
curDescription = description;
server = selectRandomServer(compositeServerSelector, curDescription);
}
} catch (InterruptedException e) {
throw new MongoInterruptedException(format("Interrupted while waiting for a server that matches %s", serverSelector), e);
}
}
死锁的产生
因为在log4j.propertis配置中，root上配置了所有的Appender。通过线程dump的分析和源码的解读，很容易发现是因为main线程在执行mongodb的Appender是，获取到的是rootLogger的锁，同时，BaseCluster类中每一次获取链接时利用了CountDownLatch方式将线程挂起，导致了主线程的等待，但此时main线程的锁并没有释放。而cluster-ClusterId线程中执行的mongodb代码中，通过slf4j也有Info级别的日志输出，而此时在执行Appender时，获取到的还是rootLogger对象，并又去申请该锁，因为是两个不同的线程，造成了cluster-ClusterId线程的阻塞
问题的解决
将mongodb包中的logger对象指向变为不是rootLogger即可，这样，cluster-ClusterId线程在获取Category对象锁时，与main线程已经申请到的对象锁不是同一个即可。
log4j.rootLogger=debug,console,file,mongo
#此配置是让org.mongodb的logger不继承父对象级别，即root，
log4j.additivity.org.mongodb=false
log4j.logger.org.mongodb=info,console
通过以上配置，所有org.mongodb包的日志，都会使用一个新的Logger对象，该对象只绑定了console一个Appender，此Logger对象的parent节点是rootLogger对象
AppenderAttachableImpl aai;
//中间代码省略
public void callAppenders(LoggingEvent event) {
int writes = 0;
for(Category c = this; c != null; c = c.parent) {
//cluster-ClusterId线程中，此时加锁对象即为新的logger对象
synchronized(c) {
if(c.aai != null) {
writes += c.aai.appendLoopOnAppenders(event);
}
if(!c.additive) {
break;
}
}
}
if(writes == 0) {
this.repository.emitNoAppenderWarning(this);
}
}
总结
网上也发现有log4j的死锁问题，其实跟多线程并发没什么关系，主要是一个线程在输出日志时，启动了另一个线程，但此时主线程又不释放锁，而子线程又去申请该锁导致的。
此时若是同一个线程，在synchronized(c)时，直接就是重入锁，不会产生死锁阻塞的问题。
250
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
