Flume架构与源码分析-MemoryChannel事务实现
Flume提供了可靠地日志采集功能,其高可靠是通过事务机制实现的。而对于Channel的事务我们本部分会介绍MemoryChannel和FileChannel的实现。
首先我们看下BasicChannelSemantics实现:
public abstract class BasicChannelSemantics extends AbstractChannel { //1、事务使用ThreadLocal存储,保证事务线程安全 private ThreadLocal<BasicTransactionSemantics> currentTransaction = new ThreadLocal<BasicTransactionSemantics>(); private boolean initialized = false; //2、进行一些初始化工作 protected void initialize() {} //3、提供给实现类的创建事务的回调 protected abstract BasicTransactionSemantics createTransaction(); //4、往Channel放Event,其直接委托给事务的put方法实现 @Override public void put(Event event) throws ChannelException { BasicTransactionSemantics transaction = currentTransaction.get(); Preconditions.checkState(transaction != null, "No transaction exists for this thread"); transaction.put(event); } //5、从Channel获取Event,也是直接委托给事务的take方法实现 @Override public Event take() throws ChannelException { BasicTransactionSemantics transaction = currentTransaction.get(); Preconditions.checkState(transaction != null, "No transaction exists for this thread"); return transaction.take(); } //6、获取事务,如果本实例没有初始化则先初始化;否则先从ThreadLocal获取事务,如果没有或者关闭了则创建一个并绑定到ThreadLocal。 @Override public Transaction getTransaction() { if (!initialized) { synchronized (this) { if (!initialized) { initialize(); initialized = true; } } } BasicTransactionSemantics transaction = currentTransaction.get(); if (transaction == null || transaction.getState().equals( BasicTransactionSemantics.State.CLOSED)) { transaction = createTransaction(); currentTransaction.set(transaction); } return transaction; } }
MemoryChannel事务实现
首先我们来看下MemoryChannel的实现,其是一个纯内存的Channel实现,整个事务操作都是在内存中完成。首先看下其内存结构:
1、首先由一个Channel Queue用于存储整个Channel的Event数据;
2、每个事务都有一个Take Queue和Put Queue分别用于存储事务相关的取数据和放数据,等事务提交时才完全同步到Channel Queue,或者失败把取数据回滚到Channel Queue。
MemoryChannel时设计时考虑了两个容量:Channel Queue容量和事务容量,而这两个容量涉及到了数量容量和字节数容量。
另外因为多个事务要操作Channel Queue,还要考虑Channel Queue的动态扩容问题,因此MemoryChannel使用了锁来实现;而容量问题则使用了信号量来实现。
在configure方法中进行了一些参数的初始化,如容量、Channel Queue等。首先看下Channel Queue的容量是如何计算的:
try { capacity = context.getInteger("capacity", defaultCapacity); } catch(NumberFormatException e) { capacity = defaultCapacity; } if (capacity <= 0) { capacity = defaultCapacity; }
即首先从配置文件读取数量容量,如果没有配置则是默认容量(默认100),而配置的容量小于等于0,则也是默认容量。
接下来是初始化事务数量容量:
try { transCapacity = context.getInteger("transactionCapacity", defaultTransCapacity); } catch(NumberFormatException e) { transCapacity = defaultTransCapacity; } if (transCapacity <= 0) { transCapacity = defaultTransCapacity; } Preconditions.checkState(transCapacity <= capacity, "Transaction Capacity of Memory Channel cannot be higher than " + "the capacity.");
整个过程和Channel Queue数量容量初始化类似,但是最后做了前置条件判断,事务容量必须小于等于Channel Queue容量。
接下来是字节容量限制:
try { byteCapacityBufferPercentage = context.getInteger("byteCapacityBufferPercentage", defaultByteCapacityBufferPercentage); } catch(NumberFormatException e) { byteCapacityBufferPercentage = defaultByteCapacityBufferPercentage; } try { byteCapacity = (int)((context.getLong("byteCapacity", defaultByteCapacity).longValue() * (1 - byteCapacityBufferPercentage * .01 )) /byteCapacitySlotSize); if (byteCapacity < 1) { byteCapacity = Integer.MAX_VALUE; } } catch(NumberFormatException e) { byteCapacity = (int)((defaultByteCapacity * (1 - byteCapacityBufferPercentage * .01 )) /byteCapacitySlotSize); }
byteCapacityBufferPercentage:用来确定byteCapacity的一个百分比参数,即我们定义的字节容量和实际事件容量的百分比,因为我们定义的字节容量主要考虑Event body,而忽略Event header,因此需要减去Event header部分的内存占用,可以认为该参数定义了Event header占了实际字节容量的百分比,默认20%;
byteCapacity:首先读取配置文件定义的byteCapacity,如果没有定义,则使用默认defaultByteCapacity,而defaultByteCapacity默认是JVM物理内存的80%(Runtime.getRuntime().maxMemory() * .80);那么实际byteCapacity=定义的byteCapacity * (1- Event header百分比)/ byteCapacitySlotSize;byteCapacitySlotSize默认100,即计算百分比的一个系数。
接下来定义keepAlive参数:
try { keepAlive = context.getInteger("keep-alive", defaultKeepAlive); } catch(NumberFormatException e) { keepAlive = defaultKeepAlive; }
keepAlive定义了操作Channel Queue的等待超时事件,默认3s。
接着初始化Channel Queue:
if(queue != null) { try { resizeQueue(capacity); } catch (InterruptedException e) { Thread.currentThread().interrupt(); } } else { synchronized(queueLock) { queue = new LinkedBlockingDeque<Event>(capacity); queueRemaining = new Semaphore(capacity); queueStored = new Semaphore(0); } }
首先如果Channel Queue不为null,表示动态扩容;否则进行Channel Queue的创建。
首先看下首次创建Channel Queue,首先使用queueLock锁定,即在操作Channel Queue时都需要锁定,因为之前说过Channel Queue可能动态扩容,然后初始化信号量:Channel Queue剩余容量和向Channel Queue申请存储的容量,用于事务操作中预占Channel Queue容量。
接着是调用resizeQueue动态扩容:
private void resizeQueue(int capacity) throws InterruptedException { int oldCapacity; synchronized(queueLock) { //首先计算扩容前的Channel Queue的容量 oldCapacity = queue.size() + queue.remainingCapacity(); } if(oldCapacity == capacity) {//如果新容量和老容量相等,不需要扩容 return; } else if (oldCapacity > capacity) {//如果老容量大于新容量,缩容 //首先要预占老容量-新容量的大小,以便缩容容量 if(!queueRemaining.tryAcquire(oldCapacity - capacity, keepAlive, TimeUnit.SECONDS)) { //如果获取失败,默认是记录日志然后忽略 } else { //否则,直接缩容,然后复制老Queue的数据,缩容时需要锁定queueLock,因为这一系列操作要线程安全 synchronized(queueLock) { LinkedBlockingDeque<Event> newQueue = new LinkedBlockingDeque<Event>(capacity); newQueue.addAll(queue); queue = newQueue; } } } else { //如果不是缩容,则直接扩容即可 synchronized(queueLock) { LinkedBlockingDeque<Event> newQueue = new LinkedBlockingDeque<Event>(capacity); newQueue.addAll(queue); queue = newQueue; } //增加/减少Channel Queue的新的容量 queueRemaining.release(capacity - oldCapacity); } } 到此,整个Channel Queue相关的数据初始化完毕,接着会调用start方法进行初始化: public synchronized void start() { channelCounter.start(); channelCounter.setChannelSize(queue.size()); channelCounter.setChannelCapacity(Long.valueOf( queue.size() + queue.remainingCapacity())); super.start(); }
此处初始化了一个ChannelCounter,是一个计数器,记录如当前队列放入Event数、取出Event数、成功数等。
之前已经分析了大部分Channel会把put和take直接委托给事务去完成,因此接下来看下MemoryTransaction的实现。
首先看下MemoryTransaction的初始化:
private class MemoryTransaction extends BasicTransactionSemantics { private LinkedBlockingDeque<Event> takeList; private LinkedBlockingDeque<Event> putList; private final ChannelCounter channelCounter; private int putByteCounter = 0; private int takeByteCounter = 0; public MemoryTransaction(int transCapacity, ChannelCounter counter) { putList = new LinkedBlockingDeque<Event>(transCapacity); takeList = new LinkedBlockingDeque<Event>(transCapacity); channelCounter = counter; }
可以看出MemoryTransaction涉及到两个事务容量大小定义的队列(链表阻塞队列)、队列字节计数器、另外一个是Channel操作的计数器。
事务中的放入操作如下:
protected void doPut(Event event) throws InterruptedException { //1、增加放入事件计数器 channelCounter.incrementEventPutAttemptCount(); //2、estimateEventSize计算当前Event body大小 int eventByteSize = (int)Math.ceil(estimateEventSize(event)/byteCapacitySlotSize); //3、往事务队列的putList中放入Event,如果满了,则抛异常回滚事务 if (!putList.offer(event)) { throw new ChannelException( "Put queue for MemoryTransaction of capacity " + putList.size() + " full, consider committing more frequently, " + "increasing capacity or increasing thread count"); } //4、增加放入队列字节数计数器 putByteCounter += eventByteSize; }
整个doPut操作相对来说比较简单,就是往事务putList队列放入Event,如果满了则直接抛异常回滚事务;否则放入putList暂存,等事务提交时转移到Channel Queue。另外需要增加放入队列的字节数计数器,以便之后做字节容量限制。
接下来是事务中的取出操作:
protected Event doTake() throws InterruptedException { //1、增加取出事件计数器 channelCounter.incrementEventTakeAttemptCount(); //2、如果takeList队列没有剩余容量,即当前事务已经消费了最大容量的Event if(takeList.remainingCapacity() == 0) { throw new ChannelException("Take list for MemoryTransaction, capacity " + takeList.size() + " full, consider committing more frequently, " + "increasing capacity, or increasing thread count"); } //3、queueStored试图获取一个信号量,超时直接返回null if(!queueStored.tryAcquire(keepAlive, TimeUnit.SECONDS)) { return null; } //4、从Channel Queue获取一个Event Event event; synchronized(queueLock) {//对Channel Queue的操作必须加queueLock,因为之前说的动态扩容问题 event = queue.poll(); } //5、因为信号量的保证,Channel Queue不应该返回null,出现了就不正常了 Preconditions.checkNotNull(event, "Queue.poll returned NULL despite semaphore " + "signalling existence of entry"); //6、暂存到事务的takeList队列 takeList.put(event); //7、计算当前Event body大小并增加取出队列字节数计数器 int eventByteSize = (int)Math.ceil(estimateEventSize(event)/byteCapacitySlotSize); takeByteCounter += eventByteSize; return event; }
接下来是提交事务:
protected void doCommit() throws InterruptedException { //1、计算改变的Event数量,即取出数量-放入数量;如果放入的多,那么改变的Event数量将是负数 int remainingChange = takeList.size() - putList.size(); //2、 如果remainingChange小于0,则需要获取Channel Queue剩余容量的信号量 if(remainingChange < 0) { //2.1、首先获取putByteCounter个字节容量信号量,如果失败说明超过字节容量限制了,回滚事务 if(!bytesRemaining.tryAcquire(putByteCounter, keepAlive, TimeUnit.SECONDS)) { throw new ChannelException("Cannot commit transaction. Byte capacity " + "allocated to store event body " + byteCapacity * byteCapacitySlotSize + "reached. Please increase heap space/byte capacity allocated to " + "the channel as the sinks may not be keeping up with the sources"); } //2.2、获取Channel Queue的-remainingChange个信号量用于放入-remainingChange个Event,如果获取不到,则释放putByteCounter个字节容量信号量,并抛出异常回滚事务 if(!queueRemaining.tryAcquire(-remainingChange, keepAlive, TimeUnit.SECONDS)) { bytesRemaining.release(putByteCounter); throw new ChannelFullException("Space for commit to queue couldn't be acquired." + " Sinks are likely not keeping up with sources, or the buffer size is too tight"); } } int puts = putList.size(); int takes = takeList.size(); synchronized(queueLock) {//操作Channel Queue时一定要锁定queueLock if(puts > 0 ) { while(!putList.isEmpty()) { //3.1、如果有Event,则循环放入Channel Queue if(!queue.offer(putList.removeFirst())) { //3.2、如果放入Channel Queue失败了,说明信号量控制出问题了,这种情况不应该发生 throw new RuntimeException("Queue add failed, this shouldn't be able to happen"); } } } //4、操作成功后,清空putList和takeList队列 putList.clear(); takeList.clear(); } //5.1、释放takeByteCounter个字节容量信号量 bytesRemaining.release(takeByteCounter); //5.2、重置字节计数器 takeByteCounter = 0; putByteCounter = 0; //5.3、释放puts个queueStored信号量,这样doTake方法就可以获取数据了 queueStored.release(puts); //5.4、释放remainingChange个queueRemaining信号量 if(remainingChange > 0) { queueRemaining.release(remainingChange); } //6、ChannelCounter一些数据计数 if (puts > 0) { channelCounter.addToEventPutSuccessCount(puts); } if (takes > 0) { channelCounter.addToEventTakeSuccessCount(takes); } channelCounter.setChannelSize(queue.size()); }
此处涉及到两个信号量:
queueStored表示Channel Queue已存储事件容量(已存储的事件数量),队列取出事件时-1,放入事件成功时+N,取出失败时-N,即Channel Queue存储了多少事件。queueStored信号量默认为0。当doTake取出Event时减少一个queueStored信号量,当doCommit提交事务时需要增加putList 队列大小的queueStored信号量,当doRollback回滚事务时需要减少takeList队列大小的queueStored信号量。
queueRemaining表示Channel Queue可存储事件容量(可存储的事件数量),取出事件成功时+N,放入事件成功时-N。queueRemaining信号量默认为Channel Queue容量。其在提交事务时首先通过remainingChange = takeList.size() - putList.size()计算获得需要增加多少变更事件;如果小于0表示放入的事件比取出的多,表示有- remainingChange个事件放入,此时应该减少-queueRemaining信号量;而如果大于0,则表示取出的事件比放入的多,表示有queueRemaining个事件取出,此时应该增加queueRemaining信号量;即消费事件时减少信号量,生产事件时增加信号量。
而bytesRemaining是字节容量信号量,超出容量则回滚事务。
最后看下回滚事务:
protected void doRollback() { int takes = takeList.size(); synchronized(queueLock) { //操作Channel Queue时一定锁住queueLock //1、前置条件判断,检查是否有足够容量回滚事务 Preconditions.checkState(queue.remainingCapacity() >= takeList.size(), "Not enough space in memory channel " + "queue to rollback takes. This should never happen, please report"); //2、回滚事务的takeList队列到Channel Queue while(!takeList.isEmpty()) { queue.addFirst(takeList.removeLast()); } putList.clear(); } //3、释放putByteCounter个bytesRemaining信号量 bytesRemaining.release(putByteCounter); //4、计数器重置 putByteCounter = 0; takeByteCounter = 0; //5、释放takeList队列大小个已存储事件容量 queueStored.release(takes); channelCounter.setChannelSize(queue.size()); } }
也就是说在回滚时,需要把takeList中暂存的事件回滚到Channel Queue,并回滚queueStored信号量。