Zookeeper事务日志预分配空间解读

前言

Zookeeper的通过快照日志和事务日志将内存信息保存下来，记录下来每次请求的具体信息。

尤其是其事务日志，每次处理事务请求时都需要将其记录下来。

Zookeeper事务日志的默认存储方式是磁盘文件，那么Zookeeper的总体性能就受限与磁盘文件的写入速度。

针对这个瓶颈，Zookeeper做了什么优化操作呢，本文我们就一起来了解下。

1.事务日志的预分配

事务日志的添加，我们需要从FileTxnLog.append()方法看起

public class FileTxnLog implements TxnLog {
volatile BufferedOutputStream logStream = null;
volatile OutputArchive oa;
volatile FileOutputStream fos = null;
// 追加事务日志
public synchronized boolean append(TxnHeader hdr, Record txn)
throws IOException
{
if (hdr == null) {
return false;
}
if (hdr.getZxid() <= lastZxidSeen) {
LOG.warn("Current zxid " + hdr.getZxid()
+ " is <= " + lastZxidSeen + " for "
+ hdr.getType());
} else {
lastZxidSeen = hdr.getZxid();
}
// 默认logStream为空
if (logStream==null) {
if(LOG.isInfoEnabled()){
LOG.info("Creating new log file: " + Util.makeLogName(hdr.getZxid()));
}
// 以下代码为创建事务日志文件
// 根据当前事务ID来创建具体文件名，并写入文件头信息
logFileWrite = new File(logDir, Util.makeLogName(hdr.getZxid()));
fos = new FileOutputStream(logFileWrite);
logStream=new BufferedOutputStream(fos);
oa = BinaryOutputArchive.getArchive(logStream);
FileHeader fhdr = new FileHeader(TXNLOG_MAGIC,VERSION, dbId);
fhdr.serialize(oa, "fileheader");
// Make sure that the magic number is written before padding.
logStream.flush();
filePadding.setCurrentSize(fos.getChannel().position());
streamsToFlush.add(fos);
}
// 预分配代码在这里
filePadding.padFile(fos.getChannel());
byte[] buf = Util.marshallTxnEntry(hdr, txn);
if (buf == null || buf.length == 0) {
throw new IOException("Faulty serialization for header " +
"and txn");
}
Checksum crc = makeChecksumAlgorithm();
crc.update(buf, 0, buf.length);
oa.writeLong(crc.getValue(), "txnEntryCRC");
Util.writeTxnBytes(oa, buf);
return true;
}
}

创建FileTxnLog对象时，其logStream属性为null，所以当第一次处理事务请求时，会先根据当前事务ID来创建一个文件。

1.1 事务日志预分配

public class FilePadding {
long padFile(FileChannel fileChannel) throws IOException {
// 针对新文件而言，newFileSize=64M
long newFileSize = calculateFileSizeWithPadding(fileChannel.position(), currentSize, preAllocSize);
if (currentSize != newFileSize) {
// 将文件扩充到64M，全部用0来填充
fileChannel.write((ByteBuffer) fill.position(0), newFileSize - fill.remaining());
currentSize = newFileSize;
}
return currentSize;
}
// size计算
public static long calculateFileSizeWithPadding(long position, long fileSize, long preAllocSize) {
// If preAllocSize is positive and we are within 4KB of the known end of the file calculate a new file size
// 初始时候position=0，fileSize为0，preAllocSize为系统参数执行，默认为64M
if (preAllocSize > 0 && position + 4096 >= fileSize) {
// If we have written more than we have previously preallocated we need to make sure the new
// file size is larger than what we already have
// Q:这里确实没看懂...
if (position > fileSize) {
fileSize = position + preAllocSize;
fileSize -= fileSize % preAllocSize;
} else {
fileSize += preAllocSize;
}
}
return fileSize;
}
}

预分配的过程比较简单，就是看下当前文件的剩余空间是否<4096，如果是，则扩容。

Q：

这里有一个不太明白的问题，position > fileSize的场景是怎样的呢？

2.创建新的事务日志文件时机

通过上述代码分析我们知道，当logStream=null时，就会创建一个新的事务日志文件，那么logStream对象什么时候为空呢？

搜索代码，只看到FileTxnLog.rollLog()方法会主动将logStream设置为null

public class FileTxnLog implements TxnLog {
public synchronized void rollLog() throws IOException {
if (logStream != null) {
this.logStream.flush();
this.logStream = null;
oa = null;
}
}
}

那么根据这个线索，我们来搜索下rollLog的调用链

SyncRequestProcessor.run() -> ZKDatabase.rollLog() -> FileTxnSnapLog.rollLog() -> FileTxnLog.rollLog()

最终看到是在SyncRequestProcessor.run()方法中发起调用的，而且只有这一条调用链，我们来分析下

2.1 SyncRequestProcessor.run()

public class SyncRequestProcessor extends ZooKeeperCriticalThread implements RequestProcessor {
public void run() {
try {
int logCount = 0;
setRandRoll(r.nextInt(snapCount/2));
while (true) {
...
if (si != null) {
// 追加事务日志
if (zks.getZKDatabase().append(si)) {
logCount++;
if (logCount > (snapCount / 2 + randRoll)) {
setRandRoll(r.nextInt(snapCount/2));
// 注意：在这里发起了rollLog
zks.getZKDatabase().rollLog();
...
}
} else if (toFlush.isEmpty()) {
...
}
toFlush.add(si);
if (toFlush.size() > 1000) {
flush(toFlush);
}
}
}
} catch (Throwable t) {
handleException(this.getName(), t);
running = false;
}
LOG.info("SyncRequestProcessor exited!");
}
}