flume源码学习9-HDFSEventSink实现

HDFSEventSink用于把数据从channel中拿出来（主动pull的形式）然后放到hdfs中，HDFSEventSink在启动时会启动两个线程池callTimeoutPool 和timedRollerPool ，callTimeoutPool 用于运行append/flush等操作hdfs的task（通过callWithTimeout方法调用，并实现timeout功能），用于运行翻转文件的计划任务timedRollerPool：

callTimeoutPool = Executors.newFixedThreadPool(threadsPoolSize,
new ThreadFactoryBuilder().setNameFormat(timeoutName).build());
timedRollerPool = Executors.newScheduledThreadPool(rollTimerPoolSize,
new ThreadFactoryBuilder().setNameFormat(rollerName).build());

channel到sink的操作最终调用了sink的process方法（由SinkProcessor实现类调用），比如HDFSEventSink的process方法,每个process方法中都是一个事务，用来提供原子性操作，process方法调用Channel的take方法从Channel中取出Event，每个transaction中最多的Event数量由hdfs.batchSize设定，默认是100，对每一个Event有如下操作：
1.获取文件的完整路径和名称lookupPath
2.声明一个BucketWriter对象和HDFSWriter 对象，HDFSWriter由hdfs.fileType设定，负责实际数据的写入，BucketWriter可以理解成对hdfs文件和写入方法的封装，每个lookupPath对应一个BucketWriter对象，对应关系写入到sfWriters中（这里sfWriters是一个WriterLinkedHashMap对象，WriterLinkedHashMap是LinkedHashMap的子类（private static class WriterLinkedHashMap extends LinkedHashMap<String, BucketWriter>），用来存放文件到BucketWriter的对应关系，在start方法中初始化：
this.sfWriters = new WriterLinkedHashMap( maxOpenFiles);
长度为hdfs.maxOpenFiles的设置，默认为5000，这个代表最多能打开的文件数量)
3.调用BucketWriter的append方法写入数据
4.当操作的Event数量达到hdfs.batchSize设定后，循环调用每个BucketWriter对象的flush方法，并提交transaction
5.如果出现异常则回滚事务
6.最后关闭transaction
process方法最后返回的是代表Sink状态的Status对象（BACKOFF或者READY），这个可以用于判断Sink的健康状态，比如failover的SinkProcessor就根据这个来判断Sink是否可以提供服务

主要方法分析：
1.构造函数声明一个HDFSWriterFactory对象
在后面会使用HDFSWriterFactory的getWriter方法会根据file类型返回对应的HDFSWriter实现类
2.configure
1)通过configure方法会根据Context设置各种参数项
比如：

inUseSuffix = context.getString( "hdfs.inUseSuffix", defaultInUseSuffix ); //正在写入的文件的后缀名，默认为".tmp"
rollInterval = context.getLong( "hdfs.rollInterval", defaultRollInterval ); //文件翻转时间，默认30
rollSize = context.getLong( "hdfs.rollSize", defaultRollSize ); //文件翻转大小，默认1024
rollCount = context.getLong( "hdfs.rollCount", defaultRollCount ); //默认为10
batchSize = context.getLong( "hdfs.batchSize", defaultBatchSize ); //默认为100
idleTimeout = context.getInteger( "hdfs.idleTimeout", 0); //默认为
String codecName = context.getString( "hdfs.codeC"); //压缩格式
fileType = context.getString( "hdfs.fileType", defaultFileType ); //默认为HDFSWriterFactory.SequenceFileType，即sequencefile
maxOpenFiles = context.getInteger( "hdfs.maxOpenFiles", defaultMaxOpenFiles ); //默认为5000
callTimeout = context.getLong( "hdfs.callTimeout", defaultCallTimeout ); //BucketWriter超时时间，默认为10000
threadsPoolSize = context.getInteger( "hdfs.threadsPoolSize",
defaultThreadPoolSize); //操作append/open/close/flush任务的线程池大小，默认为10
rollTimerPoolSize = context.getInteger( "hdfs.rollTimerPoolSize",
defaultRollTimerPoolSize); //文件翻转计时器线程池大小，默认为1
tryCount = context.getInteger( "hdfs.closeTries", defaultTryCount ); //尝试close文件的此数（大于0）
retryInterval = context.getLong( "hdfs.retryInterval", defaultRetryInterval); //间隔时间（大于0）

2)获取压缩格式

if (codecName == null) { //如果hdfs.codeC没有设置
codeC = null; //则没有压缩功能
compType = CompressionType. NONE;
} else {
codeC = getCodec(codecName);  //调用getCodec方法获取压缩格式
// TODO : set proper compression type
compType = CompressionType. BLOCK; //压缩类型为BLOCK类型
}

3）hdfs文件翻转相关设置，在实例化BucketWriter对象时会用到

needRounding = context.getBoolean( "hdfs.round", false );
if(needRounding) {
String unit = context.getString( "hdfs.roundUnit", "second" );
if (unit.equalsIgnoreCase( "hour")) {
this.roundUnit = Calendar.HOUR_OF_DAY;
} else if (unit.equalsIgnoreCase("minute" )) {
this.roundUnit = Calendar.MINUTE;
} else if (unit.equalsIgnoreCase("second" )){
this.roundUnit = Calendar.SECOND;
} else {
LOG.warn("Rounding unit is not valid, please set one of" +
"minute, hour, or second. Rounding will be disabled" );
needRounding = false ;
}
this. roundValue = context.getInteger("hdfs.roundValue" , 1);
if(roundUnit == Calendar. SECOND || roundUnit == Calendar.MINUTE){
Preconditions. checkArgument(roundValue > 0 && roundValue <= 60,
"Round value" +
"must be > 0 and <= 60");
} else if (roundUnit == Calendar.HOUR_OF_DAY){
Preconditions. checkArgument(roundValue > 0 && roundValue <= 24,
"Round value" +
"must be > 0 and <= 24");
}
}

4）最后初始化一个SinkCounter对象用来记录sink的性能数据

if (sinkCounter == null) {
sinkCounter = new SinkCounter(getName());
}

3.start方法用来启动线程池等

public void start() {
String timeoutName = "hdfs-" + getName() + "-call-runner-%d" ;
callTimeoutPool = Executors. newFixedThreadPool(threadsPoolSize,
new ThreadFactoryBuilder().setNameFormat(timeoutName).build());
String rollerName = "hdfs-" + getName() + "-roll-timer-%d" ;
timedRollerPool = Executors. newScheduledThreadPool(rollTimerPoolSize,
new ThreadFactoryBuilder().setNameFormat(rollerName).build());
this. sfWriters = new WriterLinkedHashMap(maxOpenFiles); //初始化WriterLinkedHashMap对象
sinkCounter.start();
super.start();
}

4.process方法，从channel中pull出数据并发送到hdfs中（每一个transaction中最多可以有batchSize条Event），获取对应的bucket，序列化数据并写入hdfs文件

public Status process() throws EventDeliveryException {
Channel channel = getChannel(); //获取对应的channel
Transaction transaction = channel.getTransaction(); //获取Transaction 对象，提供事务功能
List<BucketWriter> writers = Lists. newArrayList();
transaction.begin(); //事务开始
try {
int txnEventCount = 0;
for (txnEventCount = 0; txnEventCount < batchSize; txnEventCount++) {//这里一个transaction存放的数据最多由hdfs.batchSize指定
Event event = channel.take(); //循环调用Channel的take方法获取Event
if (event == null) {
break;
}
// reconstruct the path name by substituting place holders
String realPath = BucketPath. escapeString(filePath, event.getHeaders(),
timeZone, needRounding, roundUnit , roundValue ); //设置文件路径
String realName = BucketPath. escapeString(fileName, event.getHeaders(),
timeZone, needRounding, roundUnit , roundValue ); //设置文件名称
String lookupPath = realPath + DIRECTORY_DELIMITER + realName; //完整的文件名称
BucketWriter bucketWriter = sfWriters.get(lookupPath);  //根据文件获取对应的BucketWriter对象
// we haven't seen this file yet, so open it and cache the handle
if (bucketWriter == null) {
HDFSWriter hdfsWriter = writerFactory.getWriter(fileType ); //根据文件类型获取HDFSWriter 对象
WriterCallback idleCallback = null;
if(idleTimeout != 0) {
idleCallback = new WriterCallback() {
@Override
public void run(String bucketPath) {
sfWriters.remove(bucketPath); //回调方法
}
};
}
bucketWriter = new BucketWriter(rollInterval , rollSize , rollCount ,
batchSize, context , realPath, realName, inUsePrefix, inUseSuffix,
suffix, codeC, compType, hdfsWriter, timedRollerPool,
proxyTicket, sinkCounter , idleTimeout , idleCallback, lookupPath); //实例化BucketWriter
sfWriters.put(lookupPath, bucketWriter); //这里sfWriters是一个WriterLinkedHashMap对象，WriterLinkedHashMap是LinkedHashMap的子类，用来存放文件到BucketWriter的对应关系，在start方法中初始化：this .sfWriters = new WriterLinkedHashMap(maxOpenFiles);大小为hdfs.maxOpenFiles的设置，默认为5000
}
// track the buckets getting written in this transaction
if (!writers.contains(bucketWriter)) { //List<BucketWriter> writers = Lists.newArrayList();
writers.add(bucketWriter);
}
// Write the data to HDFS
append(bucketWriter, event); //调用append方法写入Event数据
}
if (txnEventCount == 0) {
sinkCounter.incrementBatchEmptyCount();
} else if (txnEventCount == batchSize ) {
sinkCounter.incrementBatchCompleteCount();
} else {
sinkCounter.incrementBatchUnderflowCount();
}
// flush all pending buckets before committing the transaction
for (BucketWriter bucketWriter : writers) {
flush(bucketWriter); //调用flush方法
}
transaction.commit(); //事务提交
if (txnEventCount < 1) {
return Status.BACKOFF ;
} else {
sinkCounter.addToEventDrainSuccessCount(txnEventCount);
return Status.READY ;
}
} catch (IOException eIO) { //如果异常则回滚事务
transaction.rollback();
LOG.warn( "HDFS IO error", eIO);
return Status. BACKOFF;
} catch (Throwable th) {
transaction.rollback();
LOG.error( "process failed", th);
if (th instanceof Error) {
throw (Error) th;
} else {
throw new EventDeliveryException(th);
}
} finally {
transaction.close();
}
}

5.同时定义了几个操作BucketWriter的方法append，flush，close

1） private void append(final BucketWriter bucketWriter, final Event event)
throws IOException, InterruptedException {
// Write the data to HDFS
callWithTimeout(new Callable<Void>() { //注意这里使用callWithTimeout提供了调用的超时功能
public Void call() throws Exception {
bucketWriter.append(event); //调用BucketWriter.append方法写入Event数据
return null;
}
});
}
2）flush-->BucketWriter.flush()
3) close-->BucketWriter.close()