Flume NG源码分析（五）使用ThriftSource通过RPC方式收集日志

上一篇说了利用ExecSource从本地日志文件异步的收集日志，这篇说说采用RPC方式同步收集日志的方式。笔者对Thrift比较熟悉，所以用ThriftSource来介绍RPC的日志收集方式。

整体的结构图如下：

1. ThriftSource包含了一个Thrift Server，以及一个Thrift Service服务的实现。这里的Thrift Service是由ThriftSourceProtocol定义

2. 应用程序调用Thrift Service的客户端，以RPC的方式将日志发送到ThriftSource的Thrfit Service的服务端，完成应用层序日志的收集

先来看下ThriftSourceProtocol定义的Thrfit服务。Thrift服务定义在flume-ng-sdk工程的flume.thrift中

1. 定义了ThriftFlumeEvent数据结构，日志封装成Event来Flume NG中传递

2. 定义了ThriftSourceProtocol服务，有两个接口，append和appendBatch

namespace java org.apache.flume.thrift

struct ThriftFlumeEvent {
  1: required map  headers,
  2: required binary body,
}

enum Status {
  OK,
  FAILED,
  ERROR,
  UNKNOWN
}

service ThriftSourceProtocol {
  Status append(1: ThriftFlumeEvent event),
  Status appendBatch(1: list events),
}

Thrift生成的中间文件是ThrfitSourceProtocol，由服务器端和客户端共享

public class ThriftSourceProtocol {

  public interface Iface {

    public Status append(ThriftFlumeEvent event) throws org.apache.thrift.TException;

    public Status appendBatch(List events) throws org.apache.thrift.TException;

  }

。。。。。。
}

ThrfitSource中的ThriftSourceHandler私有类实现了ThrfitSourceProtocol这个服务，append接口的实现逻辑如下

1. 把ThriftFlumeEvent转化成SimpleEvent

2. 修改计数器

3. 把SimpleEvent交给ChannelProcessor来处理，传递到下游的Channel中去

可以看到ThriftSouceHandler的实现逻辑和ExecRunnable的逻辑基本是一样的

private class ThriftSourceHandler implements ThriftSourceProtocol.Iface {

    @Override
    public Status append(ThriftFlumeEvent event) throws TException {
      Event flumeEvent = EventBuilder.withBody(event.getBody(),
        event.getHeaders());

      sourceCounter.incrementAppendReceivedCount();
      sourceCounter.incrementEventReceivedCount();

      try {
        getChannelProcessor().processEvent(flumeEvent);
      } catch (ChannelException ex) {
        logger.warn("Thrift source " + getName() + " could not append events " +
          "to the channel.", ex);
        return Status.FAILED;
      }
      sourceCounter.incrementAppendAcceptedCount();
      sourceCounter.incrementEventAcceptedCount();
      return Status.OK;
    }

    @Override
    public Status appendBatch(List events) throws TException {
      sourceCounter.incrementAppendBatchReceivedCount();
      sourceCounter.addToEventReceivedCount(events.size());

      List flumeEvents = Lists.newArrayList();
      for(ThriftFlumeEvent event : events) {
        flumeEvents.add(EventBuilder.withBody(event.getBody(),
          event.getHeaders()));
      }

      try {
        getChannelProcessor().processEventBatch(flumeEvents);
      } catch (ChannelException ex) {
        logger.warn("Thrift source %s could not append events to the " +
          "channel.", getName());
        return Status.FAILED;
      }

      sourceCounter.incrementAppendBatchAcceptedCount();
      sourceCounter.addToEventAcceptedCount(events.size());
      return Status.OK;
    }
  }

}

有了Thrfit服务实现后，ThrfitSource定义了Thrfit Server。默认是TThreadedSelectorServer，当TThreadedSelectorServer ClassNotFound后，创建TThreadPoolServer，还是没找到的话，那么ThriftSource启动失败。

关于Thrfit Server的更多介绍可以看这篇Thrift源码分析（七）-- TServer服务器分析

Thrift Server的创建和启动主要做了几件事情

1. 创建ServerSocket，这里是TNonblockingServerSocket，非阻塞的ServerSocket

2. 创建服务器参数类TNonblockingServer.AbstractNonblockingServerArgs，所有的服务器的属性设置都是在Args类里传递的

3. TThreadedSelectorServer是一个Reactor模式的服务器实现，需要传递一个线程池。这里是Executors.newFixedThreadPool(maxThreads, threadFactory);

4. 设置编解码协议，这里是TFastFramedTransport协议

5. 设置Thrift服务的实现类Processor，这里是上面定义的ThrfitSourceHandler类

6. 启动Thrift服务器，这里在单独的线程中启动了Thrift服务器。servingExecutor.submit(new Runnable() {public void run() {server.serve();}})

在单独的线程启动Thrift服务器主要的目的是在原来的线程中可以处理一下Thrfit服务器停止后的清理工作。

Class serverClass = null;
    Class argsClass = null;
    TServer.AbstractServerArgs args = null;
    /*
     * Use reflection to determine if TThreadedSelectServer is available. If
     * it is not available, use TThreadPoolServer
     */
    try {
      serverClass = Class.forName("org.apache.thrift" +
        ".server.TThreadedSelectorServer");

      argsClass = Class.forName("org.apache.thrift" +
        ".server.TThreadedSelectorServer$Args");

      // Looks like TThreadedSelectorServer is available, so continue..
      ExecutorService sourceService;
      ThreadFactory threadFactory = new ThreadFactoryBuilder().setNameFormat(
        "Flume Thrift IPC Thread %d").build();
      if (maxThreads == 0) {
        sourceService = Executors.newCachedThreadPool(threadFactory);
      } else {
        sourceService = Executors.newFixedThreadPool(maxThreads, threadFactory);
      }
      serverTransport = new TNonblockingServerSocket(
        new InetSocketAddress(bindAddress, port));
      args = (TNonblockingServer.AbstractNonblockingServerArgs) argsClass
        .getConstructor(TNonblockingServerTransport.class)
        .newInstance(serverTransport);
      Method m = argsClass.getDeclaredMethod("executorService",
        ExecutorService.class);
      m.invoke(args, sourceService);
    }


 try {

      args.protocolFactory(new TCompactProtocol.Factory());
      args.inputTransportFactory(new TFastFramedTransport.Factory());
      args.outputTransportFactory(new TFastFramedTransport.Factory());
      args.processor(new ThriftSourceProtocol
        .Processor(new ThriftSourceHandler()));
     
      server = (TServer) serverClass.getConstructor(argsClass).newInstance
        (args);
    } catch (Throwable ex) {
      throw new FlumeException("Cannot start Thrift Source.", ex);
    }


    servingExecutor = Executors.newSingleThreadExecutor(new
      ThreadFactoryBuilder().setNameFormat("Flume Thrift Source I/O Boss")
      .build());

    /**
     * Start serving.
     */
    servingExecutor.submit(new Runnable() {
      @Override
      public void run() {
        server.serve();
      }
    });

总结一下，采用RPC的方式收集日志有几个步骤

1. 定义RPC服务来收集日志

2. 实现RPC服务，并提供客户端给应用程序。应用程序使用客户端来将日志封装成Event，通过RPC调用传递给RPC类型的Source

3. RPC类型的Source启动RPC Server，提供RPC服务，将接收到的Event传递给下游的Channel