【Flume】【源码分析】flume中LoadBalancingSinkProcessor负载均衡实现机制的源码分析

基于上一篇文章http://blog.csdn.net/simonchi/article/details/42520193  相对比较细致的分析后，该文章将对LoadBalancingSinkProcessor源码进行选择性的重要逻辑代码进行讲解

首先读取配置，当然是重写congifure方法

 

public void configure(Context context) {
    Preconditions.checkState(getSinks().size() > 1,
        "The LoadBalancingSinkProcessor cannot be used for a single sink. "
        + "Please configure more than one sinks and try again.");

    String selectorTypeName = context.getString(CONFIG_SELECTOR,
        SELECTOR_NAME_ROUND_ROBIN);

    Boolean shouldBackOff = context.getBoolean(CONFIG_BACKOFF, false);

    selector = null;

    if (selectorTypeName.equalsIgnoreCase(SELECTOR_NAME_ROUND_ROBIN)) {
      selector = new RoundRobinSinkSelector(shouldBackOff);
    } else if (selectorTypeName.equalsIgnoreCase(SELECTOR_NAME_RANDOM)) {
      selector = new RandomOrderSinkSelector(shouldBackOff);
    } else {
      try {
        @SuppressWarnings("unchecked")
        Class<? extends SinkSelector> klass = (Class<? extends SinkSelector>)
            Class.forName(selectorTypeName);

        selector = klass.newInstance();
      } catch (Exception ex) {
        throw new FlumeException("Unable to instantiate sink selector: "
            + selectorTypeName, ex);
      }
    }

    selector.setSinks(getSinks());
    selector.configure(
        new Context(context.getSubProperties(CONFIG_SELECTOR_PREFIX)));

    LOGGER.debug("Sink selector: " + selector + " initialized");
  }

该方法最重要的就是后面实例化了selector对象，通过Class.forName去加载该类，就是配置文件中selector对应的值

 

 

selector.setSinks(getSinks());

这行是给该selector设置了sinks，sinks是从配置文件中sinks读取的

 

再来看核心逻辑process()方法

 

public Status process() throws EventDeliveryException {
    Status status = null;

    Iterator<Sink> sinkIterator = selector.createSinkIterator();
    while (sinkIterator.hasNext()) {
      Sink sink = sinkIterator.next();
      try {
        status = sink.process();
        break;
      } catch (Exception ex) {
        selector.informSinkFailed(sink);
        LOGGER.warn("Sink failed to consume event. "
            + "Attempting next sink if available.", ex);
      }
    }

    if (status == null) {
      throw new EventDeliveryException("All configured sinks have failed");
    }

    return status;
  }

首先是创建了一个sink遍历器

 

我们来看下round_robin的实现

 

public Iterator<T> createIterator() {
    List<Integer> activeIndices = getIndexList();
    int size = activeIndices.size();
    // possible that the size has shrunk so gotta adjust nextHead for that
    if (nextHead >= size) {
      nextHead = 0;
    }
    int begin = nextHead++;
    if (nextHead == activeIndices.size()) {
      nextHead = 0;
    }

    int[] indexOrder = new int[size];

    for (int i = 0; i < size; i++) {
      indexOrder[i] = activeIndices.get((begin + i) % size);
    }

    return new SpecificOrderIterator<T>(indexOrder, getObjects());
  }

1、首先获取当前有效的sink
2、顺序指定好

 

 

for (int i = 0; i < size; i++) {
      indexOrder[i] = activeIndices.get((begin + i) % size);
    }

这其实就是round_robin的算法

 

 

while (sinkIterator.hasNext()) {
      Sink sink = sinkIterator.next();
      try {
        status = sink.process();
        break;
      } catch (Exception ex) {
        selector.informSinkFailed(sink);
        LOGGER.warn("Sink failed to consume event. "
            + "Attempting next sink if available.", ex);
      }
    }

循环内部一次选择有效的sink进行处理

 

异常部分，我们发现触发了informSinkFailed()方法，我们来看看该方法

 

public void informFailure(T failedObject) {
    //If there is no backoff this method is a no-op.
    if (!shouldBackOff) {
      return;
    }
    FailureState state = stateMap.get(failedObject);
    long now = System.currentTimeMillis();
    long delta = now - state.lastFail;

    /*
     * When do we increase the backoff period?
     * We basically calculate the time difference between the last failure
     * and the current one. If this failure happened within one hour of the
     * last backoff period getting over, then we increase the timeout,
     * since the object did not recover yet. Else we assume this is a fresh
     * failure and reset the count.
     */
    long lastBackoffLength = Math.min(maxTimeout, 1000 * (1 << state.sequentialFails));
    long allowableDiff = lastBackoffLength + CONSIDER_SEQUENTIAL_RANGE;
    if (allowableDiff > delta) {
      if (state.sequentialFails < EXP_BACKOFF_COUNTER_LIMIT) {
        state.sequentialFails++;
      }
    } else {
      state.sequentialFails = 1;
    }
    state.lastFail = now;
    //Depending on the number of sequential failures this component had, delay
    //its restore time. Each time it fails, delay the restore by 1000 ms,
    //until the maxTimeOut is reached.
    state.restoreTime = now + Math.min(maxTimeout, 1000 * (1 << state.sequentialFails));
  }

实现如上：

 

当然该方法是针对配置中选择惩罚的机制，也就是backoff=true，所以第一行，如果你配置的不选择惩罚，当然就不会执行该方法了；

惩罚机制：

首先选择该失败sink，读取其是否有过失败记录，因为失败的sink有状态记录的，就是代码中那些字段

这些状态的变来变去，读者自行阅读吧，就是个惩罚机制，没什么好说的！！！

 

这样总的看起来，flume内置的两种SinkProcessor其实没什么东西，只要你按照这种结构，就可以开发自己的sinkprocessor了