hadoop 2.4 namenode ha 源码分析(ActiveStandbyElector)

2019独角兽企业重金招聘Python工程师标准>>>

在hadoop nn的HA，对于主备节点的选举，是通过ActiveStandbyElector来实现的。源码上有针对该类的解释。

小弟英文不才，翻译一下。该类主要使用了zookeeper实现了主节点的选举，对于成功选举的主节点，会在zookeeper上创建零时节点。如果创建成功，NN会变成active，而其余nn节点会成备用节点。

下面还是来具体分析一下ActiveStandbyElector类的作用，ActiveStandbyElector主要实现了选举，选举流程主要是通过创建零时节点的方式实现，如果创建成功。可以认为是获取到对应的LOCK，该节点可以成为active。如果没有成功创建该节点，可以认为为standby节点，对于standby节点，需要一直监听该LOCK节点的状态。如果发生节点的事件，就去尝试选举。基本流程就是这样。

下面，来看一下ActiveStandbyElector类的主要方法和流程。对于熟悉zookeeper的同学来说，zookeeper的必须要实现watcher接口，其中可以实现自己的各种事件的处理逻辑。

在ActiveStandbyElector中，采用了

内部类来实现Watcher接口，其process方法，调用了processWatchEvent来实现具体的业务处理。

下面来分析该processWatchEvent的具体逻辑：

 //处理zk的事件

synchronized void processWatchEvent(ZooKeeper zk, WatchedEvent event) {
    Event.EventType eventType = event.getType();
    if (isStaleClient(zk)) return;
    LOG.debug("Watcher event type: " + eventType + " with state:"
        + event.getState() + " for path:" + event.getPath()
        + " connectionState: " + zkConnectionState
        + " for " + this);

    if (eventType == Event.EventType.None) {
    	//会话本身的时间，如连接。失去连接。
      // the connection state has changed
      switch (event.getState()) {
      case SyncConnected:
        LOG.info("Session connected.");
        // if the listener was asked to move to safe state then it needs to
        // be undone
        ConnectionState prevConnectionState = zkConnectionState;
        zkConnectionState = ConnectionState.CONNECTED;
        if (prevConnectionState == ConnectionState.DISCONNECTED &&
            wantToBeInElection) {
          monitorActiveStatus();//监控节点
        }
        break;
      case Disconnected:
        LOG.info("Session disconnected. Entering neutral mode...");

        // ask the app to move to safe state because zookeeper connection
        // is not active and we dont know our state
        zkConnectionState = ConnectionState.DISCONNECTED;
        enterNeutralMode();
        break;
      case Expired:
        // the connection got terminated because of session timeout
        // call listener to reconnect
        LOG.info("Session expired. Entering neutral mode and rejoining...");
        enterNeutralMode();
        reJoinElection(0);//参与选举
        break;
      case SaslAuthenticated:
        LOG.info("Successfully authenticated to ZooKeeper using SASL.");
        break;
      default:
        fatalError("Unexpected Zookeeper watch event state: "
            + event.getState());
        break;
      }

      return;
    }
    
    // a watch on lock path in zookeeper has fired. so something has changed on
    // the lock. ideally we should check that the path is the same as the lock
    // path but trusting zookeeper for now
    //节点事件
    String path = event.getPath();
    if (path != null) {
      switch (eventType) {
      case NodeDeleted:
        if (state == State.ACTIVE) {
          enterNeutralMode();//该方法目前未实现
        }
        joinElectionInternal();//开始选举
        break;
      case NodeDataChanged:
        monitorActiveStatus();//继续监控该节点，尝试成为active
        break;
      default:
        LOG.debug("Unexpected node event: " + eventType + " for path: " + path);
        monitorActiveStatus();
      }

      return;
    }

    // some unexpected error has occurred
    fatalError("Unexpected watch error from Zookeeper");
  }

而joinElectionInternal，选举的核心方法就是，

选举就是通过对zkLokFilePath节点的创建，来完成。这个采用了zk的异步回调。

从该类的定义，可以看出，本身就是实现了zk的两个接口。

StatCallback需要实现的方法，如下：

对于两个方法的实现，ActiveStandbyElector内部实现几乎是一样的。这里不再贴上源码，有兴趣的可以自己去看源码。

贴上实现方法，有注释。呵呵

public synchronized void processResult(int rc, String path, Object ctx,
      String name) {
    if (isStaleClient(ctx)) return;
    LOG.debug("CreateNode result: " + rc + " for path: " + path
        + " connectionState: " + zkConnectionState +
        "  for " + this);

    Code code = Code.get(rc);//为了方便使用，这里自定义了一组状态
    if (isSuccess(code)) {//成功返回,成功创建zklocakpath节点
      // we successfully created the znode. we are the leader. start monitoring
      if (becomeActive()) {//要将本节点上的NN变成active
        monitorActiveStatus();//继续监控节点状态
      } else {
        reJoinElectionAfterFailureToBecomeActive();//失败，继续选举尝试
      }
      return;
    }

    if (isNodeExists(code)) {//节点存在，说明已经有active，wait即可
      if (createRetryCount == 0) {
        // znode exists and we did not retry the operation. so a different
        // instance has created it. become standby and monitor lock.
        becomeStandby();
      }
      // if we had retried then the znode could have been created by our first
      // attempt to the server (that we lost) and this node exists response is
      // for the second attempt. verify this case via ephemeral node owner. this
      // will happen on the callback for monitoring the lock.
      monitorActiveStatus();//不过努力成为active的动作不能停
      return;
    }

    String errorMessage = "Received create error from Zookeeper. code:"
        + code.toString() + " for path " + path;
    LOG.debug(errorMessage);

    if (shouldRetry(code)) {
      if (createRetryCount < maxRetryNum) {
        LOG.debug("Retrying createNode createRetryCount: " + createRetryCount);
        ++createRetryCount;
        createLockNodeAsync();
        return;
      }
      errorMessage = errorMessage
          + ". Not retrying further znode create connection errors.";
    } else if (isSessionExpired(code)) {
      // This isn't fatal - the client Watcher will re-join the election
      LOG.warn("Lock acquisition failed because session was lost");
      return;
    }

    fatalError(errorMessage);
  }

对于becomeStandby，becomeActive这些状态的改变，有ZKFailoverController来实现。这里只做调用，不做实现。下面会分析ZKFailoverController是如何协调完成一些了的选举，状态监控，重新选举等过程。

转载于:https://my.oschina.net/hadooper/blog/339051