zookeeper原理之Leader选举完成之后的处理逻辑

通过 lookForLeader 方法选举完成以后，会设置当前节点的 PeerState，要么为 Leading、要么就是 FOLLOWING、或者 OBSERVING到这里，只是表示当前的 leader 选出来了，但是 QuorumPeer.run 方法里面还没执行完，我们再回过头看看后续的处理过程。

QuorumPeer.run

分别来看看 case 为 FOLLOWING 和 LEADING，会做什么事情：

@Override
   public void run() {
       setName("QuorumPeer" + "[myid=" + getId() + "]" + cnxnFactory.getLocalAddress());
       while (running) {
           switch (getPeerState()) {
               case LOOKING:
               case OBSERVING:
               case FOLLOWING:
                   try {
                       LOG.info("FOLLOWING");
                       setFollower(makeFollower(logFactory));
                       follower.followLeader();
                   } catch (Exception e) {
                       LOG.warn("Unexpected exception",e);
                   } finally {
                       follower.shutdown();
                       setFollower(null);
                       setPeerState(ServerState.LOOKING);
                   }
                   break;
               case LEADING:
                   LOG.info("LEADING");
                   try {
                       setLeader(makeLeader(logFactory));
                       leader.lead();
                       setLeader(null);
                   } catch (Exception e) {
                       LOG.warn("Unexpected exception",e);
                   } finally {
                       if (leader != null) {
                           leader.shutdown("Forcing shutdown");
                           setLeader(null);
                       }
                       setPeerState(ServerState.LOOKING);
                   }
                   break;
           }
           //省略部分代码
       }

makeFollower

初始化一个 Follower 对象，构建一个 FollowerZookeeperServer，表示 follower 节点的请求处理服务。

protected Follower makeFollower(FileTxnSnapLog logFactory) throws IOException {
       return new Follower(this, new FollowerZooKeeperServer(logFactory, this, new ZooKeeperServer.BasicDataTreeBuilder(), this.zkDb));
   }

follower.followLeader();

protected Follower
    makeFollower(FileTxnSnapLog logFactory) throws IOException {
        return new Follower(this, new FollowerZooKeeperServer(logFactory, this,new ZooKeeperServer.BasicDataTreeBuilder(), this.zkDb));
    }
    follower.followLeader();
    void followLeader() throws InterruptedException {
        //省略部分代码
        try {
            //根据 sid 找到对应 leader，拿到 lead 连接信息
            QuorumServer leaderServer = findLeader();
            try {//连接到 Leader
                connectToLeader(leaderServer.addr, leaderServer.hostname);
                //将 Follower 的 zxid 及 myid 等信息封装好发送到 Leader，同步 epoch。 也就是意味着接下来 follower 节点只同步新 epoch 的数据信息
                long newEpochZxid = registerWithLeader(Leader.FOLLOWERINFO);
                //如果 leader 的 epoch 比当前 follow 节点的 poch 还小，抛异常
                long newEpoch = ZxidUtils.getEpochFromZxid(newEpochZxid);
                if (newEpoch < self.getAcceptedEpoch()) {
                    LOG.error("Proposed leader epoch " + ZxidUtils.zxidToString(newEpochZxid) + " is less than our accepted epoch " + ZxidUtils.zxidToString(self.getAcceptedEpoch()));
                    throw new IOException("Error: Epoch of leader is lower");
                }//和 leader 进行数据同步
                syncWithLeader(newEpochZxid);
                QuorumPacket qp = new QuorumPacket();
                while (this.isRunning()) { //接受 Leader 消息，执行并反馈给 leader，线程在此自旋
                    readPacket(qp);//从 leader 读取数据包
                    processPacket(qp);//处理 packet
                }
            } catch (Exception e) {
                LOG.warn("Exception when following the leader", e);
                try {
                    sock.close();
                } catch (IOException e1) {
                    e1.printStackTrace();
                }
                // clear pending revalidations
                pendingRevalidations.clear();
            }
        } finally {
            zk.unregisterJMX((Learner)this);
        } 
    }

makeLeader

初始化一个 Leader 对象，构建一个 LeaderZookeeperServer，用于表示leader 节点的请求处理服务。

protected Leader makeLeader(FileTxnSnapLog logFactory) throws IOException {
       return new Leader(this, new LeaderZooKeeperServer(logFactory, this,new ZooKeeperServer.BasicDataTreeBuilder(), this.zkDb));
   }