Zookeeper leader选举原理

创建传输层收发选票线程和队列

addToRecvQueue

sendworker

选票参与leader选举

发送线程投递选票

addToSendQueue

应用层

sockect

QuorumCnxManager.Listener listener = qcm.listener;listener.start();

poll()接收线程获取投票

sid(key)

updateProposal()更新选票为投自己zookeeper刚开始都是投的自己为leader

queue

public boolean containsQuorum(Set<Long> ackSet) {        return (ackSet.size() > half);    }当前投票leader 数大于 总参与选举的zookeeper的一半时，leader选举成功

termPredicate()过半选举逻辑

所有选票放入发送队列

Zookeeper leader选举源码中的核心方法

keysid(3)

接收线程获取选票msg

super.start(); leader选举核心逻辑makeLEStrategy().lookForLeader()

sendqueue

totalOrderPredicate选举核心方法return ((newEpoch > curEpoch) || ((newEpoch == curEpoch) && ((newZxid > curZxid) || ((newZxid == curZxid) && (newId > curId)))));事务zxid大的优先为leader事务id 一致，默认myid(curid)大的为leader

mySid == sidaddToRecvQueue()

WorkerSenderwsThread.start();

(sid)RecvWorker

termPredicate()Zookeeper过半选举机制

WorkerReceiverwrThread.start();

queueSendMap

recvqueue.offer(n);

keysid(myid)

未选举出leader进行下一轮选举

recvqueue

recvqueue.poll

value

recvQueue

keysid(2)

leader选举逻辑makeLEStrategy().lookForLeader()

poll()

createElectionAlgorithm()创建socket、收发队列

发送方选举状态为Looking状态且选举周期小于当前选举周期，将选票回发给发送方sendqueue.offer(notmsg);

server.1=127.0.0.1:2001:3001:participantserver.2=127.0.0.1:2002:3002:participantserver.3=127.0.0.1:2003:3003:participant

sendNotifications()经过一轮选举后，将选举结果发送出去

quorumPeer.start();入口方法

从接收队列获取选票

self.setPeerState((proposedLeader == self.getId()) ? ServerState.LEADING: learningState());各新选举结果，设置当前节点为leading或者following

QuorumCnxManager.Listener listener = qcm.listener;listener.start();创建socket创建SendWorker、RecvWorker、recvQueue、queueSendMap

updateProposal()更新当前选票为上面选举出来的leader

senderWorkerMap

创建应用层收发选票线程和队列

config

sendNotifications()每个server生成一张选票

传输层

FastLeaderElection