Zookeeper的选举算法和脑裂问题深度讲解
ZK介绍
ZK = zookeeper
ZK是微服务解决方案中拥有服务注册发现最为核心的环境,是微服务的基石。作为服务注册发现模块,并不是只有ZK一种产品,目前得到行业认可的还有:Eureka、Consul。
这里我们只聊ZK,这个工具本身很小zip包就几兆,安装非常傻瓜,能够支持集群部署。
官网地址:https://zookeeper.apache.org/
背景
在集群环境下ZK的leader&follower的概念,已经节点异常ZK面临的问题以及如何解决。ZK本身是java语言开发,也开源到Github上但官方文档对内部介绍的很少,零散的博客很多,有些写的很不错。
提问:
ZK集群单节点状态(每个节点有且只有一个状态),ZK的定位一定需要一个leader节点处于lading状态。
- looking:寻找leader状态,当前集群没有leader,进入leader选举流程。
- following:跟随者状态,接受leading节点同步和指挥。
- leading:领导者状态。
- observing:观察者状态,表名当前服务器是observer。
过半选举算法
ZK中有三种选举算法,分别是LeaderElection,FastLeaderElection,AuthLeaderElection,FastLeaderElection和AuthLeaderElection是类似的选举算法,唯一区别是后者加入了认证信息, FastLeaderElection比LeaderElection更高效,后续的版本只保留FastLeaderElection。
理解:
在集群环境下多个节点启动,ZK首先需要在多个节点中选出一个节点作为leader并处于Leading状态,这样就面临一个选举问题,同时选举规则是什么样的。“过半选举算法”:投票选举中获得票数过半的节点胜出,即状态从looking变为leading,效率更高。
官网资料描述:Clustered (Multi-Server) Setup,如下图:
As long as a majority of the ensemble are up, the service will be available. Because Zookeeper requires a majority, it is best to use an odd number of machines. For example, with four machines ZooKeeper can only handle the failure of a single machine; if two machines fail, the remaining two machines do not constitute a majority. However, with five machines ZooKeeper can handle the failure of two machines.
以5台服务器讲解思路:
- 服务器1启动,此时只有它一台服务器启动了,它发出去的Vote没有任何响应,所以它的选举状态一直是LOOKING状态;
- 服务器2启动,它与最开始启动的服务器1进行通信,互相交换自己的选举结果,由于两者都没有历史数据,所以id值较大的服务器2胜出,但是由于没有达到超过半数以上的服务器都同意选举它(这个例子中的半数以上是3),所以服务器1,2还是继续保持LOOKING状态.
- 服务器3启动,根据前面的理论,分析有三台服务器选举了它,服务器3成为服务器1,2,3中的老大,所以它成为了这次选举的leader.
- 服务器4启动,根据前面的分析,理论上服务器4应该是服务器1,2,3,4中最大的,但是由于前面已经有半数以上的服务器选举了服务器3,所以它只能接收当小弟的命了.
- 服务器5启动,同4一样,当小弟.
源码解析:
URL: FastLeaderElection
/** * Starts a new round of leader election. Whenever our QuorumPeer * changes its state to LOOKING, this method is invoked, and it * sends notifications to all other peers. */ public Vote lookForLeader() throws InterruptedException { try { self.jmxLeaderElectionBean = new LeaderElectionBean(); MBeanRegistry.getInstance().register(self.jmxLeaderElectionBean, self.jmxLocalPeerBean); } catch (Exception e) { LOG.warn("Failed to register with JMX", e); self.jmxLeaderElectionBean = null; } self.start_fle = Time.currentElapsedTime(); try { Map<Long, Vote> recvset = new HashMap<Long, Vote>(); Map<Long, Vote> outofelection = new HashMap<Long, Vote>(); int notTimeout = minNotificationInterval; synchronized (this) { logicalclock.incrementAndGet(); updateProposal(getInitId(), getInitLastLoggedZxid(), getPeerEpoch()); } LOG.info("New election. My id = " + self.getId() + ", proposed zxid=0x" + Long.toHexString(proposedZxid)); sendNotifications(); SyncedLearnerTracker voteSet; /* * Loop in which we exchange notifications until we find a leader */ while ((self.getPeerState() == ServerState.LOOKING) && (!stop)) { /* * Remove next notification from queue, times out after 2 times * the termination time */ Notification n = recvqueue.poll(notTimeout, TimeUnit.MILLISECONDS); /* * Sends more notifications if haven't received enough. * Otherwise processes new notification. */ if (n == null) { if (manager.haveDelivered()) { sendNotifications(); } else { manager.connectAll(); } /* * Exponential backoff */ int tmpTimeOut = notTimeout * 2; notTimeout = (tmpTimeOut < maxNotificationInterval ? tmpTimeOut : maxNotificationInterval); LOG.info("Notification time out: " + notTimeout); } else if (validVoter(n.sid) && validVoter(n.leader)) { /* * Only proceed if the vote comes from a replica in the current or next * voting view for a replica in the current or next voting view. */ switch (n.state) { case LOOKING: if (getInitLastLoggedZxid() == -1) { LOG.debug("Ignoring notification as our zxid is -1"); break; } if (n.zxid == -1) { LOG.debug("Ignoring notification from member with -1 zxid {}", n.sid); break; } // If notification > current, replace and send messages out if (n.electionEpoch > logicalclock.get()) { logicalclock.set(n.electionEpoch); recvset.clear(); if (totalOrderPredicate(n.leader, n.zxid, n.peerEpoch, getInitId(), getInitLastLoggedZxid(), getPeerEpoch())) { updateProposal(n.leader, n.zxid, n.peerEpoch); } else { updateProposal(getInitId(), getInitLastLoggedZxid(), getPeerEpoch()); } sendNotifications(); } else if (n.electionEpoch < logicalclock.get()) { if (LOG.isDebugEnabled()) { LOG.debug( "Notification election epoch is smaller than logicalclock. n.electionEpoch = 0x" + Long.toHexString(n.electionEpoch) + ", logicalclock=0x" + Long.toHexString(logicalclock.get())); } break; } else if (totalOrderPredicate(n.leader, n.zxid, n.peerEpoch, proposedLeader, proposedZxid, proposedEpoch)) { updateProposal(n.leader, n.zxid, n.peerEpoch); sendNotifications(); } if (LOG.isDebugEnabled()) { LOG.debug("Adding vote: from=" + n.sid + ", proposed leader=" + n.leader + ", proposed zxid=0x" + Long.toHexString(n.zxid) + ", proposed election epoch=0x" + Long.toHexString(n.electionEpoch)); } // don't care about the version if it's in LOOKING state recvset.put(n.sid, new Vote(n.leader, n.zxid, n.electionEpoch, n.peerEpoch)); voteSet = getVoteTracker(recvset, new Vote(proposedLeader, proposedZxid, logicalclock.get(), proposedEpoch)); if (voteSet.hasAllQuorums()) { // Verify if there is any change in the proposed leader while ((n = recvqueue.poll(finalizeWait, TimeUnit.MILLISECONDS)) != null) { if (totalOrderPredicate(n.leader, n.zxid, n.peerEpoch, proposedLeader, proposedZxid, proposedEpoch)) { recvqueue.put(n); break; } } /* * This predicate is true once we don't read any new * relevant message from the reception queue */ if (n == null) { setPeerState(proposedLeader, voteSet); Vote endVote = new Vote(proposedLeader, proposedZxid, logicalclock.get(), proposedEpoch); leaveInstance(endVote); return endVote; } } break; case OBSERVING: LOG.debug("Notification from observer: {}", n.sid); break; case FOLLOWING: case LEADING: /* * Consider all notifications from the same epoch * together. */ if (n.electionEpoch == logicalclock.get()) { recvset.put(n.sid, new Vote(n.leader, n.zxid, n.electionEpoch, n.peerEpoch)); voteSet = getVoteTracker(recvset, new Vote(n.version, n.leader, n.zxid, n.electionEpoch, n.peerEpoch, n.state)); if (voteSet.hasAllQuorums() && checkLeader(outofelection, n.leader, n.electionEpoch)) { setPeerState(n.leader, voteSet); Vote endVote = new Vote(n.leader, n.zxid, n.electionEpoch, n.peerEpoch); leaveInstance(endVote); return endVote; } } /* * Before joining an established ensemble, verify that * a majority are following the same leader. */ outofelection.put(n.sid, new Vote(n.version, n.leader, n.zxid, n.electionEpoch, n.peerEpoch, n.state)); voteSet = getVoteTracker(outofelection, new Vote(n.version, n.leader, n.zxid, n.electionEpoch, n.peerEpoch, n.state)); if (voteSet.hasAllQuorums() && checkLeader(outofelection, n.leader, n.electionEpoch)) { synchronized (this) { logicalclock.set(n.electionEpoch); setPeerState(n.leader, voteSet); } Vote endVote = new Vote(n.leader, n.zxid, n.electionEpoch, n.peerEpoch); leaveInstance(endVote); return endVote; } break; default: LOG.warn("Notification state unrecoginized: " + n.state + " (n.state), " + n.sid + " (n.sid)"); break; } } else { if (!validVoter(n.leader)) { LOG.warn("Ignoring notification for non-cluster member sid {} from sid {}", n.leader, n.sid); } if (!validVoter(n.sid)) { LOG.warn("Ignoring notification for sid {} from non-quorum member sid {}", n.leader, n.sid); } } } return null; } finally { try { if (self.jmxLeaderElectionBean != null) { MBeanRegistry.getInstance().unregister(self.jmxLeaderElectionBean); } } catch (Exception e) { LOG.warn("Failed to unregister with JMX", e); } self.jmxLeaderElectionBean = null; LOG.debug("Number of connection processing threads: {}", manager.getConnectionThreadCount()); } }
/* * We return true if one of the following three cases hold: * 1- New epoch is higher * 2- New epoch is the same as current epoch, but new zxid is higher * 3- New epoch is the same as current epoch, new zxid is the same * as current zxid, but server id is higher. */ return ((newEpoch > curEpoch) || ((newEpoch == curEpoch) && ((newZxid > curZxid) || ((newZxid == curZxid) && (newId > curId)))));
脑裂问题
脑裂问题出现在集群中leader死掉,follower选出了新leader而原leader又复活了的情况下,因为ZK的过半机制是允许损失一定数量的机器而扔能正常提供给服务,当leader死亡判断不一致时就会出现多个leader。
方案:
ZK的过半机制一定程度上也减少了脑裂情况的出现,起码不会出现三个leader同时。ZK中的Epoch机制(时钟)每次选举都是递增+1,当通信时需要判断epoch是否一致,小于自己的则抛弃,大于自己则重置自己,等于则选举;
// If notification > current, replace and send messages out if (n.electionEpoch > logicalclock.get()) { logicalclock.set(n.electionEpoch); recvset.clear(); if (totalOrderPredicate(n.leader, n.zxid, n.peerEpoch, getInitId(), getInitLastLoggedZxid(), getPeerEpoch())) { updateProposal(n.leader, n.zxid, n.peerEpoch); } else { updateProposal(getInitId(), getInitLastLoggedZxid(), getPeerEpoch()); } sendNotifications(); } else if (n.electionEpoch < logicalclock.get()) { if (LOG.isDebugEnabled()) { LOG.debug( "Notification election epoch is smaller than logicalclock. n.electionEpoch = 0x" + Long.toHexString(n.electionEpoch) + ", logicalclock=0x" + Long.toHexString(logicalclock.get())); } break; } else if (totalOrderPredicate(n.leader, n.zxid, n.peerEpoch, proposedLeader, proposedZxid, proposedEpoch)) { updateProposal(n.leader, n.zxid, n.peerEpoch); sendNotifications(); }
归纳
在日常的ZK运维时需要注意以上场景在极端情况下出现问题,特别是脑裂的出现,可以采用:
过半选举策略下部署原则:
- 服务器群部署要单数,如:3、5、7、...,单数是最容易选出leader的配置量。
- ZK允许节点最大损失数,原则就是“保证过半选举正常”,多了就是浪费。
详细的算法逻辑是很复杂要考虑很多情况,其中有个Epoch的概念(自增长),分为:LogicEpoch和ElectionEpoch,每次投票都有判断每个投票周期是否一致等等。
在思考ZK策略时经常遇到这样的问题(上文中两块),梳理了一下思路以便于理解也作为后续回顾,特别感谢下面几篇博文的支持,感谢分享;
作者:Owen Jia
可以关注他的博客:Owen Blog
参考博文资料:
- 机试算法讲解: 第44题 深度优先搜索之素数环问题
- 机试算法讲解: 第43题 递归之汉诺塔问题
- zookeeper选举原理讲解
- 该深度优先搜索算法查找钥匙的程序之问题~~
- [未完成]N皇后问题的深度算法思路
- 极限元算法专家:深度学习在语音生成问题上的典型应用 | 学术分享会总结文
- zookeeper能启动但无法选举问题
- 作业调度问题深度搜索定界算法
- Zookeeper中的FastLeaderElection选举算法简述
- 啊哈算法之xxx+xxx=xxx问题(深度优先实现)
- zookeeper选举算法
- 机试算法讲解: 第45题 深度优先搜索之寻找沙特石油存储区
- 基于栈的应用--列车车厢重排问题、开关盒布线问题、离线等价类问题、迷宫老鼠(深度优先、回溯)问题的算法实现
- 算法还是算力?周志华微博引爆深度学习的“鸡生蛋,蛋生鸡”问题
- zookeeper选举算法之FastLeaderElection
- zookeeper与kafka的选举算法
- ZooKeeper系统模型之Leader选举算法分析。
- 机试算法讲解: 第46题 深度优先搜索之能否逃出魔掌
- zookeeper FastLeader选举算法
- zookeeper3.3.3源码分析(二)FastLeader选举算法