Shulou(Shulou.com)06/01 Report--

Zookeeper election source process is what kind of, many novices are not very clear about this, in order to help you solve this problem, the following editor will explain for you in detail, people with this need can come to learn, I hope you can gain something.

Overview of Cluster

Zookeper is usually deployed in a cluster in a production environment to ensure high availability. The following is a cluster deployment structure diagram shown on zookeeper's official website:

As you can see from the figure above, each node of the zookeeper server maintains communication with the master node, and each node stores backups of data and logs, and the cluster is available only when most nodes are available. This article is mainly based on zookeeper 3.8.0 explanation, mainly through the dimension of the source code to analyze the zookeeper election process for zookeeper source code compilation you can refer to: compile and run Zookeeper source code

Cluster node status

The cluster node state is defined in the QuorumPeer#ServerState enumeration, which mainly contains four states: LOOKING, FOLLOWING, LEADING and OBSERVING. Here is the code and description of the definition.

Public enum ServerState {/ / looks for leader status. When the server is in this state, it assumes that there is no leader in the pre-cluster, so it needs to enter the leader election state. LOOKING, / / follower status. Indicates that the current server role is follower. FOLLOWING, / / leadership status. Indicates that the current server role is leader. LEADING, / / Observer status. Indicates that the current server role is observer. OBSERVING} Leader election process initiation and initialization

QuorumPeerMain is the startup class of zookeeper, launched by the main method

/ / do not show non-core code public static void main (String [] args) {QuorumPeerMain main = new QuorumPeerMain (); main.initializeAndRun (args);} protected void initializeAndRun (String [] args) throws ConfigException, IOException, AdminServerException {/ / start if in cluster mode (args.length = = 1 & & config.isDistributed ()) {runFromConfig (config) } else {}} public void runFromConfig (QuorumPeerConfig config) throws IOException, AdminServerException {/ / quorumPeer starts quorumPeer.start ();}

QuorumPeer is a thread instance class. When the start method is called, it will cause the QuorumPeer#run () method to determine the cluster status and finally enter a series of operations such as whether to perform elections or synchronize cluster node data information. The following is the core code:

@ Override public void run () {try {while (running) {switch (getPeerState ()) {case LOOKING: / / vote for yourself setCurrentVote (makeLEStrategy (). LookForLeader ()); break; case OBSERVING: setObserver (makeObserver (logFactory)) Observer.observeLeader (); break; case FOLLOWING: setFollower (makeFollower (logFactory)); follower.followLeader (); break; case LEADING: setLeader (makeLeader (logFactory)); leader.lead () SetLeader (null); break;} finally {}} to conduct elections

FastLeaderElection is the core class of the election, in which there is a process of dealing with votes and ballots.

Public Vote lookForLeader () throws InterruptedException {/ / create a ballot box for the current election cycle Map recvset = new HashMap (); / / create a ballot box. This ballot box is different from recvset. / / store voting Map outofelection = new HashMap () if Leader already exists in the current cluster; int notTimeout = minNotificationInterval; synchronized (this) {/ / increment local election cycle logicalclock.incrementAndGet (); / / vote for yourself updateProposal (getInitId (), getInitLastLoggedZxid (), getPeerEpoch ());} / / broadcast voting sendNotifications (); SyncedLearnerTracker voteSet = null / / if the status of the current server is Looking and the stop parameter is false, then while ((self.getPeerState () = = ServerState.LOOKING) & & (! stop)) {if (n.electionEpoch > logicalclock.get ()) {logicalclock.set (n.electionEpoch); recvset.clear () / / totalOrderPredicate voting competition 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 (totalOrderPredicate (n.leader, n.zxid, n.peerEpoch, proposedLeader, proposedZxid, proposedEpoch) {updateProposal (n.leader, n.zxid, n.peerEpoch); sendNotifications ();} / / listen for Notification n = recvqueue.poll (notTimeout, TimeUnit.MILLISECONDS) received by the communication layer / put into the ballot box recvset.put (n.sid, new Vote (n.leader, n.zxid, n.electionEpoch, n.peerEpoch); / / more than half of the logic voteSet = getVoteTracker (recvset, new Vote (proposedLeader, proposedZxid, logicalclock.get (), proposedEpoch));}}

TotalOrderPredicate is mainly the logic of voting. Let's take a look at the code:

Protected boolean totalOrderPredicate (long newId, long newZxid, long newEpoch, long curId, long curZxid, long curEpoch) {if (self.getQuorumVerifier (). GetWeight (newId) = = 0) {return false } return ((newEpoch > curEpoch) | | (newEpoch = = curEpoch) & & (newZxid > curZxid) | | (newZxid = = curZxid) & & (newId > curId);}

The election process looks like this, but in fact, officials have also given comments:

Hongmeng official Strategic Cooperation to build HarmonyOS Technology Community

First of all, compare the number of elections, the high number of terms means that the latest election, winning

Then compare zxid, that is, to see whose data is the latest, the latest winner.

Finally, compare serverid, which is specified in the configuration file. The node with the largest id wins the election and notifies the other nodes through sendNotifications ().

Process summary

Earlier, I roughly explained zookeeper from the startup process to the election, the synchronization of election results, and how to vote in the election result confirmation process, but zookeeper, as a high-performance and reliable distributed coordination middleware, is also excellent in many design details.

Voting process

In general, the larger the zxid, the more likely it is to become leader during the voting process, mainly because the larger the zxid, the more data the node has, so that the comparison process between node transaction revocation and log file synchronization can be reduced in the process of data synchronization, so as to improve performance. The following is the process of electing five zookeeper nodes.

Note: (sid, zxid), the current scenario is server1, server2 fails, zxid of server3 is 9, zxid of server4 and server5 is 8. After two rounds of elections, sever3 was finally selected as the leader node.

Multi-layer network architecture

In the previous analysis, I omitted the NIO operation of communication between Zookeeper nodes. In this part, zookeeper divides them into transport layer and business layer. Network layer packets are processed by SendWorker and RecvWorker, and business layer data is processed by WorkerSender and WorkerReceiver.

Here will involve multi-threaded operation, zookeeper in the source code also gives a lot of log information, for beginners have a certain degree of difficulty, you can refer to the following Zookeeper election source code process of this part of the flow chart to assist analysis.

Leader election source code process

Combined with the above carding, I have done a more detailed comb on the process of zookeeper start-up and election.

Is it helpful for you to read the above content? If you want to know more about the relevant knowledge or read more related articles, please follow the industry information channel, thank you for your support.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.