elasticsearch的zenDiscovery和master选举机制原理分析

private void innerJoinCluster() {
        DiscoveryNode masterNode = null;
        final Thread currentThread = Thread.currentThread();
　　　　　//一直阻塞直到找到master节点，在集群刚刚启动，或者集群master丢失的情况，这种阻塞能够保证集群一致性
        while (masterNode == null && joinThreadControl.joinThreadActive(currentThread)) {
            masterNode = findMaster();
        }
　　　　　　//有可能自己会被选举为master（集群启动，或者加入时正在选举）
　　　　　　if (clusterService.localNode().equals(masterNode)) {
　　　　　　//如果本身是master，则需要向其它所有节点发送集群状态更新
            clusterService.submitStateUpdateTask("zen-disco-join (elected_as_master)", Priority.IMMEDIATE, new ProcessedClusterStateNonMasterUpdateTask() {
                @Override
                public ClusterState execute(ClusterState currentState) {
　　　　　　　　　　　　//选举时错误的，之前的master状态良好，则不更新状态，仍旧使用之前状态。
                    if (currentState.nodes().masterNode() != null) {
                       return currentState;
                    }
                    DiscoveryNodes.Builder builder = new DiscoveryNodes.Builder(currentState.nodes()).masterNodeId(currentState.nodes().localNode().id());
                    // update the fact that we are the master...
                    ClusterBlocks clusterBlocks = ClusterBlocks.builder().blocks(currentState.blocks()).removeGlobalBlock(discoverySettings.getNoMasterBlock()).build();
                    currentState = ClusterState.builder(currentState).nodes(builder).blocks(clusterBlocks).build();
                    // eagerly run reroute to remove dead nodes from routing table
                    RoutingAllocation.Result result = allocationService.reroute(currentState);
                    return ClusterState.builder(currentState).routingResult(result).build();
                }
                @Override
                public void onFailure(String source, Throwable t) {
                    logger.error("unexpected failure during [{}]", t, source);
                    joinThreadControl.markThreadAsDoneAndStartNew(currentThread);
                }
                @Override
                public void clusterStateProcessed(String source, ClusterState oldState, ClusterState newState) {
                    if (newState.nodes().localNodeMaster()) {
                        // we only starts nodesFD if we are master (it may be that we received a cluster state while pinging)
                        joinThreadControl.markThreadAsDone(currentThread);
                        nodesFD.updateNodesAndPing(newState); // start the nodes FD
                    } else {
                        // if we're not a master it means another node published a cluster state while we were pinging
                        // make sure we go through another pinging round and actively join it
                        joinThreadControl.markThreadAsDoneAndStartNew(currentThread);
                    }
                    sendInitialStateEventIfNeeded();
                    long count = clusterJoinsCounter.incrementAndGet();
                    logger.trace("cluster joins counter set to [{}] (elected as master)", count);
                }
            });
        } else {
            // 找到的节点不是我，试图连接该master
            final boolean success = joinElectedMaster(masterNode);
            // finalize join through the cluster state update thread
            final DiscoveryNode finalMasterNode = masterNode;
            clusterService.submitStateUpdateTask("finalize_join (" + masterNode + ")", new ClusterStateNonMasterUpdateTask() {
                @Override
                public ClusterState execute(ClusterState currentState) throws Exception {
                    if (!success) {
                        // failed to join. Try again...
                        joinThreadControl.markThreadAsDoneAndStartNew(currentThread);
                        return currentState;
                    }
                    if (currentState.getNodes().masterNode() == null) {
                        // Post 1.3.0, the master should publish a new cluster state before acking our join request. we now should have
                        // a valid master.
                        logger.debug("no master node is set, despite of join request completing. retrying pings.");
                        joinThreadControl.markThreadAsDoneAndStartNew(currentThread);
                        return currentState;
                    }
                    if (!currentState.getNodes().masterNode().equals(finalMasterNode)) {
                        return joinThreadControl.stopRunningThreadAndRejoin(currentState, "master_switched_while_finalizing_join");
                    }
                    // Note: we do not have to start master fault detection here because it's set at {@link #handleNewClusterStateFromMaster }
                    // when the first cluster state arrives.
                    joinThreadControl.markThreadAsDone(currentThread);
                    return currentState;
                }
                @Override
                public void onFailure(String source, @Nullable Throwable t) {
                    logger.error("unexpected error while trying to finalize cluster join", t);
                    joinThreadControl.markThreadAsDoneAndStartNew(currentThread);
                }
            });
        }
    }

private DiscoveryNode findMaster() {
　　　　　　//ping集群中的节点
        ZenPing.PingResponse[] fullPingResponses = pingService.pingAndWait(pingTimeout);
        if (fullPingResponses == null) {return null;
        }// 过滤所得到的ping响应，虑除client节点，单纯的data节点
        List<ZenPing.PingResponse> pingResponses = Lists.newArrayList();
        for (ZenPing.PingResponse pingResponse : fullPingResponses) {
            DiscoveryNode node = pingResponse.node();
            if (masterElectionFilterClientNodes && (node.clientNode() || (!node.masterNode() && !node.dataNode()))) {
                // filter out the client node, which is a client node, or also one that is not data and not master (effectively, client)
            } else if (masterElectionFilterDataNodes && (!node.masterNode() && node.dataNode())) {
                // filter out data node that is not also master
            } else {
                pingResponses.add(pingResponse);
            }
        }
       final DiscoveryNode localNode = clusterService.localNode();
        List<DiscoveryNode> pingMasters = newArrayList();
　　　　　//获取所有ping响应中的master节点，如果master节点是节点本身则过滤掉。pingMasters列表结果要么为空（本节点是master）要么是同一个节点（出现不同节点则集群出现了问题
不过没关系，后面会进行选举）
        for (ZenPing.PingResponse pingResponse : pingResponses) {
            if (pingResponse.master() != null) {
                if (!localNode.equals(pingResponse.master())) {
                    pingMasters.add(pingResponse.master());
                }
            }
        }
        // nodes discovered during pinging
        Set<DiscoveryNode> activeNodes = Sets.newHashSet();
        // nodes discovered who has previously been part of the cluster and do not ping for the very first time
        Set<DiscoveryNode> joinedOnceActiveNodes = Sets.newHashSet();
        Version minimumPingVersion = localNode.version();
　　　　for (ZenPing.PingResponse pingResponse : pingResponses) {
   　　　　 activeNodes.add(pingResponse.node());
    　　　　minimumPingVersion = Version.smallest(pingResponse.node().version(), minimumPingVersion);
    　　　　if (pingResponse.hasJoinedOnce() != null && pingResponse.hasJoinedOnce()) {
        　　joinedOnceActiveNodes.add(pingResponse.node());
    　　　　}
　　　　}

//本节点暂时是master也要加入候选节点进行选举
        if (localNode.masterNode()) {
            activeNodes.add(localNode);
            long joinsCounter = clusterJoinsCounter.get();
            if (joinsCounter > 0) {
                logger.trace("adding local node to the list of active nodes who has previously joined the cluster (joins counter is [{}})", joinsCounter);
                joinedOnceActiveNodes.add(localNode);
            }
        }
　　　　　　//pingMasters为空，则本节点是master节点，
　　　　if (pingMasters.isEmpty()) {
            if (electMaster.hasEnoughMasterNodes(activeNodes)) {//保证选举数量,说明有足够多的节点选举本节点为master，但是这还不够，本节点还需要再选举一次，如果
　　　　　　　　　　本次选举节点仍旧是自己，那么本节点才能成为master。这里就体现了master选举的第二条原则。
                DiscoveryNode master = electMaster.electMaster(joinedOnceActiveNodes);
                if (master != null) {
                    return master;
                }
                return electMaster.electMaster(activeNodes);
            } else {
                // if we don't have enough master nodes, we bail, because there are not enough master to elect from
                logger.trace("not enough master nodes [{}]", activeNodes);
                return null;
            }
        } else {
　　　　　　　　//pingMasters不为空（pingMasters列表中应该都是同一个节点），本节点没有被选举为master，那就接受之前的选举。
            return electMaster.electMaster(pingMasters);
        }
    }

clusterService.submitStateUpdateTask("zen-disco-master_failed (" + masterNode + ")", Priority.IMMEDIATE, new ProcessedClusterStateNonMasterUpdateTask() {           
 　　　　　　@Override
            public ClusterState execute(ClusterState currentState) {
                //获取到当前集群状态下的所有节点
                DiscoveryNodes discoveryNodes = DiscoveryNodes.builder(currentState.nodes())
                        // make sure the old master node, which has failed, is not part of the nodes we publish
                        .remove(masterNode.id())
                        .masterNodeId(null).build();
　　　　　　　　　　//rejoin过程仍然是重复findMaster过程
　　　　　　　　　　if (rejoin) {
                    return rejoin(ClusterState.builder(currentState).nodes(discoveryNodes).build(), "master left (reason = " + reason + ")");
                }
　　　　　　　　　　//无法达到选举数量，进行findMaster过程
                if (!electMaster.hasEnoughMasterNodes(discoveryNodes)) {
                    return rejoin(ClusterState.builder(currentState).nodes(discoveryNodes).build(), "not enough master nodes after master left (reason = " + reason + ")");
                }
　　　　　　　　　　//在当前集群状态下，如果候选节点数量达到预期数量，那么选举出来的节点一定是同一个节点，因为所有的节点看到的集群states是一致的
                final DiscoveryNode electedMaster = electMaster.electMaster(discoveryNodes); // elect master
                final DiscoveryNode localNode = currentState.nodes().localNode();
              ....
            }