elasticsearch 启动

本地编译

参考：https://blog.csdn.net/heartsdance/article/details/119737124

elasticsearch启动过程

@startuml
!theme plain
Node -[#red]> GatewayMetaState: start() 
Node -[#red]> Coordinator: doStart() 设置coordinationState 协调状态
Node -[#red]> TransportService : acceptIncomingRequests()=true开始接受请求
Node -[#red]> Coordinator : startInitialJoin()开始加入集群
Coordinator -[#red]> Coordinator : becomeCandidate()成为候选者
Coordinator -[#red]> CoordinatorPeerFinder:activate()
CoordinatorPeerFinder -[#red]> CoordinatorPeerFinder:handleWakeUp() 启动周期性监控节点更新
Coordinator -[#red]> Coordinator : onFoundPeersUpdated()查找节点变更 节点mode 是否变更为CANDIDATE 并进行投票
Coordinator -[#red]> ClusterBootstrapService:onFoundPeersUpdated()集群引导
Coordinator -[#red]> ClusterBootstrapService:startBootstrap()开始集群引导 如果执行过不再执行 bootstrappingPermitted = false
ClusterBootstrapService -[#red]> ClusterBootstrapService:doBootstrap()
ClusterBootstrapService -[#red]> Coordinator:setInitialConfiguration()
alt getStateForMasterService().getLastAcceptedConfiguration() 为空

      Coordinator -> CoordinationState:setInitialState()设置ClusterState
      Coordinator -> PreVoteCollector:update()更新预投票的响应
      Coordinator -> Coordinator:startElectionScheduler() 启动选举
      Coordinator -> ElectionSchedulerFactory:electionScheduler = startElectionScheduler() 创建选举调度器
      ElectionSchedulerFactory -> PreVoteCollector:start()开始选举
      PreVoteCollector -> PreVotingRound:start() 广播其他节点开始投票

end

@enduml

选举

 

@startuml
!theme plain
NodeA -[#red]> NodeB: internal:cluster/request_pre_vote: PreVoteRequest(transportService.getLocalNode(), currentTerm)

NodeB -> PreVoteCollector:handlePreVoteRequest()
PreVoteCollector -> NodeB:PreVoteResponse()
NodeB -[#red]> NodeA:PreVoteResponse(currentTerm,lastAcceptedTerm,lastAcceptedVersion)
NodeA -> PreVoteCollector:handlePreVoteResponse()
alt response.lastAcceptedTerm > NodeA.term ||\n(response.lastAcceptedTerm==NodeA.term&&\nresponse.lastAcceptedVersion>NodeA.VersionOrMetaDataVersion)

    PreVoteCollector -> NodeA: 比本节点的高 丢弃预选票

else 比本节点的高低(term version)

    PreVoteCollector -> PreVoteCollector: preVotesReceived.put(sender, response)收集请求
    PreVoteCollector -> PreVoteCollector : voteCollection.addJoinVote()收集选票
    
    alt ElectionStrategy.isElectionQuorum(voteCollection)是获得多数预选票==false
     PreVoteCollector -> NodeA: 没收集到足够的预选票 返回
     else  获得多数的预选票 开始选举
       PreVoteCollector -> Coordinator: startElection
       NodeA -[#red]> NodeB: 通知其他节点可以选我作为term+1进入下一个纪元 leaderinternal:internal:cluster/coordination/start_join:StartJoinRequest(getLocalNode(), Math.max(getCurrentTerm(), maxTermSeen) + 1) 

 NodeB->Coordinator:joinLeaderInTerm() 和本地节点比较
Coordinator->CoordinationState:handleStartJoin()处理加入请求
   alt startJoinRequest.getTerm() <= getCurrentTerm()
        CoordinationState->Coordinator: request的term小于本地的直接返回
   else
        CoordinationState->CoordinationState:更新本地persistedState.setCurrentTerm(startJoinRequest.getTerm()) 清理之前的收集的选票
   end

 Coordinator->NodeB:optionalJoin=Join(localNode,startJoinRequest.getSourceNode(),getCurrentTerm(), getLastAcceptedTerm())
  NodeB-[#red]>NodeA:选我作为leader的选票响应internal:cluster/coordination/join:JoinRequest(transportService.getLocalNode(), optionalJoin)
NodeA->Coordinator:handleJoinRequest(JoinRequest)
Coordinator->Coordinator:processJoinRequest()
Coordinator->CoordinationState:handleJoin()
Coordinator->Coordinator:ensureTermAtLeast() 如果检测到 其他加入节点的版本高于本地 这放弃本次投票 加入到更高的投票中去 出现这种原因是因为预投票是多投票
alt coordinationState.get().electionWon() 节点已经是leader
        Coordinator->Coordinator: handleJoinIgnoringExceptions()
   else
        Coordinator->CoordinationState:收集选票joinVotes.addJoinVote(join)
        CoordinationState->CoordinationState:electionWon=isElectionQuorum(VoteCollection joinVotes) 判断是否获得多数的选票
         alt prevElectionWon == false && coordState.electionWon() 获得多数选票
        Coordinator->Coordinator: becomeLeader() 成为leader mode = Mode.LEADER 
        Coordinator->CandidateJoinAccumulator:close()关闭Candidate mode进入leader
        CandidateJoinAccumulator->MasterService:submitStateUpdateTasks() 提交更新任务
        MasterService->Batcher:submitTasks()批量任务提交
        Batcher->UpdateTask:run()执行任务
        UpdateTask->Batcher:run() 
        Batcher->MasterService:runTasks()执行批量任务
        MasterService->MasterService:calculateTaskOutputs()检测集群状态是否有变更
        alt askOutputs.clusterStateUnchanged() 集群状态没有变更
          
       else
         MasterService->Coordinator:publish(ClusterChangedEvent,startTimeMillis) 发集群状态变更事件
       Coordinator->CoordinationState:组装 PublishRequest
       Coordinator->Coordinator: 创建CoordinatorPublication 用于广播其他节点
       CoordinatorPublication->CoordinatorPublication:new  注册AckListener响应请求，timeoutHandler 注册发布超时监控 
       Coordinator->CoordinatorPublication:start()
       CoordinatorPublication->PublicationTarget:sendPublishRequest() forEach 广播通知其他节点
       PublicationTarget->Publication:sendPublishRequest() responseActionListener=new PublishResponseHandler() 设置ResponseHandler
       
       end
   end
    end
end
@enduml

发布 

 

@startuml
!theme plain
Master-> Publication:start() forEach
Publication-> PublicationContext:sendPublishRequest() 
   alt sendFullVersion || !previousState.nodes().nodeExists(destination) 全量发送
        PublicationContext->PublicationTransportHandler: sendFullClusterState()
   else  diff 发送
        PublicationContext->PublicationTransportHandler:sendClusterStateDiff()
   end

Master -[#red]> NodeB: internal:internal:cluster/coordination/publish_state: BytesTransportRequest(bytes, node.getVersion())

NodeB -> Coordinator:handlePublishRequest()
Coordinator -> CoordinationState:handlePublishRequest() 
CoordinationState->Coordinator:PublishResponse
Coordinator->NodeB: new PublishWithJoinResponse
NodeB->Master: PublishWithJoinResponse
Master -[#red]>PublishResponseHandler:onResponse()
PublishResponseHandler -> CoordinationState:handlePublishResponse()
CoordinationState->CoordinationState:publishVotes.addVote(sourceNode) 收集publish响应
alt isPublishQuorum(publishVotes) 半数
   CoordinationState->CoordinationState:new ApplyCommitRequest()

else
   CoordinationState->CoordinationState:Optional.empty()
end

alt assert applyCommitRequest.isPresent() == false
   CoordinationState->PublicationTarget: forEach sendApplyCommit()
   PublicationTarget->Publication:sendApplyCommit()
   Master->NodeB:internal:cluster/coordination/commit_state
   NodeB->Coordinator:handleApplyCommit()
   Coordinator->CoordinationState:handleCommit()
   CoordinationState->PersistedState:markLastAcceptedStateAsCommitted() 标记已提交
  NodeB->Master: TransportResponse.Empty
  Master->ApplyCommitResponseHandler:onResponse()
  ApplyCommitResponseHandler->PublicationTarget:setAppliedCommit()
  PublicationTarget->Publication:onPossibleCompletion()
  
end

@enduml