Copycat - configure

Copycat server之间的configure是如何，何时被同步的？

 

大家可以看到，只有leader可以同步配置

 

1. 显式的调用LeaderState.configure

LeaderState.configure

/**
   * Commits the given configuration.
   */
  protected CompletableFuture<Long> configure(Collection<Member> members) {
    final long index;
    try (ConfigurationEntry entry = context.getLog().create(ConfigurationEntry.class)) {
      entry.setTerm(context.getTerm())
        .setTimestamp(System.currentTimeMillis())
        .setMembers(members);
      index = context.getLog().append(entry); //先把configuration写入local log
      LOGGER.debug("{} - Appended {}", context.getCluster().member().address(), entry);

      // Store the index of the configuration entry in order to prevent other configurations from
      // being logged and committed concurrently. This is an important safety property of Raft.
      configuring = index; //configuring用于互斥
      context.getClusterState().configure(new Configuration(entry.getIndex(), entry.getTerm(), entry.getTimestamp(), entry.getMembers())); //更新ClusterState
    }

    return appender.appendEntries(index).whenComplete((commitIndex, commitError) -> { //调用appendEntries同步configuration
      context.checkThread();
      if (isOpen()) {
        // Reset the configuration index to allow new configuration changes to be committed.
        configuring = 0; //configuring完成
      }
    });
  }

何处显式调用？

private void appendInitialEntries() {

  // Append a configuration entry to propagate the leader's cluster configuration.
  configure(context.getCluster().members());

 

public CompletableFuture<JoinResponse> join(final JoinRequest request) {

  configure(members).whenComplete((index, error) -> {

 

@Override
public CompletableFuture<ReconfigureResponse> reconfigure(final ReconfigureRequest request) {

  configure(members).whenComplete((index, error) -> {

 

@Override
public CompletableFuture<LeaveResponse> leave(final LeaveRequest request) {

  configure(members).whenComplete((index, error) -> {

 

2. 在发送hb，和appendEntries时，也会自动发生同步

因为上面调用configure也是会调用到

LeaderAppender.appendEntries(MemberState member)

和普通append不同的是，会走到这个逻辑，

    // If the member term is less than the current term or the member's configuration index is less
    // than the local configuration index, send a configuration update to the member.
    // Ensure that only one configuration attempt per member is attempted at any given time by storing the
    // member state in a set of configuring members.
    // Once the configuration is complete sendAppendRequest will be called recursively.
    else if (member.getConfigTerm() < context.getTerm() || member.getConfigIndex() < context.getClusterState().getConfiguration().index()) {
      if (member.canConfigure()) {
        sendConfigureRequest(member, buildConfigureRequest(member));
      }
    }

注意虽然是在appendEntries里面，

这里发出的是sendConfigureRequest，而不是appendRequest

因为leader的configuration发生变化，所以member.getConfigIndex一定是小的，所以要更新

AbstractAppender

/**
   * Connects to the member and sends a configure request.
   */
  protected void sendConfigureRequest(MemberState member, ConfigureRequest request) {
    // Start the configure to the member.
    member.startConfigure();

    context.getConnections().getConnection(member.getMember().serverAddress()).whenComplete((connection, error) -> {
      context.checkThread();

      if (open) {
        if (error == null) {
          sendConfigureRequest(connection, member, request);
        } else {
          // Complete the configure to the member.
          member.completeConfigure(); // 将configuring设置成false，表示configuring结束

          // Trigger reactions to the request failure.
          handleConfigureRequestFailure(member, request, error);
        }
      }
    });
  }

  /**
   * Sends a configuration message.
   */
  protected void sendConfigureRequest(Connection connection, MemberState member, ConfigureRequest request) {
    LOGGER.debug("{} - Sent {} to {}", context.getCluster().member().address(), request, member.getMember().serverAddress());
    connection.<ConfigureRequest, ConfigureResponse>send(request).whenComplete((response, error) -> {
      context.checkThread();

      // Complete the configure to the member.
      member.completeConfigure();

      if (open) {
        if (error == null) {
          LOGGER.debug("{} - Received {} from {}", context.getCluster().member().address(), response, member.getMember().serverAddress());
          handleConfigureResponse(member, request, response);
        } else {
          LOGGER.warn("{} - Failed to configure {}", context.getCluster().member().address(), member.getMember().serverAddress());
          handleConfigureResponseFailure(member, request, error);
        }
      }
    });
  }

 

LeaderAppender

@Override
  protected void handleConfigureResponse(MemberState member, ConfigureRequest request, ConfigureResponse response) {
    // Trigger commit futures if necessary.
    updateHeartbeatTime(member, null); // 判断是否大部分member都已经完成configuration同步

    super.handleConfigureResponse(member, request, response);
  }

 

AbstractAppender

 

/**
   * Handles a configuration response.
   */
  protected void handleConfigureResponse(MemberState member, ConfigureRequest request, ConfigureResponse response) {
    if (response.status() == Response.Status.OK) {
      handleConfigureResponseOk(member, request, response);
    } else {
      handleConfigureResponseError(member, request, response);
    }
  }

 

/**
   * Handles an OK configuration response.
   */
  @SuppressWarnings("unused")
  protected void handleConfigureResponseOk(MemberState member, ConfigureRequest request, ConfigureResponse response) {
    // Reset the member failure count and update the member's status if necessary.
    succeedAttempt(member);

    // Update the member's current configuration term and index according to the installed configuration.
    member.setConfigTerm(request.term()).setConfigIndex(request.index());

    // Recursively append entries to the member.
    appendEntries(member);
  }

 

在server上收到ConfigureRequest

ServerContext

public void connectServer(Connection connection) {

  connection.handler(ConfigureRequest.class, request -> state.configure(request));
 

在connectServer中，说明configuration只能在server间调用，client是不能调用的

configure，只在两个state中有实现

InactiveState

@Override
  public CompletableFuture<ConfigureResponse> configure(ConfigureRequest request) {
    context.checkThread();
    logRequest(request);
    updateTermAndLeader(request.term(), request.leader()); // 更新leader term

    Configuration configuration = new Configuration(request.index(), request.term(), request.timestamp(), request.members()); // 根据request创建Configuration

    // Configure the cluster membership. This will cause this server to transition to the
    // appropriate state if its type has changed.
    context.getClusterState().configure(configuration); // 调用ClusterState.configure

    // If the configuration is already committed, commit it to disk.
    // Check against the actual cluster Configuration rather than the received configuration in
    // case the received configuration was an older configuration that was not applied.
    if (context.getCommitIndex() >= context.getClusterState().getConfiguration().index()) {
      context.getClusterState().commit(); // 调用commit
    }

    return CompletableFuture.completedFuture(logResponse(ConfigureResponse.builder()
      .withStatus(Response.Status.OK) // 返回
      .build()));
  }

 

FollowerState

public CompletableFuture<ConfigureResponse> configure(ConfigureRequest request) {
    CompletableFuture<ConfigureResponse> future = super.configure(request);
    resetHeartbeatTimeout();
    return future;
  }

可以看到Follower里面只是多了resetHB

 

继续，

可以看到首先是ClusterState.configure

ClusterState中保存了当前server，所知道cluster的所有信息，其中包含Configuration对象

final class ClusterState implements Cluster, AutoCloseable {
  private final ServerContext context;
  private final ServerMember member;
  private volatile Configuration configuration;

configure

/**
   * Configures the cluster state.
   *
   * @param configuration The cluster configuration.
   * @return The cluster state.
   */
  ClusterState configure(Configuration configuration) {
    // If the configuration index is less than the currently configured index, ignore it.
    // Configurations can be persisted and applying old configurations can revert newer configurations.
    if (this.configuration != null && configuration.index() <= this.configuration.index()) //如果是老的configuration，丢弃
      return this;

    Instant time = Instant.ofEpochMilli(configuration.time());

    // Iterate through members in the new configuration, add any missing members, and update existing members.
    boolean transition = false;
    for (Member member : configuration.members()) {
      if (member.equals(this.member)) { //如果有我的配置变更
        transition = this.member.type().ordinal() < member.type().ordinal(); //看下type是否promote，只有promote才要transition，不让demote
        this.member.update(member.type(), time).update(member.clientAddress(), time);
        members.add(this.member);
      } else { //如果是更新其他member的配置
        // If the member state doesn't already exist, create it.
        MemberState state = membersMap.get(member.id());
        if (state == null) { //如果是新member，初始化
          state = new MemberState(new ServerMember(member.type(), member.serverAddress(), member.clientAddress(), time), this);
          state.resetState(context.getLog());
          this.members.add(state.getMember());
          this.remoteMembers.add(state);
          membersMap.put(member.id(), state);
          addressMap.put(member.address(), state);
          joinListeners.accept(state.getMember());
        }

        // If the member type has changed, update the member type and reset its state.
        state.getMember().update(member.clientAddress(), time);
        if (state.getMember().type() != member.type()) { //如果member的type发生了变化，更新数据
          state.getMember().update(member.type(), time);
          state.resetState(context.getLog());
        }

        // If the member status has changed, update the local member status.
        if (state.getMember().status() != member.status()) { //如果status发送变化，更新
          state.getMember().update(member.status(), time);
        }

        // Update the optimized member collections according to the member type.
        for (List<MemberState> memberType : memberTypes.values()) {
          memberType.remove(state);
        }

        List<MemberState> memberType = memberTypes.get(member.type());
        if (memberType == null) {
          memberType = new CopyOnWriteArrayList<>();
          memberTypes.put(member.type(), memberType);
        }
        memberType.add(state);
      }
    }

    // Transition the local member only if the member is being promoted and not demoted.
    // Configuration changes that demote the local member are only applied to the local server
    // upon commitment. This ensures that e.g. a leader that's removing itself from the quorum
    // can commit the configuration change prior to shutting down.
    if (transition) {  //做state transition
      context.transition(this.member.type());
    }

    // Iterate through configured members and remove any that no longer exist in the configuration.
    int i = 0;
    while (i < this.remoteMembers.size()) {
      MemberState member = this.remoteMembers.get(i);
      if (!configuration.members().contains(member.getMember())) {
        this.members.remove(member.getMember());
        this.remoteMembers.remove(i);
        for (List<MemberState> memberType : memberTypes.values()) {
          memberType.remove(member);
        }
        membersMap.remove(member.getMember().id());
        addressMap.remove(member.getMember().address());
        leaveListeners.accept(member.getMember());
      } else {
        i++;
      }
    }

    // If the local member was removed from the cluster, remove it from the members list.
    if (!configuration.members().contains(member)) {
      members.remove(member);
    }

    this.configuration = configuration; //更新configuration对象

    // Store the configuration if it's already committed.
    if (context.getCommitIndex() >= configuration.index()) { //如果这个configuration已经被commit，store到存储上
      context.getMetaStore().storeConfiguration(configuration);
    }

    // Reassign members based on availability.
    reassign(); //更新passive member的assignment，因为member变了，所以follower所对应的passive可能需要重新分配

    return this;
  }

 

ClusterState.commit

/**
   * Commit the current configuration to disk.
   *
   * @return The cluster state.
   */
  ClusterState commit() {
    // Apply the configuration to the local server state.
    context.transition(member.type());
    if (!configuration.members().contains(member) && leaveFuture != null) {
      leaveFuture.complete(null);
    }

    // If the local stored configuration is older than the committed configuration, overwrite it.
    if (context.getMetaStore().loadConfiguration().index() < configuration.index()) {
      context.getMetaStore().storeConfiguration(configuration);
    }
    return this;
  }

逻辑，就是把configuration存入本地盘

 

 

reconfigure

在什么地方被调用，

 

ServerMember

@Override
  public CompletableFuture<Void> promote() {
    return configure(Type.values()[type.ordinal() + 1]);
  }

可以用reconfigure来更新配置，比如promote，demote

 

/**
   * Demotes the server to the given type.
   */
  CompletableFuture<Void> configure(Member.Type type) {
    CompletableFuture<Void> future = new CompletableFuture<>();
    cluster.getContext().getThreadContext().executor().execute(() -> configure(type, future));
    return future;
  }

 

/**
   * Recursively reconfigures the cluster.
   */
  private void configure(Member.Type type, CompletableFuture<Void> future) {
    // Set a timer to retry the attempt to leave the cluster.
    configureTimeout = cluster.getContext().getThreadContext().schedule(cluster.getContext().getElectionTimeout(), () -> {
      configure(type, future); //设定schedule，反复重试
    });

    // Attempt to leave the cluster by submitting a LeaveRequest directly to the server state.
    // Non-leader states should forward the request to the leader if there is one. Leader states
    // will log, replicate, and commit the reconfiguration.
    cluster.getContext().getServerState().reconfigure(ReconfigureRequest.builder()
      .withIndex(cluster.getConfiguration().index())
      .withTerm(cluster.getConfiguration().term())
      .withMember(new ServerMember(type, serverAddress(), clientAddress(), updated))
      .build()).whenComplete((response, error) -> {
      if (error == null) {
        if (response.status() == Response.Status.OK) {
          cancelConfigureTimer(); //如果成功就cancel掉schedule
          cluster.configure(new Configuration(response.index(), response.term(), response.timestamp(), response.members())); //更新clusterState中的configuration配置
          future.complete(null);
        } else if (response.error() == null || response.error() == CopycatError.Type.NO_LEADER_ERROR) {
          cancelConfigureTimer();
          configureTimeout = cluster.getContext().getThreadContext().schedule(cluster.getContext().getElectionTimeout().multipliedBy(2), () -> configure(type, future));
        } else {
          cancelConfigureTimer();
          future.completeExceptionally(response.error().createException());
        }
      }
    });
  }

 

处理ReconfigureRequest

public void connectServer(Connection connection) {

  connection.handler(ReconfigureRequest.class, request -> state.reconfigure(request));

 

ReserveState

@Override
  public CompletableFuture<ReconfigureResponse> reconfigure(ReconfigureRequest request) {
    context.checkThread();
    logRequest(request);

    if (context.getLeader() == null) {
      return CompletableFuture.completedFuture(logResponse(ReconfigureResponse.builder()
        .withStatus(Response.Status.ERROR)
        .withError(CopycatError.Type.NO_LEADER_ERROR)
        .build()));
    } else {
      return this.<ReconfigureRequest, ReconfigureResponse>forward(request)
        .exceptionally(error -> ReconfigureResponse.builder()
          .withStatus(Response.Status.ERROR)
          .withError(CopycatError.Type.NO_LEADER_ERROR)
          .build())
        .thenApply(this::logResponse);
    }
  }

只是forward到leader

 

LeaderState

@Override
  public CompletableFuture<ReconfigureResponse> reconfigure(final ReconfigureRequest request) {
    // If another configuration change is already under way, reject the configuration.
    // If the leader index is 0 or is greater than the commitIndex, reject the promote requests.
    // Configuration changes should not be allowed until the leader has committed a no-op entry.
    // See https://groups.google.com/forum/#!topic/raft-dev/t4xj6dJTP6E
    if (configuring() || initializing()) { //如果正在configure或leader初始化，不能做配置变更
      return CompletableFuture.completedFuture(logResponse(ReconfigureResponse.builder()
        .withStatus(Response.Status.ERROR)
        .build()));
    }

    // If the configuration request index is less than the last known configuration index for
    // the leader, fail the request to ensure servers can't reconfigure an old configuration.
    if (request.index() > 0 && request.index() < context.getClusterState().getConfiguration().index() || request.term() != context.getClusterState().getConfiguration().term()
      && (existingMember.type() != request.member().type() || existingMember.status() != request.member().status())) {
      return CompletableFuture.completedFuture(logResponse(ReconfigureResponse.builder()
        .withStatus(Response.Status.ERROR)
        .withError(CopycatError.Type.CONFIGURATION_ERROR)
        .build()));
    }

    Member member = request.member();

    // If the client address is being set or has changed, update the configuration.
    if (member.clientAddress() != null && (existingMember.clientAddress() == null || !existingMember.clientAddress().equals(member.clientAddress()))) {
      existingMember.update(member.clientAddress(), Instant.now());
    }

    // Update the member type.
    existingMember.update(request.member().type(), Instant.now());

    Collection<Member> members = context.getCluster().members();

    CompletableFuture<ReconfigureResponse> future = new CompletableFuture<>();
    configure(members).whenComplete((index, error) -> { //调用configure
      context.checkThread();
      if (isOpen()) {
        if (error == null) { //如果成功，更新local配置
          future.complete(logResponse(ReconfigureResponse.builder()
            .withStatus(Response.Status.OK)
            .withIndex(index)
            .withTerm(context.getClusterState().getConfiguration().term())
            .withTime(context.getClusterState().getConfiguration().time())
            .withMembers(members)
            .build()));
        } else {
          future.complete(logResponse(ReconfigureResponse.builder()
            .withStatus(Response.Status.ERROR)
            .withError(CopycatError.Type.INTERNAL_ERROR)
            .build()));
        }
      }
    });
    return future;
  }