예제 #1
0
    @Override
    public State<?, ?> handle(
        ClusterContext context, Message<ClusterMessage> message, MessageHolder outgoing)
        throws Throwable {
      switch (message.getMessageType()) {
        case addClusterListener:
          {
            context.addClusterListener(message.<ClusterListener>getPayload());

            break;
          }

        case removeClusterListener:
          {
            context.removeClusterListener(message.<ClusterListener>getPayload());

            break;
          }

        case create:
          {
            String name = message.getPayload();
            context.getLogger(ClusterState.class).info("Creating cluster: " + name);
            context.created(name);
            return entered;
          }

        case join:
          {
            // Send configuration request to all instances
            Object[] args = message.<Object[]>getPayload();
            String name = (String) args[0];
            URI[] clusterInstanceUris = (URI[]) args[1];
            context.joining(name, Iterables.<URI, URI>iterable(clusterInstanceUris));

            for (URI potentialClusterInstanceUri : clusterInstanceUris) {
              outgoing.offer(to(ClusterMessage.configurationRequest, potentialClusterInstanceUri));
            }
            context.timeouts.setTimeout(
                "discovery",
                timeout(
                    ClusterMessage.configurationTimeout,
                    message,
                    new ClusterMessage.ConfigurationTimeoutState(4)));
            return acquiringConfiguration;
          }
      }
      return this;
    }
 @Override
 public void bookInstance(InstanceId instanceId, Message message) {
   if (message.getPayload() == null) {
     throw new IllegalArgumentException("null payload for booking instance: " + message);
   }
   bookedInstances.put(instanceId, message);
 }
예제 #3
0
    @Override
    public State<?, ?> handle(
        ClusterContext context, Message<ClusterMessage> message, MessageHolder outgoing)
        throws Throwable {
      switch (message.getMessageType()) {
        case configurationChanged:
          {
            ClusterMessage.ConfigurationChangeState state = message.getPayload();
            if (state.isLeaving(context.getMe())) {
              context.timeouts.cancelTimeout("leave");

              context.left();

              return start;
            } else {
              state.apply(context);
              return leaving;
            }
          }

        case leaveTimedout:
          {
            context
                .getLogger(ClusterState.class)
                .warn(
                    "Failed to leave. Cluster may consider this" + " instance still a " + "member");
            context.left();
            return start;
          }
      }

      return this;
    }
예제 #4
0
    @Override
    public State<?, ?> handle(
        ClusterContext context, Message<ClusterMessage> message, MessageHolder outgoing)
        throws Throwable {
      switch (message.getMessageType()) {
        case configurationChanged:
          {
            ClusterMessage.ConfigurationChangeState state = message.getPayload();

            if (context.getMe().equals(state.getJoin())) {
              context.timeouts.cancelTimeout("join");

              context.joined();
              outgoing.offer(
                  message.copyHeadersTo(
                      internal(ClusterMessage.joinResponse, context.getConfiguration())));
              return entered;
            } else {
              state.apply(context);
              return this;
            }
          }

        case joiningTimeout:
          {
            context
                .getLogger(ClusterState.class)
                .info("Join timeout for " + message.getHeader(Message.CONVERSATION_ID));

            // Go back to requesting configurations from potential members
            for (URI potentialClusterInstanceUri : context.getJoiningInstances()) {
              outgoing.offer(to(ClusterMessage.configurationRequest, potentialClusterInstanceUri));
            }
            context.timeouts.setTimeout(
                "discovery",
                timeout(
                    ClusterMessage.configurationTimeout,
                    message,
                    new ClusterMessage.ConfigurationTimeoutState(4)));

            return acquiringConfiguration;
          }

        case joinFailure:
          {
            // This causes an exception from the join() method
            return start;
          }
      }

      return this;
    }
예제 #5
0
    @Override
    public HeartbeatState handle(
        HeartbeatContext context, Message<HeartbeatMessage> message, MessageHolder outgoing)
        throws Throwable {
      switch (message.getMessageType()) {
        case addHeartbeatListener:
          {
            context.addHeartbeatListener((HeartbeatListener) message.getPayload());
            break;
          }

        case removeHeartbeatListener:
          {
            context.removeHeartbeatListener((HeartbeatListener) message.getPayload());
            break;
          }

        case join:
          {
            for (InstanceId instanceId : context.getClusterContext().getOtherInstances()) {
              // Setup heartbeat timeouts for the other instance
              context
                  .getClusterContext()
                  .timeouts
                  .setTimeout(
                      HeartbeatMessage.i_am_alive + "-" + instanceId,
                      timeout(HeartbeatMessage.timed_out, message, instanceId));

              // Send first heartbeat immediately
              outgoing.offer(timeout(HeartbeatMessage.sendHeartbeat, message, instanceId));
            }

            return heartbeat;
          }
      }

      return this;
    }
예제 #6
0
    @Override
    public HeartbeatState handle(
        HeartbeatContext context, Message<HeartbeatMessage> message, MessageHolder outgoing)
        throws Throwable {
      switch (message.getMessageType()) {
        case i_am_alive:
          {
            HeartbeatMessage.IAmAliveState state = message.getPayload();

            if (context.getClusterContext().isMe(state.getServer())) {
              break;
            }

            context.getClusterContext().getLogger(HeartbeatState.class).debug("Received " + state);

            if (state.getServer() == null) {
              break;
            }

            if (context.alive(state.getServer())) {
              // Send suspicions messages to all non-failed servers
              for (InstanceId aliveServer : context.getAlive()) {
                if (!aliveServer.equals(context.getClusterContext().getMyId())) {
                  URI aliveServerUri =
                      context.getClusterContext().configuration.getUriForId(aliveServer);
                  outgoing.offer(
                      Message.to(
                          HeartbeatMessage.suspicions,
                          aliveServerUri,
                          new HeartbeatMessage.SuspicionsState(
                              context.getSuspicionsFor(context.getClusterContext().getMyId()))));
                }
              }
            }

            context
                .getClusterContext()
                .timeouts
                .cancelTimeout(HeartbeatMessage.i_am_alive + "-" + state.getServer());
            context
                .getClusterContext()
                .timeouts
                .setTimeout(
                    HeartbeatMessage.i_am_alive + "-" + state.getServer(),
                    timeout(HeartbeatMessage.timed_out, message, state.getServer()));

            // Check if this server knows something that we don't
            if (message.hasHeader("last-learned")) {
              long lastLearned = Long.parseLong(message.getHeader("last-learned"));
              if (lastLearned
                  > context.getLearnerContext().getLastKnownLearnedInstanceInCluster()) {
                outgoing.offer(internal(LearnerMessage.catchUp, lastLearned));
              }
            }

            break;
          }

        case timed_out:
          {
            InstanceId server = message.getPayload();
            context
                .getClusterContext()
                .getLogger(HeartbeatState.class)
                .debug("Received timed out for server " + server);
            // Check if this node is no longer a part of the cluster
            if (context.getClusterContext().getConfiguration().getMembers().containsKey(server)) {
              context.suspect(server);

              context
                  .getClusterContext()
                  .timeouts
                  .setTimeout(
                      HeartbeatMessage.i_am_alive + "-" + server,
                      timeout(HeartbeatMessage.timed_out, message, server));

              // Send suspicions messages to all non-failed servers
              for (InstanceId aliveServer : context.getAlive()) {
                if (!aliveServer.equals(context.getClusterContext().getMyId())) {
                  URI sendTo =
                      context.getClusterContext().getConfiguration().getUriForId(aliveServer);
                  outgoing.offer(
                      Message.to(
                          HeartbeatMessage.suspicions,
                          sendTo,
                          new HeartbeatMessage.SuspicionsState(
                              context.getSuspicionsFor(context.getClusterContext().getMyId()))));
                }
              }
            } else {
              // If no longer part of cluster, then don't bother
              context.serverLeftCluster(server);
            }
            break;
          }

        case sendHeartbeat:
          {
            InstanceId to = message.getPayload();

            // Check if this node is no longer a part of the cluster
            if (!context.getClusterContext().isMe(to)
                && context.getClusterContext().getConfiguration().getMembers().containsKey(to)) {
              URI toSendTo = context.getClusterContext().getConfiguration().getUriForId(to);
              // Send heartbeat message to given server
              outgoing.offer(
                  to(
                          HeartbeatMessage.i_am_alive,
                          toSendTo,
                          new HeartbeatMessage.IAmAliveState(context.getClusterContext().getMyId()))
                      .setHeader(
                          "last-learned",
                          context.getLearnerContext().getLastLearnedInstanceId() + ""));

              // Set new timeout to send heartbeat to this host
              context
                  .getClusterContext()
                  .timeouts
                  .setTimeout(
                      HeartbeatMessage.sendHeartbeat + "-" + to,
                      timeout(HeartbeatMessage.sendHeartbeat, message, to));
            }
            break;
          }

        case reset_send_heartbeat:
          {
            InstanceId to = message.getPayload();

            if (!context.getClusterContext().isMe(to)) {
              String timeoutName = HeartbeatMessage.sendHeartbeat + "-" + to;
              context.getClusterContext().timeouts.cancelTimeout(timeoutName);
              context
                  .getClusterContext()
                  .timeouts
                  .setTimeout(timeoutName, timeout(HeartbeatMessage.sendHeartbeat, message, to));
            }
            break;
          }

        case suspicions:
          {
            HeartbeatMessage.SuspicionsState suspicions = message.getPayload();
            context
                .getClusterContext()
                .getLogger(HeartbeatState.class)
                .debug("Received suspicions as " + suspicions);

            URI from = new URI(message.getHeader(Message.FROM));
            InstanceId fromId = context.getClusterContext().getConfiguration().getServerId(from);
            /*
             * Remove ourselves from the suspicions received - we just received a message,
             * it's not normal to be considered failed. Whatever it was, it was transient and now it has
             * passed.
             */
            suspicions.getSuspicions().remove(context.getClusterContext().getMyId());
            context.suspicions(fromId, suspicions.getSuspicions());

            break;
          }

        case leave:
          {
            context.getClusterContext().getLogger(HeartbeatState.class).debug("Received leave");
            return start;
          }

        case addHeartbeatListener:
          {
            context.addHeartbeatListener((HeartbeatListener) message.getPayload());
            break;
          }

        case removeHeartbeatListener:
          {
            context.removeHeartbeatListener((HeartbeatListener) message.getPayload());
            break;
          }
      }

      return this;
    }
예제 #7
0
    @Override
    public AcceptorState handle(
        AcceptorContext context, Message<AcceptorMessage> message, MessageHolder outgoing)
        throws Throwable {
      switch (message.getMessageType()) {
        case prepare:
          {
            AcceptorMessage.PrepareState prepareState = message.getPayload();
            InstanceId instanceId = new InstanceId(message);
            AcceptorInstance instance = context.getAcceptorInstance(instanceId);

            if (prepareState.getBallot() >= instance.getBallot()) {
              context.promise(instance, prepareState.getBallot());

              outgoing.offer(
                  message.copyHeadersTo(
                      Message.respond(
                          ProposerMessage.promise,
                          message,
                          new ProposerMessage.PromiseState(
                              prepareState.getBallot(), instance.getValue())),
                      InstanceId.INSTANCE));
            } else {
              // Optimization - explicit reject
              context
                  .getLogger(AcceptorState.class)
                  .debug("Reject " + instanceId + " ballot:" + instance.getBallot());
              outgoing.offer(
                  message.copyHeadersTo(
                      Message.respond(
                          ProposerMessage.rejectPrepare,
                          message,
                          new ProposerMessage.RejectPrepare(instance.getBallot())),
                      InstanceId.INSTANCE));
            }
            break;
          }

        case accept:
          {
            // Task 4
            AcceptorMessage.AcceptState acceptState = message.getPayload();
            InstanceId instanceId = new InstanceId(message);
            AcceptorInstance instance = context.getAcceptorInstance(instanceId);

            if (acceptState.getBallot() == instance.getBallot()) {
              context.accept(instance, acceptState.getValue());
              instance.accept(acceptState.getValue());

              outgoing.offer(
                  message.copyHeadersTo(
                      Message.respond(
                          ProposerMessage.accepted, message, new ProposerMessage.AcceptedState()),
                      InstanceId.INSTANCE));
            } else {
              context
                  .getLogger(AcceptorState.class)
                  .debug(
                      "Reject "
                          + instanceId
                          + " accept ballot:"
                          + acceptState.getBallot()
                          + " actual ballot:"
                          + instance.getBallot());
              outgoing.offer(
                  message.copyHeadersTo(
                      Message.respond(
                          ProposerMessage.rejectAccept,
                          message,
                          new ProposerMessage.RejectAcceptState()),
                      InstanceId.INSTANCE));
            }
            break;
          }

        case leave:
          {
            context.leave();
            return start;
          }
      }

      return this;
    }
예제 #8
0
    @Override
    public State<?, ?> handle(
        ClusterContext context, Message<ClusterMessage> message, MessageHolder outgoing)
        throws Throwable {
      List<URI> discoveredInstances = context.getDiscoveredInstances();
      switch (message.getMessageType()) {
        case configurationResponse:
          {
            context.timeouts.cancelTimeout("discovery");

            ClusterMessage.ConfigurationResponseState state = message.getPayload();

            context.getLogger(ClusterState.class).info("Joining cluster " + state.getClusterName());
            if (!context.getConfiguration().getName().equals(state.getClusterName())) {
              context
                  .getLogger(ClusterState.class)
                  .warn(
                      "Joined cluster name is different than "
                          + "the one configured. Expected "
                          + context.getConfiguration().getName()
                          + ", got "
                          + state.getClusterName()
                          + ".");
            }

            List<URI> memberList = new ArrayList<URI>(state.getMembers());
            context.learnerContext.setLastDeliveredInstanceId(
                state.getLatestReceivedInstanceId().getId());
            context.learnerContext.learnedInstanceId(state.getLatestReceivedInstanceId().getId());
            context.proposerContext.nextInstanceId =
                state.getLatestReceivedInstanceId().getId() + 1;

            context.acquiredConfiguration(memberList, state.getRoles());

            if (!memberList.contains(context.me)) {
              context
                  .getLogger(ClusterState.class)
                  .info(
                      String.format(
                          "%s joining:%s, " + "last delivered:%d",
                          context.me.toString(),
                          context.getConfiguration().toString(),
                          state.getLatestReceivedInstanceId().getId()));

              ClusterMessage.ConfigurationChangeState newState =
                  new ClusterMessage.ConfigurationChangeState();
              newState.join(context.me);

              // Let the coordinator propose this if possible
              URI coordinator = state.getRoles().get(ClusterConfiguration.COORDINATOR);
              if (coordinator != null) {
                outgoing.offer(to(ProposerMessage.propose, coordinator, newState));
              } else {
                outgoing.offer(
                    to(
                        ProposerMessage.propose,
                        new URI(message.getHeader(Message.FROM)),
                        newState));
              }

              context
                  .getLogger(ClusterState.class)
                  .debug("Setup join timeout for " + message.getHeader(Message.CONVERSATION_ID));
              context.timeouts.setTimeout(
                  "join",
                  timeout(
                      ClusterMessage.joiningTimeout,
                      message,
                      new URI(message.getHeader(Message.FROM))));

              return joining;
            } else {
              // Already in (probably due to crash of this server previously), go to entered state
              context.joined();
              outgoing.offer(internal(ClusterMessage.joinResponse, context.getConfiguration()));

              return entered;
            }
          }

        case configurationTimeout:
          {
            ClusterMessage.ConfigurationTimeoutState state = message.getPayload();
            if (state.getRemainingPings() > 0) {
              // Send out requests again
              for (URI potentialClusterInstanceUri : context.getJoiningInstances()) {
                outgoing.offer(
                    to(ClusterMessage.configurationRequest, potentialClusterInstanceUri));
              }
              context.timeouts.setTimeout(
                  "join",
                  timeout(
                      ClusterMessage.configurationTimeout,
                      message,
                      new ClusterMessage.ConfigurationTimeoutState(state.getRemainingPings() - 1)));
            } else {
              // No responses
              // Check if we picked up any other instances' requests during this phase
              if (!discoveredInstances.isEmpty()) {
                Collections.sort(discoveredInstances);
                /*
                 * The assumption here is that the lowest in the list of discovered instances
                 * will create the cluster. Keep in mind that this is run on all instances so
                 * everyone will pick the same one.
                 * If the one picked up is configured to not init a cluster then the timeout
                 * set in else{} will take care of that.
                 */
                if (discoveredInstances.get(0).compareTo(context.getMe()) >= 0) {
                  discoveredInstances.clear();

                  // I'm supposed to create the cluster - fail the join
                  outgoing.offer(
                      internal(
                          ClusterMessage.joinFailure,
                          new TimeoutException("Join failed, timeout waiting for configuration")));
                  return start;
                } else {
                  discoveredInstances.clear();

                  // Someone else is supposed to create the cluster - restart the join discovery
                  for (URI potentialClusterInstanceUri : context.getJoiningInstances()) {
                    outgoing.offer(
                        to(ClusterMessage.configurationRequest, potentialClusterInstanceUri));
                  }
                  context.timeouts.setTimeout(
                      "discovery",
                      timeout(
                          ClusterMessage.configurationTimeout,
                          message,
                          new ClusterMessage.ConfigurationTimeoutState(4)));
                }
              } else {
                // Join failed
                outgoing.offer(
                    internal(
                        ClusterMessage.joinFailure,
                        new TimeoutException("Join failed, timeout waiting for configuration")));
                return start;
              }
            }

            return this;
          }

        case configurationRequest:
          {
            // We're listening for existing clusters, but if all instances start up at the same time
            // and look for each other, this allows us to pick that up
            URI joiningInstanceUri = new URI(message.getHeader(Message.FROM));
            if (!discoveredInstances.contains(joiningInstanceUri)) {
              discoveredInstances.add(joiningInstanceUri);
            }
          }
      }

      return this;
    }
예제 #9
0
    @Override
    public State<?, ?> handle(
        ClusterContext context, Message<ClusterMessage> message, MessageHolder outgoing)
        throws Throwable {
      switch (message.getMessageType()) {
        case addClusterListener:
          {
            context.addClusterListener(message.<ClusterListener>getPayload());

            break;
          }

        case removeClusterListener:
          {
            context.removeClusterListener(message.<ClusterListener>getPayload());

            break;
          }

        case configurationRequest:
          {
            outgoing.offer(
                message.copyHeadersTo(
                    respond(
                        ClusterMessage.configurationResponse,
                        message,
                        new ClusterMessage.ConfigurationResponseState(
                            context.getConfiguration().getRoles(),
                            context.getConfiguration().getMembers(),
                            new InstanceId(context.learnerContext.getLastDeliveredInstanceId()),
                            context.getConfiguration().getName()))));
            break;
          }

        case configurationChanged:
          {
            ClusterMessage.ConfigurationChangeState state = message.getPayload();
            state.apply(context);
            break;
          }

        case leave:
          {
            List<URI> nodeList = new ArrayList<URI>(context.getConfiguration().getMembers());
            if (nodeList.size() == 1) {
              context
                  .getLogger(ClusterState.class)
                  .info("Shutting down cluster: " + context.getConfiguration().getName());
              context.left();

              return start;

            } else {
              context.getLogger(ClusterState.class).info("Leaving:" + nodeList);

              ClusterMessage.ConfigurationChangeState newState =
                  new ClusterMessage.ConfigurationChangeState();
              newState.leave(context.me);

              outgoing.offer(internal(AtomicBroadcastMessage.broadcast, newState));
              context.timeouts.setTimeout("leave", timeout(ClusterMessage.leaveTimedout, message));

              return leaving;
            }
          }
      }

      return this;
    }