@Override
public boolean process(Message<? extends MessageType> message) {
if (!message.isInternal() && !message.getMessageType().equals(HeartbeatMessage.i_am_alive)) {
// We assume the FROM header always exists.
String from = message.getHeader(Message.FROM);
if (!from.equals(message.getHeader(Message.TO))) {
InstanceId theId;
if (message.hasHeader(Message.INSTANCE_ID)) {
// INSTANCE_ID is there since after 1.9.6
theId = new InstanceId(Integer.parseInt(message.getHeader(Message.INSTANCE_ID)));
} else {
theId = clusterContext.getConfiguration().getIdForUri(URI.create(from));
}
if (theId != null
&& clusterContext.getConfiguration().getMembers().containsKey(theId)
&& !clusterContext.isMe(theId)) {
output.offer(
message.copyHeadersTo(
Message.internal(
HeartbeatMessage.i_am_alive, new HeartbeatMessage.IAmAliveState(theId)),
Message.FROM,
Message.INSTANCE_ID));
}
}
}
return true;
}
@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;
}
public int tick() {
// Deliver messages whose delivery time has passed
now += tickDuration;
// logger.debug( "tick:"+now );
Iterator<MessageDelivery> iter = messageDeliveries.iterator();
while (iter.hasNext()) {
MessageDelivery messageDelivery = iter.next();
if (messageDelivery.getMessageDeliveryTime() <= now) {
long delay =
strategy.messageDelay(
messageDelivery.getMessage(), messageDelivery.getServer().toString());
if (delay != NetworkLatencyStrategy.LOST) {
messageDelivery.getServer().process(messageDelivery.getMessage());
}
iter.remove();
}
}
// Check and trigger timeouts
for (TestProtocolServer testServer : participants.values()) {
testServer.tick(now);
}
// Get all sent messages from all test servers
List<Message> messages = new ArrayList<Message>();
for (TestProtocolServer testServer : participants.values()) {
testServer.sendMessages(messages);
}
// Now send them and figure out latency
for (Message message : messages) {
String to = message.getHeader(Message.TO);
if (to.equals(Message.BROADCAST)) {
for (Map.Entry<String, TestProtocolServer> testServer : participants.entrySet()) {
if (!testServer.getKey().equals(message.getHeader(Message.FROM))) {
long delay = strategy.messageDelay(message, testServer.getKey());
if (delay == NetworkLatencyStrategy.LOST) {
logger.info("Broadcasted message to " + testServer.getKey() + " was lost");
} else {
logger.info("Broadcast to " + testServer.getKey() + ": " + message);
messageDeliveries.add(
new MessageDelivery(now + delay, message, testServer.getValue()));
}
}
}
} else {
long delay = 0;
if (message.getHeader(Message.TO).equals(message.getHeader(Message.FROM))) {
logger.info("Sending message to itself; zero latency");
} else {
delay = strategy.messageDelay(message, to);
}
if (delay == NetworkLatencyStrategy.LOST) {
logger.info("Send message to " + to + " was lost");
} else {
TestProtocolServer server = participants.get(to);
logger.info("Send to " + to + ": " + message);
messageDeliveries.add(new MessageDelivery(now + delay, message, server));
}
}
}
return messageDeliveries.size();
}
@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;
}
@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;
}