private ClusterState handleJoinRequest(final DiscoveryNode node) {
if (!master) {
throw new ElasticsearchIllegalStateException(
"Node [" + localNode + "] not master for join request from [" + node + "]");
}
ClusterState state = clusterService.state();
if (!transportService.addressSupported(node.address().getClass())) {
// TODO, what should we do now? Maybe inform that node that its crap?
logger.warn("received a wrong address type from [{}], ignoring...", node);
} else {
// try and connect to the node, if it fails, we can raise an exception back to the client...
transportService.connectToNode(node);
state = clusterService.state();
// validate the join request, will throw a failure if it fails, which will get back to the
// node calling the join request
membership.sendValidateJoinRequestBlocking(node, state, pingTimeout);
clusterService.submitStateUpdateTask(
"zen-disco-receive(join from node[" + node + "])",
Priority.URGENT,
new ClusterStateUpdateTask() {
@Override
public ClusterState execute(ClusterState currentState) {
if (currentState.nodes().nodeExists(node.id())) {
// the node already exists in the cluster
logger.warn("received a join request for an existing node [{}]", node);
// still send a new cluster state, so it will be re published and possibly update
// the other node
return ClusterState.builder(currentState).build();
}
DiscoveryNodes.Builder builder = DiscoveryNodes.builder(currentState.nodes());
for (DiscoveryNode existingNode : currentState.nodes()) {
if (node.address().equals(existingNode.address())) {
builder.remove(existingNode.id());
logger.warn(
"received join request from node [{}], but found existing node {} with same address, removing existing node",
node,
existingNode);
}
}
latestDiscoNodes = builder.build();
// add the new node now (will update latestDiscoNodes on publish)
return ClusterState.builder(currentState)
.nodes(latestDiscoNodes.newNode(node))
.build();
}
@Override
public void onFailure(String source, Throwable t) {
logger.error("unexpected failure during [{}]", t, source);
}
});
}
return state;
}
@Override
protected void doStart() throws ElasticsearchException {
Map<String, String> nodeAttributes = discoveryNodeService.buildAttributes();
// note, we rely on the fact that its a new id each time we start, see FD and "kill -9" handling
final String nodeId = getNodeUUID(settings);
localNode =
new DiscoveryNode(
settings.get("name"),
nodeId,
transportService.boundAddress().publishAddress(),
nodeAttributes,
version);
latestDiscoNodes =
new DiscoveryNodes.Builder().put(localNode).localNodeId(localNode.id()).build();
nodesFD.updateNodes(latestDiscoNodes);
pingService.start();
// do the join on a different thread, the DiscoveryService waits for 30s anyhow till it is
// discovered
asyncJoinCluster();
}
private void innerJoinCluster() {
boolean retry = true;
while (retry) {
if (lifecycle.stoppedOrClosed()) {
return;
}
retry = false;
DiscoveryNode masterNode = findMaster();
if (masterNode == null) {
logger.trace("no masterNode returned");
retry = true;
continue;
}
if (localNode.equals(masterNode)) {
this.master = true;
nodesFD.start(); // start the nodes FD
clusterService.submitStateUpdateTask(
"zen-disco-join (elected_as_master)",
Priority.URGENT,
new ProcessedClusterStateUpdateTask() {
@Override
public ClusterState execute(ClusterState currentState) {
DiscoveryNodes.Builder builder =
new DiscoveryNodes.Builder()
.localNodeId(localNode.id())
.masterNodeId(localNode.id())
// put our local node
.put(localNode);
// update the fact that we are the master...
latestDiscoNodes = builder.build();
ClusterBlocks clusterBlocks =
ClusterBlocks.builder()
.blocks(currentState.blocks())
.removeGlobalBlock(NO_MASTER_BLOCK)
.build();
return ClusterState.builder(currentState)
.nodes(latestDiscoNodes)
.blocks(clusterBlocks)
.build();
}
@Override
public void onFailure(String source, Throwable t) {
logger.error("unexpected failure during [{}]", t, source);
}
@Override
public void clusterStateProcessed(
String source, ClusterState oldState, ClusterState newState) {
sendInitialStateEventIfNeeded();
}
});
} else {
this.master = false;
try {
// first, make sure we can connect to the master
transportService.connectToNode(masterNode);
} catch (Exception e) {
logger.warn("failed to connect to master [{}], retrying...", e, masterNode);
retry = true;
continue;
}
// send join request
try {
membership.sendJoinRequestBlocking(masterNode, localNode, pingTimeout);
} catch (Exception e) {
if (e instanceof ElasticsearchException) {
logger.info(
"failed to send join request to master [{}], reason [{}]",
masterNode,
((ElasticsearchException) e).getDetailedMessage());
} else {
logger.info(
"failed to send join request to master [{}], reason [{}]",
masterNode,
e.getMessage());
}
if (logger.isTraceEnabled()) {
logger.trace("detailed failed reason", e);
}
// failed to send the join request, retry
retry = true;
continue;
}
masterFD.start(masterNode, "initial_join");
// no need to submit the received cluster state, we will get it from the master when it
// publishes
// the fact that we joined
}
}
}
@Inject
public ZenDiscovery(
Settings settings,
ClusterName clusterName,
ThreadPool threadPool,
TransportService transportService,
ClusterService clusterService,
NodeSettingsService nodeSettingsService,
DiscoveryNodeService discoveryNodeService,
ZenPingService pingService,
Version version) {
super(settings);
this.clusterName = clusterName;
this.threadPool = threadPool;
this.clusterService = clusterService;
this.transportService = transportService;
this.discoveryNodeService = discoveryNodeService;
this.pingService = pingService;
this.version = version;
// also support direct discovery.zen settings, for cases when it gets extended
this.pingTimeout =
settings.getAsTime(
"discovery.zen.ping.timeout",
settings.getAsTime(
"discovery.zen.ping_timeout",
componentSettings.getAsTime(
"ping_timeout",
componentSettings.getAsTime("initial_ping_timeout", timeValueSeconds(3)))));
this.sendLeaveRequest = componentSettings.getAsBoolean("send_leave_request", true);
this.masterElectionFilterClientNodes =
settings.getAsBoolean("discovery.zen.master_election.filter_client", true);
this.masterElectionFilterDataNodes =
settings.getAsBoolean("discovery.zen.master_election.filter_data", false);
logger.debug(
"using ping.timeout [{}], master_election.filter_client [{}], master_election.filter_data [{}]",
pingTimeout,
masterElectionFilterClientNodes,
masterElectionFilterDataNodes);
this.electMaster = new ElectMasterService(settings);
nodeSettingsService.addListener(new ApplySettings());
this.masterFD = new MasterFaultDetection(settings, threadPool, transportService, this);
this.masterFD.addListener(new MasterNodeFailureListener());
this.nodesFD = new NodesFaultDetection(settings, threadPool, transportService);
this.nodesFD.addListener(new NodeFailureListener());
this.publishClusterState =
new PublishClusterStateAction(
settings, transportService, this, new NewClusterStateListener());
this.pingService.setNodesProvider(this);
this.membership =
new MembershipAction(settings, transportService, this, new MembershipListener());
transportService.registerHandler(
RejoinClusterRequestHandler.ACTION, new RejoinClusterRequestHandler());
}
