use of org.apache.ignite.internal.processors.cluster.ChangeGlobalStateFinishMessage in project ignite by apache.
the class GridDiscoveryManager method start.
/**
* {@inheritDoc}
*/
@Override
public void start() throws IgniteCheckedException {
long totSysMemory = -1;
try {
totSysMemory = U.<Long>property(os, "totalPhysicalMemorySize");
} catch (RuntimeException ignored) {
// No-op.
}
ctx.addNodeAttribute(ATTR_PHY_RAM, totSysMemory);
ctx.addNodeAttribute(ATTR_OFFHEAP_SIZE, requiredOffheap());
ctx.addNodeAttribute(ATTR_DATA_REGIONS_OFFHEAP_SIZE, configuredOffheap());
DiscoverySpi spi = getSpi();
discoOrdered = discoOrdered();
histSupported = historySupported();
isLocDaemon = ctx.isDaemon();
hasRslvrs = !ctx.config().isClientMode() && !F.isEmpty(ctx.config().getSegmentationResolvers());
segChkFreq = ctx.config().getSegmentCheckFrequency();
if (hasRslvrs) {
if (segChkFreq < 0)
throw new IgniteCheckedException("Segment check frequency cannot be negative: " + segChkFreq);
if (segChkFreq > 0 && segChkFreq < 2000)
U.warn(log, "Configuration parameter 'segmentCheckFrequency' is too low " + "(at least 2000 ms recommended): " + segChkFreq);
int segResAttemp = ctx.config().getSegmentationResolveAttempts();
if (segResAttemp < 1)
throw new IgniteCheckedException("Segment resolve attempts cannot be negative or zero: " + segResAttemp);
checkSegmentOnStart();
}
metricsUpdateTask = ctx.timeout().schedule(new MetricsUpdater(), METRICS_UPDATE_FREQ, METRICS_UPDATE_FREQ);
spi.setMetricsProvider(createMetricsProvider());
if (ctx.security().enabled()) {
if (isSecurityCompatibilityMode())
ctx.addNodeAttribute(ATTR_SECURITY_COMPATIBILITY_MODE, true);
spi.setAuthenticator(new DiscoverySpiNodeAuthenticator() {
@Override
public SecurityContext authenticateNode(ClusterNode node, SecurityCredentials cred) {
try {
return ctx.security().authenticateNode(node, cred);
} catch (IgniteCheckedException e) {
throw U.convertException(e);
}
}
@Override
public boolean isGlobalNodeAuthentication() {
return ctx.security().isGlobalNodeAuthentication();
}
});
}
spi.setListener(new DiscoverySpiListener() {
private long gridStartTime;
/**
* {@inheritDoc}
*/
@Override
public void onLocalNodeInitialized(ClusterNode locNode) {
for (IgniteInClosure<ClusterNode> lsnr : locNodeInitLsnrs) lsnr.apply(locNode);
if (locNode instanceof TcpDiscoveryNode) {
final TcpDiscoveryNode node = (TcpDiscoveryNode) locNode;
if (consistentId != null)
node.setConsistentId(consistentId);
}
}
@Override
public void onDiscovery(final int type, final long topVer, final ClusterNode node, final Collection<ClusterNode> topSnapshot, final Map<Long, Collection<ClusterNode>> snapshots, @Nullable DiscoverySpiCustomMessage spiCustomMsg) {
synchronized (discoEvtMux) {
onDiscovery0(type, topVer, node, topSnapshot, snapshots, spiCustomMsg);
}
}
/**
* @param type Event type.
* @param topVer Event topology version.
* @param node Event node.
* @param topSnapshot Topology snapsjot.
* @param snapshots Topology snapshots history.
* @param spiCustomMsg Custom event.
*/
private void onDiscovery0(final int type, final long topVer, final ClusterNode node, final Collection<ClusterNode> topSnapshot, final Map<Long, Collection<ClusterNode>> snapshots, @Nullable DiscoverySpiCustomMessage spiCustomMsg) {
DiscoveryCustomMessage customMsg = spiCustomMsg == null ? null : ((CustomMessageWrapper) spiCustomMsg).delegate();
if (skipMessage(type, customMsg))
return;
final ClusterNode locNode = localNode();
if (snapshots != null)
topHist = snapshots;
boolean verChanged;
if (type == EVT_NODE_METRICS_UPDATED)
verChanged = false;
else {
if (type != EVT_NODE_SEGMENTED && type != EVT_CLIENT_NODE_DISCONNECTED && type != EVT_CLIENT_NODE_RECONNECTED && type != EVT_DISCOVERY_CUSTOM_EVT) {
minorTopVer = 0;
verChanged = true;
} else
verChanged = false;
}
if (type == EVT_NODE_FAILED || type == EVT_NODE_LEFT) {
for (DiscoCache c : discoCacheHist.values()) c.updateAlives(node);
updateClientNodes(node.id());
}
boolean locJoinEvt = type == EVT_NODE_JOINED && node.id().equals(locNode.id());
ChangeGlobalStateFinishMessage stateFinishMsg = null;
if (type == EVT_NODE_FAILED || type == EVT_NODE_LEFT)
stateFinishMsg = ctx.state().onNodeLeft(node);
final AffinityTopologyVersion nextTopVer;
if (type == EVT_DISCOVERY_CUSTOM_EVT) {
assert customMsg != null;
boolean incMinorTopVer;
if (customMsg instanceof ChangeGlobalStateMessage) {
incMinorTopVer = ctx.state().onStateChangeMessage(new AffinityTopologyVersion(topVer, minorTopVer), (ChangeGlobalStateMessage) customMsg, discoCache());
} else if (customMsg instanceof ChangeGlobalStateFinishMessage) {
ctx.state().onStateFinishMessage((ChangeGlobalStateFinishMessage) customMsg);
Snapshot snapshot = topSnap.get();
// Topology version does not change, but need create DiscoCache with new state.
DiscoCache discoCache = snapshot.discoCache.copy(snapshot.topVer, ctx.state().clusterState());
topSnap.set(new Snapshot(snapshot.topVer, discoCache));
incMinorTopVer = false;
} else {
incMinorTopVer = ctx.cache().onCustomEvent(customMsg, new AffinityTopologyVersion(topVer, minorTopVer), node);
}
if (incMinorTopVer) {
minorTopVer++;
verChanged = true;
}
nextTopVer = new AffinityTopologyVersion(topVer, minorTopVer);
} else
nextTopVer = new AffinityTopologyVersion(topVer, minorTopVer);
ctx.cache().onDiscoveryEvent(type, customMsg, node, nextTopVer, ctx.state().clusterState());
if (type == EVT_DISCOVERY_CUSTOM_EVT) {
for (Class cls = customMsg.getClass(); cls != null; cls = cls.getSuperclass()) {
List<CustomEventListener<DiscoveryCustomMessage>> list = customEvtLsnrs.get(cls);
if (list != null) {
for (CustomEventListener<DiscoveryCustomMessage> lsnr : list) {
try {
lsnr.onCustomEvent(nextTopVer, node, customMsg);
} catch (Exception e) {
U.error(log, "Failed to notify direct custom event listener: " + customMsg, e);
}
}
}
}
}
DiscoCache discoCache;
// event notifications, since SPI notifies manager about all events from this listener.
if (verChanged) {
Snapshot snapshot = topSnap.get();
if (customMsg == null) {
discoCache = createDiscoCache(nextTopVer, ctx.state().clusterState(), locNode, topSnapshot);
} else if (customMsg instanceof ChangeGlobalStateMessage) {
discoCache = createDiscoCache(nextTopVer, ctx.state().pendingState((ChangeGlobalStateMessage) customMsg), locNode, topSnapshot);
} else
discoCache = customMsg.createDiscoCache(GridDiscoveryManager.this, nextTopVer, snapshot.discoCache);
discoCacheHist.put(nextTopVer, discoCache);
assert snapshot.topVer.compareTo(nextTopVer) < 0 : "Topology version out of order [this.topVer=" + topSnap + ", topVer=" + topVer + ", node=" + node + ", nextTopVer=" + nextTopVer + ", evt=" + U.gridEventName(type) + ']';
topSnap.set(new Snapshot(nextTopVer, discoCache));
} else
// Current version.
discoCache = discoCache();
final DiscoCache discoCache0 = discoCache;
// If this is a local join event, just save it and do not notify listeners.
if (locJoinEvt) {
if (gridStartTime == 0)
gridStartTime = getSpi().getGridStartTime();
topSnap.set(new Snapshot(nextTopVer, discoCache));
startLatch.countDown();
DiscoveryEvent discoEvt = new DiscoveryEvent();
discoEvt.node(ctx.discovery().localNode());
discoEvt.eventNode(node);
discoEvt.type(EVT_NODE_JOINED);
discoEvt.topologySnapshot(topVer, new ArrayList<>(F.view(topSnapshot, FILTER_NOT_DAEMON)));
discoWrk.discoCache = discoCache;
if (!isLocDaemon && !ctx.clientDisconnected()) {
ctx.cache().context().exchange().onLocalJoin(discoEvt, discoCache);
ctx.authentication().onLocalJoin();
}
IgniteInternalFuture<Boolean> transitionWaitFut = ctx.state().onLocalJoin(discoCache);
locJoin.onDone(new DiscoveryLocalJoinData(discoEvt, discoCache, transitionWaitFut, ctx.state().clusterState().active()));
return;
} else if (type == EVT_CLIENT_NODE_DISCONNECTED) {
assert locNode.isClient() : locNode;
assert node.isClient() : node;
((IgniteKernal) ctx.grid()).onDisconnected();
locJoin = new GridFutureAdapter<>();
registeredCaches.clear();
registeredCacheGrps.clear();
for (AffinityTopologyVersion histVer : discoCacheHist.keySet()) {
Object rmvd = discoCacheHist.remove(histVer);
assert rmvd != null : histVer;
}
topHist.clear();
topSnap.set(new Snapshot(AffinityTopologyVersion.ZERO, createDiscoCache(AffinityTopologyVersion.ZERO, ctx.state().clusterState(), locNode, Collections.singleton(locNode))));
} else if (type == EVT_CLIENT_NODE_RECONNECTED) {
assert locNode.isClient() : locNode;
assert node.isClient() : node;
boolean clusterRestarted = gridStartTime != getSpi().getGridStartTime();
gridStartTime = getSpi().getGridStartTime();
((IgniteKernal) ctx.grid()).onReconnected(clusterRestarted);
ctx.cache().context().exchange().onLocalJoin(localJoinEvent(), discoCache);
ctx.cluster().clientReconnectFuture().listen(new CI1<IgniteFuture<?>>() {
@Override
public void apply(IgniteFuture<?> fut) {
try {
fut.get();
discoWrk.addEvent(type, nextTopVer, node, discoCache0, topSnapshot, null);
} catch (IgniteException ignore) {
// No-op.
}
}
});
return;
}
if (type == EVT_CLIENT_NODE_DISCONNECTED || type == EVT_NODE_SEGMENTED || !ctx.clientDisconnected())
discoWrk.addEvent(type, nextTopVer, node, discoCache, topSnapshot, customMsg);
if (stateFinishMsg != null)
discoWrk.addEvent(EVT_DISCOVERY_CUSTOM_EVT, nextTopVer, node, discoCache, topSnapshot, stateFinishMsg);
}
});
spi.setDataExchange(new DiscoverySpiDataExchange() {
@Override
public DiscoveryDataBag collect(DiscoveryDataBag dataBag) {
assert dataBag != null;
assert dataBag.joiningNodeId() != null;
if (ctx.localNodeId().equals(dataBag.joiningNodeId())) {
for (GridComponent c : ctx.components()) c.collectJoiningNodeData(dataBag);
} else {
for (GridComponent c : ctx.components()) c.collectGridNodeData(dataBag);
}
return dataBag;
}
@Override
public void onExchange(DiscoveryDataBag dataBag) {
assert dataBag != null;
assert dataBag.joiningNodeId() != null;
if (ctx.localNodeId().equals(dataBag.joiningNodeId())) {
// NodeAdded msg reached joining node after round-trip over the ring.
IGridClusterStateProcessor stateProc = ctx.state();
stateProc.onGridDataReceived(dataBag.gridDiscoveryData(stateProc.discoveryDataType().ordinal()));
for (GridComponent c : ctx.components()) {
if (c.discoveryDataType() != null && c != stateProc)
c.onGridDataReceived(dataBag.gridDiscoveryData(c.discoveryDataType().ordinal()));
}
} else {
// Discovery data from newly joined node has to be applied to the current old node.
IGridClusterStateProcessor stateProc = ctx.state();
JoiningNodeDiscoveryData data0 = dataBag.newJoinerDiscoveryData(stateProc.discoveryDataType().ordinal());
assert data0 != null;
stateProc.onJoiningNodeDataReceived(data0);
for (GridComponent c : ctx.components()) {
if (c.discoveryDataType() != null && c != stateProc) {
JoiningNodeDiscoveryData data = dataBag.newJoinerDiscoveryData(c.discoveryDataType().ordinal());
if (data != null)
c.onJoiningNodeDataReceived(data);
}
}
}
}
});
startSpi();
registeredDiscoSpi = true;
try {
U.await(startLatch);
} catch (IgniteInterruptedException e) {
throw new IgniteCheckedException("Failed to start discovery manager (thread has been interrupted).", e);
}
// Start segment check worker only if frequency is greater than 0.
if (hasRslvrs && segChkFreq > 0) {
segChkWrk = new SegmentCheckWorker();
segChkThread = new IgniteThread(segChkWrk);
segChkThread.start();
}
locNode = spi.getLocalNode();
checkAttributes(discoCache().remoteNodes());
// Start discovery worker.
new IgniteThread(discoWrk).start();
if (log.isDebugEnabled())
log.debug(startInfo());
}
use of org.apache.ignite.internal.processors.cluster.ChangeGlobalStateFinishMessage in project ignite by apache.
the class GridDhtPartitionsExchangeFuture method finishExchangeOnCoordinator.
/**
* @param sndResNodes Additional nodes to send finish message to.
*/
private void finishExchangeOnCoordinator(@Nullable Collection<ClusterNode> sndResNodes) {
try {
AffinityTopologyVersion resTopVer = exchCtx.events().topologyVersion();
if (log.isInfoEnabled()) {
log.info("finishExchangeOnCoordinator [topVer=" + initialVersion() + ", resVer=" + resTopVer + ']');
}
Map<Integer, CacheGroupAffinityMessage> idealAffDiff = null;
if (exchCtx.mergeExchanges()) {
synchronized (mux) {
if (mergedJoinExchMsgs != null) {
for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> e : mergedJoinExchMsgs.entrySet()) {
msgs.put(e.getKey(), e.getValue());
updatePartitionSingleMap(e.getKey(), e.getValue());
}
}
}
assert exchCtx.events().hasServerJoin() || exchCtx.events().hasServerLeft();
exchCtx.events().processEvents(this);
if (exchCtx.events().hasServerLeft())
idealAffDiff = cctx.affinity().onServerLeftWithExchangeMergeProtocol(this);
else
cctx.affinity().onServerJoinWithExchangeMergeProtocol(this, true);
for (CacheGroupDescriptor desc : cctx.affinity().cacheGroups().values()) {
if (desc.config().getCacheMode() == CacheMode.LOCAL)
continue;
CacheGroupContext grp = cctx.cache().cacheGroup(desc.groupId());
GridDhtPartitionTopology top = grp != null ? grp.topology() : cctx.exchange().clientTopology(desc.groupId(), events().discoveryCache());
top.beforeExchange(this, true, true);
}
}
Map<Integer, CacheGroupAffinityMessage> joinedNodeAff = null;
for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> e : msgs.entrySet()) {
GridDhtPartitionsSingleMessage msg = e.getValue();
// Apply update counters after all single messages are received.
for (Map.Entry<Integer, GridDhtPartitionMap> entry : msg.partitions().entrySet()) {
Integer grpId = entry.getKey();
CacheGroupContext grp = cctx.cache().cacheGroup(grpId);
GridDhtPartitionTopology top = grp != null ? grp.topology() : cctx.exchange().clientTopology(grpId, events().discoveryCache());
CachePartitionPartialCountersMap cntrs = msg.partitionUpdateCounters(grpId, top.partitions());
if (cntrs != null)
top.collectUpdateCounters(cntrs);
}
Collection<Integer> affReq = msg.cacheGroupsAffinityRequest();
if (affReq != null) {
joinedNodeAff = CacheGroupAffinityMessage.createAffinityMessages(cctx, resTopVer, affReq, joinedNodeAff);
}
}
if (firstDiscoEvt.type() == EVT_DISCOVERY_CUSTOM_EVT) {
assert firstDiscoEvt instanceof DiscoveryCustomEvent;
if (activateCluster() || changedBaseline())
assignPartitionsStates();
DiscoveryCustomMessage discoveryCustomMessage = ((DiscoveryCustomEvent) firstDiscoEvt).customMessage();
if (discoveryCustomMessage instanceof DynamicCacheChangeBatch) {
if (exchActions != null) {
assignPartitionsStates();
Set<String> caches = exchActions.cachesToResetLostPartitions();
if (!F.isEmpty(caches))
resetLostPartitions(caches);
}
} else if (discoveryCustomMessage instanceof SnapshotDiscoveryMessage && ((SnapshotDiscoveryMessage) discoveryCustomMessage).needAssignPartitions())
assignPartitionsStates();
} else {
if (exchCtx.events().hasServerJoin())
assignPartitionsStates();
if (exchCtx.events().hasServerLeft())
detectLostPartitions(resTopVer);
}
if (!exchCtx.mergeExchanges() && forceAffReassignment)
idealAffDiff = cctx.affinity().onCustomEventWithEnforcedAffinityReassignment(this);
for (CacheGroupContext grpCtx : cctx.cache().cacheGroups()) {
if (!grpCtx.isLocal())
grpCtx.topology().applyUpdateCounters();
}
updateLastVersion(cctx.versions().last());
cctx.versions().onExchange(lastVer.get().order());
IgniteProductVersion minVer = exchCtx.events().discoveryCache().minimumNodeVersion();
GridDhtPartitionsFullMessage msg = createPartitionsMessage(true, minVer.compareToIgnoreTimestamp(PARTIAL_COUNTERS_MAP_SINCE) >= 0);
if (exchCtx.mergeExchanges()) {
assert !centralizedAff;
msg.resultTopologyVersion(resTopVer);
if (exchCtx.events().hasServerLeft())
msg.idealAffinityDiff(idealAffDiff);
} else if (forceAffReassignment)
msg.idealAffinityDiff(idealAffDiff);
msg.prepareMarshal(cctx);
synchronized (mux) {
finishState = new FinishState(crd.id(), resTopVer, msg);
state = ExchangeLocalState.DONE;
}
if (centralizedAff) {
assert !exchCtx.mergeExchanges();
IgniteInternalFuture<Map<Integer, Map<Integer, List<UUID>>>> fut = cctx.affinity().initAffinityOnNodeLeft(this);
if (!fut.isDone()) {
fut.listen(new IgniteInClosure<IgniteInternalFuture<Map<Integer, Map<Integer, List<UUID>>>>>() {
@Override
public void apply(IgniteInternalFuture<Map<Integer, Map<Integer, List<UUID>>>> fut) {
onAffinityInitialized(fut);
}
});
} else
onAffinityInitialized(fut);
} else {
Set<ClusterNode> nodes;
Map<UUID, GridDhtPartitionsSingleMessage> mergedJoinExchMsgs0;
synchronized (mux) {
srvNodes.remove(cctx.localNode());
nodes = U.newHashSet(srvNodes.size());
nodes.addAll(srvNodes);
mergedJoinExchMsgs0 = mergedJoinExchMsgs;
if (mergedJoinExchMsgs != null) {
for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> e : mergedJoinExchMsgs.entrySet()) {
if (e.getValue() != null) {
ClusterNode node = cctx.discovery().node(e.getKey());
if (node != null)
nodes.add(node);
}
}
}
if (!F.isEmpty(sndResNodes))
nodes.addAll(sndResNodes);
}
if (!nodes.isEmpty())
sendAllPartitions(msg, nodes, mergedJoinExchMsgs0, joinedNodeAff);
if (!stateChangeExchange())
onDone(exchCtx.events().topologyVersion(), null);
for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> e : pendingSingleMsgs.entrySet()) {
if (log.isInfoEnabled()) {
log.info("Process pending message on coordinator [node=" + e.getKey() + ", ver=" + initialVersion() + ", resVer=" + resTopVer + ']');
}
processSingleMessage(e.getKey(), e.getValue());
}
}
if (stateChangeExchange()) {
IgniteCheckedException err = null;
StateChangeRequest req = exchActions.stateChangeRequest();
assert req != null : exchActions;
boolean stateChangeErr = false;
if (!F.isEmpty(changeGlobalStateExceptions)) {
stateChangeErr = true;
err = new IgniteCheckedException("Cluster state change failed.");
cctx.kernalContext().state().onStateChangeError(changeGlobalStateExceptions, req);
} else {
boolean hasMoving = !partsToReload.isEmpty();
Set<Integer> waitGrps = cctx.affinity().waitGroups();
if (!hasMoving) {
for (CacheGroupContext grpCtx : cctx.cache().cacheGroups()) {
if (waitGrps.contains(grpCtx.groupId()) && grpCtx.topology().hasMovingPartitions()) {
hasMoving = true;
break;
}
}
}
cctx.kernalContext().state().onExchangeFinishedOnCoordinator(this, hasMoving);
}
boolean active = !stateChangeErr && req.activate();
ChangeGlobalStateFinishMessage stateFinishMsg = new ChangeGlobalStateFinishMessage(req.requestId(), active, !stateChangeErr);
cctx.discovery().sendCustomEvent(stateFinishMsg);
if (!centralizedAff)
onDone(exchCtx.events().topologyVersion(), err);
}
} catch (IgniteCheckedException e) {
if (reconnectOnError(e))
onDone(new IgniteNeedReconnectException(cctx.localNode(), e));
else
onDone(e);
}
}
Aggregations