/**
* @param nodes Nodes.
* @param id ID.
* @throws IgniteCheckedException If failed.
*/
private void sendAllPartitions(
Collection<? extends ClusterNode> nodes, GridDhtPartitionExchangeId id)
throws IgniteCheckedException {
GridDhtPartitionsFullMessage m =
new GridDhtPartitionsFullMessage(id, lastVer.get(), id.topologyVersion());
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (!cacheCtx.isLocal()) {
AffinityTopologyVersion startTopVer = cacheCtx.startTopologyVersion();
boolean ready = startTopVer == null || startTopVer.compareTo(id.topologyVersion()) <= 0;
if (ready)
m.addFullPartitionsMap(cacheCtx.cacheId(), cacheCtx.topology().partitionMap(true));
}
}
// It is important that client topologies be added after contexts.
for (GridClientPartitionTopology top : cctx.exchange().clientTopologies())
m.addFullPartitionsMap(top.cacheId(), top.partitionMap(true));
if (log.isDebugEnabled())
log.debug(
"Sending full partition map [nodeIds="
+ F.viewReadOnly(nodes, F.node2id())
+ ", exchId="
+ exchId
+ ", msg="
+ m
+ ']');
cctx.io().safeSend(nodes, m, SYSTEM_POOL, null);
}
/**
* @param cacheId Cache ID.
* @return {@code True} if local client has been added.
*/
public boolean isLocalClientAdded(int cacheId) {
if (!F.isEmpty(reqs)) {
for (DynamicCacheChangeRequest req : reqs) {
if (req.start() && F.eq(req.initiatingNodeId(), cctx.localNodeId())) {
if (CU.cacheId(req.cacheName()) == cacheId) return true;
}
}
}
return false;
}
/** @param nodes Nodes. */
private Queue<ClusterNode> fallbacks(Collection<ClusterNode> nodes) {
Queue<ClusterNode> fallbacks = new LinkedList<>();
ClusterNode node = F.first(F.view(nodes, IS_LOC_NODE));
if (node != null) fallbacks.add(node);
fallbacks.addAll(node != null ? F.view(nodes, F.not(IS_LOC_NODE)) : nodes);
return fallbacks;
}
/**
* @param cacheId Cache ID to check.
* @param topVer Topology version.
* @return {@code True} if cache was added during this exchange.
*/
public boolean isCacheAdded(int cacheId, AffinityTopologyVersion topVer) {
if (!F.isEmpty(reqs)) {
for (DynamicCacheChangeRequest req : reqs) {
if (req.start() && !req.clientStartOnly()) {
if (CU.cacheId(req.cacheName()) == cacheId) return true;
}
}
}
GridCacheContext<?, ?> cacheCtx = cctx.cacheContext(cacheId);
return cacheCtx != null && F.eq(cacheCtx.startTopologyVersion(), topVer);
}
/** {@inheritDoc} */
@Override
public Collection<ClusterNode> nodes(int p, AffinityTopologyVersion topVer) {
Collection<ClusterNode> affNodes = cctx.affinity().nodes(p, topVer);
lock.readLock().lock();
try {
assert node2part != null && node2part.valid()
: "Invalid node-to-partitions map [topVer1="
+ topVer
+ ", topVer2="
+ this.topVer
+ ", cache="
+ cctx.name()
+ ", node2part="
+ node2part
+ ']';
Collection<ClusterNode> nodes = null;
Collection<UUID> nodeIds = part2node.get(p);
if (!F.isEmpty(nodeIds)) {
Collection<UUID> affIds = new HashSet<>(F.viewReadOnly(affNodes, F.node2id()));
for (UUID nodeId : nodeIds) {
if (!affIds.contains(nodeId) && hasState(p, nodeId, OWNING, MOVING, RENTING)) {
ClusterNode n = cctx.discovery().node(nodeId);
if (n != null
&& (topVer.topologyVersion() < 0 || n.order() <= topVer.topologyVersion())) {
if (nodes == null) {
nodes = new ArrayList<>(affNodes.size() + 2);
nodes.addAll(affNodes);
}
nodes.add(n);
}
}
}
}
return nodes != null ? nodes : affNodes;
} finally {
lock.readLock().unlock();
}
}
/** @return Nodes to execute on. */
private Collection<ClusterNode> nodes() {
CacheMode cacheMode = cctx.config().getCacheMode();
switch (cacheMode) {
case LOCAL:
if (prj != null)
U.warn(
log,
"Ignoring query projection because it's executed over LOCAL cache "
+ "(only local node will be queried): "
+ this);
return Collections.singletonList(cctx.localNode());
case REPLICATED:
if (prj != null || partition() != null) return nodes(cctx, prj, partition());
return cctx.affinityNode()
? Collections.singletonList(cctx.localNode())
: Collections.singletonList(F.rand(nodes(cctx, null, partition())));
case PARTITIONED:
return nodes(cctx, prj, partition());
default:
throw new IllegalStateException("Unknown cache distribution mode: " + cacheMode);
}
}
/**
* @param cctx Cache context.
* @param prj Projection (optional).
* @return Collection of data nodes in provided projection (if any).
*/
private static Collection<ClusterNode> nodes(
final GridCacheContext<?, ?> cctx,
@Nullable final ClusterGroup prj,
@Nullable final Integer part) {
assert cctx != null;
final AffinityTopologyVersion topVer = cctx.affinity().affinityTopologyVersion();
Collection<ClusterNode> affNodes = CU.affinityNodes(cctx);
if (prj == null && part == null) return affNodes;
final Set<ClusterNode> owners =
part == null
? Collections.<ClusterNode>emptySet()
: new HashSet<>(cctx.topology().owners(part, topVer));
return F.view(
affNodes,
new P1<ClusterNode>() {
@Override
public boolean apply(ClusterNode n) {
return cctx.discovery().cacheAffinityNode(n, cctx.name())
&& (prj == null || prj.node(n.id()) != null)
&& (part == null || owners.contains(n));
}
});
}
/** {@inheritDoc} */
@Override
public GridDhtPartitionMap localPartitionMap() {
lock.readLock().lock();
try {
return new GridDhtPartitionMap(
cctx.nodeId(), updateSeq.get(), F.viewReadOnly(locParts, CU.part2state()), true);
} finally {
lock.readLock().unlock();
}
}
/**
* Starts dynamic caches.
*
* @throws IgniteCheckedException If failed.
*/
private void startCaches() throws IgniteCheckedException {
cctx.cache()
.prepareCachesStart(
F.view(
reqs,
new IgnitePredicate<DynamicCacheChangeRequest>() {
@Override
public boolean apply(DynamicCacheChangeRequest req) {
return req.start();
}
}),
exchId.topologyVersion());
}
/**
* @param rmtReducer Optional reducer.
* @param rmtTransform Optional transformer.
* @param args Arguments.
* @return Future.
*/
@SuppressWarnings("IfMayBeConditional")
private <R> CacheQueryFuture<R> execute(
@Nullable IgniteReducer<T, R> rmtReducer,
@Nullable IgniteClosure<T, R> rmtTransform,
@Nullable Object... args) {
Collection<ClusterNode> nodes = nodes();
cctx.checkSecurity(SecurityPermission.CACHE_READ);
if (nodes.isEmpty())
return new GridCacheQueryErrorFuture<>(
cctx.kernalContext(), new ClusterGroupEmptyCheckedException());
if (log.isDebugEnabled())
log.debug("Executing query [query=" + this + ", nodes=" + nodes + ']');
if (cctx.deploymentEnabled()) {
try {
cctx.deploy().registerClasses(filter, rmtReducer, rmtTransform);
cctx.deploy().registerClasses(args);
} catch (IgniteCheckedException e) {
return new GridCacheQueryErrorFuture<>(cctx.kernalContext(), e);
}
}
if (subjId == null) subjId = cctx.localNodeId();
taskHash = cctx.kernalContext().job().currentTaskNameHash();
final GridCacheQueryBean bean =
new GridCacheQueryBean(
this,
(IgniteReducer<Object, Object>) rmtReducer,
(IgniteClosure<Object, Object>) rmtTransform,
args);
final GridCacheQueryManager qryMgr = cctx.queries();
boolean loc = nodes.size() == 1 && F.first(nodes).id().equals(cctx.localNodeId());
if (type == SQL_FIELDS || type == SPI)
return (CacheQueryFuture<R>)
(loc ? qryMgr.queryFieldsLocal(bean) : qryMgr.queryFieldsDistributed(bean, nodes));
else if (type == SCAN && part != null && nodes.size() > 1)
return new CacheQueryFallbackFuture<>(nodes, bean, qryMgr);
else
return (CacheQueryFuture<R>)
(loc ? qryMgr.queryLocal(bean) : qryMgr.queryDistributed(bean, nodes));
}
/**
* @param cacheId Cache ID to check.
* @return {@code True} if cache is stopping by this exchange.
*/
private boolean stopping(int cacheId) {
boolean stopping = false;
if (!F.isEmpty(reqs)) {
for (DynamicCacheChangeRequest req : reqs) {
if (cacheId == CU.cacheId(req.cacheName())) {
stopping = req.stop();
break;
}
}
}
return stopping;
}
/**
* @param nodeId Node ID.
* @param retryCnt Number of retries.
*/
private void sendAllPartitions(final UUID nodeId, final int retryCnt) {
ClusterNode n = cctx.node(nodeId);
try {
if (n != null) sendAllPartitions(F.asList(n), exchId);
} catch (IgniteCheckedException e) {
if (e instanceof ClusterTopologyCheckedException || !cctx.discovery().alive(n)) {
log.debug(
"Failed to send full partition map to node, node left grid "
+ "[rmtNode="
+ nodeId
+ ", exchangeId="
+ exchId
+ ']');
return;
}
if (retryCnt > 0) {
long timeout = cctx.gridConfig().getNetworkSendRetryDelay();
LT.error(
log,
e,
"Failed to send full partition map to node (will retry after timeout) "
+ "[node="
+ nodeId
+ ", exchangeId="
+ exchId
+ ", timeout="
+ timeout
+ ']');
cctx.time()
.addTimeoutObject(
new GridTimeoutObjectAdapter(timeout) {
@Override
public void onTimeout() {
sendAllPartitions(nodeId, retryCnt - 1);
}
});
} else
U.error(
log,
"Failed to send full partition map [node=" + n + ", exchangeId=" + exchId + ']',
e);
}
}
/**
* @param p Partition.
* @param topVer Topology version ({@code -1} for all nodes).
* @param state Partition state.
* @param states Additional partition states.
* @return List of nodes for the partition.
*/
private List<ClusterNode> nodes(
int p,
AffinityTopologyVersion topVer,
GridDhtPartitionState state,
GridDhtPartitionState... states) {
Collection<UUID> allIds =
topVer.topologyVersion() > 0 ? F.nodeIds(CU.affinityNodes(cctx, topVer)) : null;
lock.readLock().lock();
try {
assert node2part != null && node2part.valid()
: "Invalid node-to-partitions map [topVer="
+ topVer
+ ", allIds="
+ allIds
+ ", node2part="
+ node2part
+ ", cache="
+ cctx.name()
+ ']';
Collection<UUID> nodeIds = part2node.get(p);
// Node IDs can be null if both, primary and backup, nodes disappear.
int size = nodeIds == null ? 0 : nodeIds.size();
if (size == 0) return Collections.emptyList();
List<ClusterNode> nodes = new ArrayList<>(size);
for (UUID id : nodeIds) {
if (topVer.topologyVersion() > 0 && !allIds.contains(id)) continue;
if (hasState(p, id, state, states)) {
ClusterNode n = cctx.discovery().node(id);
if (n != null && (topVer.topologyVersion() < 0 || n.order() <= topVer.topologyVersion()))
nodes.add(n);
}
}
return nodes;
} finally {
lock.readLock().unlock();
}
}
/** @return {@code True} if succeeded. */
private boolean spreadPartitions() {
try {
sendAllPartitions(rmtNodes, exchId);
return true;
} catch (IgniteCheckedException e) {
scheduleRecheck();
if (!X.hasCause(e, InterruptedException.class))
U.error(
log,
"Failed to send full partition map to nodes (will retry after timeout) [nodes="
+ F.nodeId8s(rmtNodes)
+ ", exchangeId="
+ exchId
+ ']',
e);
return false;
}
}
/**
* Starts activity.
*
* @throws IgniteInterruptedCheckedException If interrupted.
*/
public void init() throws IgniteInterruptedCheckedException {
if (isDone()) return;
if (init.compareAndSet(false, true)) {
if (isDone()) return;
try {
// Wait for event to occur to make sure that discovery
// will return corresponding nodes.
U.await(evtLatch);
assert discoEvt != null : this;
assert !dummy && !forcePreload : this;
ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx, exchId.topologyVersion());
oldestNode.set(oldest);
startCaches();
// True if client node joined or failed.
boolean clientNodeEvt;
if (F.isEmpty(reqs)) {
int type = discoEvt.type();
assert type == EVT_NODE_JOINED || type == EVT_NODE_LEFT || type == EVT_NODE_FAILED
: discoEvt;
clientNodeEvt = CU.clientNode(discoEvt.eventNode());
} else {
assert discoEvt.type() == EVT_DISCOVERY_CUSTOM_EVT : discoEvt;
boolean clientOnlyStart = true;
for (DynamicCacheChangeRequest req : reqs) {
if (!req.clientStartOnly()) {
clientOnlyStart = false;
break;
}
}
clientNodeEvt = clientOnlyStart;
}
if (clientNodeEvt) {
ClusterNode node = discoEvt.eventNode();
// Client need to initialize affinity for local join event or for stated client caches.
if (!node.isLocal()) {
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
GridDhtPartitionTopology top = cacheCtx.topology();
top.updateTopologyVersion(exchId, this, -1, stopping(cacheCtx.cacheId()));
if (cacheCtx.affinity().affinityTopologyVersion() == AffinityTopologyVersion.NONE) {
initTopology(cacheCtx);
top.beforeExchange(this);
} else
cacheCtx.affinity().clientEventTopologyChange(discoEvt, exchId.topologyVersion());
}
if (exchId.isLeft())
cctx.mvcc().removeExplicitNodeLocks(exchId.nodeId(), exchId.topologyVersion());
onDone(exchId.topologyVersion());
skipPreload = cctx.kernalContext().clientNode();
return;
}
}
if (cctx.kernalContext().clientNode()) {
skipPreload = true;
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
GridDhtPartitionTopology top = cacheCtx.topology();
top.updateTopologyVersion(exchId, this, -1, stopping(cacheCtx.cacheId()));
}
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
initTopology(cacheCtx);
}
if (oldestNode.get() != null) {
rmtNodes =
new ConcurrentLinkedQueue<>(
CU.aliveRemoteServerNodesWithCaches(cctx, exchId.topologyVersion()));
rmtIds = Collections.unmodifiableSet(new HashSet<>(F.nodeIds(rmtNodes)));
ready.set(true);
initFut.onDone(true);
if (log.isDebugEnabled()) log.debug("Initialized future: " + this);
sendPartitions();
} else onDone(exchId.topologyVersion());
return;
}
assert oldestNode.get() != null;
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (isCacheAdded(cacheCtx.cacheId(), exchId.topologyVersion())) {
if (cacheCtx
.discovery()
.cacheAffinityNodes(cacheCtx.name(), topologyVersion())
.isEmpty())
U.quietAndWarn(log, "No server nodes found for cache client: " + cacheCtx.namex());
}
cacheCtx.preloader().onExchangeFutureAdded();
}
List<String> cachesWithoutNodes = null;
if (exchId.isLeft()) {
for (String name : cctx.cache().cacheNames()) {
if (cctx.discovery().cacheAffinityNodes(name, topologyVersion()).isEmpty()) {
if (cachesWithoutNodes == null) cachesWithoutNodes = new ArrayList<>();
cachesWithoutNodes.add(name);
// Fire event even if there is no client cache started.
if (cctx.gridEvents().isRecordable(EventType.EVT_CACHE_NODES_LEFT)) {
Event evt =
new CacheEvent(
name,
cctx.localNode(),
cctx.localNode(),
"All server nodes have left the cluster.",
EventType.EVT_CACHE_NODES_LEFT,
0,
false,
null,
null,
null,
null,
false,
null,
false,
null,
null,
null);
cctx.gridEvents().record(evt);
}
}
}
}
if (cachesWithoutNodes != null) {
StringBuilder sb =
new StringBuilder(
"All server nodes for the following caches have left the cluster: ");
for (int i = 0; i < cachesWithoutNodes.size(); i++) {
String cache = cachesWithoutNodes.get(i);
sb.append('\'').append(cache).append('\'');
if (i != cachesWithoutNodes.size() - 1) sb.append(", ");
}
U.quietAndWarn(log, sb.toString());
U.quietAndWarn(log, "Must have server nodes for caches to operate.");
}
assert discoEvt != null;
assert exchId.nodeId().equals(discoEvt.eventNode().id());
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
GridClientPartitionTopology clientTop =
cctx.exchange().clearClientTopology(cacheCtx.cacheId());
long updSeq = clientTop == null ? -1 : clientTop.lastUpdateSequence();
// Update before waiting for locks.
if (!cacheCtx.isLocal())
cacheCtx
.topology()
.updateTopologyVersion(exchId, this, updSeq, stopping(cacheCtx.cacheId()));
}
// Grab all alive remote nodes with order of equal or less than last joined node.
rmtNodes =
new ConcurrentLinkedQueue<>(
CU.aliveRemoteServerNodesWithCaches(cctx, exchId.topologyVersion()));
rmtIds = Collections.unmodifiableSet(new HashSet<>(F.nodeIds(rmtNodes)));
for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> m : singleMsgs.entrySet())
// If received any messages, process them.
onReceive(m.getKey(), m.getValue());
for (Map.Entry<UUID, GridDhtPartitionsFullMessage> m : fullMsgs.entrySet())
// If received any messages, process them.
onReceive(m.getKey(), m.getValue());
AffinityTopologyVersion topVer = exchId.topologyVersion();
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
// Must initialize topology after we get discovery event.
initTopology(cacheCtx);
cacheCtx.preloader().updateLastExchangeFuture(this);
}
IgniteInternalFuture<?> partReleaseFut = cctx.partitionReleaseFuture(topVer);
// Assign to class variable so it will be included into toString() method.
this.partReleaseFut = partReleaseFut;
if (log.isDebugEnabled()) log.debug("Before waiting for partition release future: " + this);
while (true) {
try {
partReleaseFut.get(2 * cctx.gridConfig().getNetworkTimeout(), TimeUnit.MILLISECONDS);
break;
} catch (IgniteFutureTimeoutCheckedException ignored) {
// Print pending transactions and locks that might have led to hang.
dumpPendingObjects();
}
}
if (log.isDebugEnabled()) log.debug("After waiting for partition release future: " + this);
if (!F.isEmpty(reqs)) blockGateways();
if (exchId.isLeft())
cctx.mvcc().removeExplicitNodeLocks(exchId.nodeId(), exchId.topologyVersion());
IgniteInternalFuture<?> locksFut = cctx.mvcc().finishLocks(exchId.topologyVersion());
while (true) {
try {
locksFut.get(2 * cctx.gridConfig().getNetworkTimeout(), TimeUnit.MILLISECONDS);
break;
} catch (IgniteFutureTimeoutCheckedException ignored) {
U.warn(
log,
"Failed to wait for locks release future. "
+ "Dumping pending objects that might be the cause: "
+ cctx.localNodeId());
U.warn(log, "Locked entries:");
Map<IgniteTxKey, Collection<GridCacheMvccCandidate>> locks =
cctx.mvcc().unfinishedLocks(exchId.topologyVersion());
for (Map.Entry<IgniteTxKey, Collection<GridCacheMvccCandidate>> e : locks.entrySet())
U.warn(log, "Locked entry [key=" + e.getKey() + ", mvcc=" + e.getValue() + ']');
}
}
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.isLocal()) continue;
// Notify replication manager.
GridCacheContext drCacheCtx =
cacheCtx.isNear() ? cacheCtx.near().dht().context() : cacheCtx;
if (drCacheCtx.isDrEnabled()) drCacheCtx.dr().beforeExchange(topVer, exchId.isLeft());
// Partition release future is done so we can flush the write-behind store.
cacheCtx.store().forceFlush();
// Process queued undeploys prior to sending/spreading map.
cacheCtx.preloader().unwindUndeploys();
GridDhtPartitionTopology top = cacheCtx.topology();
assert topVer.equals(top.topologyVersion())
: "Topology version is updated only in this class instances inside single ExchangeWorker thread.";
top.beforeExchange(this);
}
for (GridClientPartitionTopology top : cctx.exchange().clientTopologies()) {
top.updateTopologyVersion(exchId, this, -1, stopping(top.cacheId()));
top.beforeExchange(this);
}
} catch (IgniteInterruptedCheckedException e) {
onDone(e);
throw e;
} catch (Throwable e) {
U.error(
log,
"Failed to reinitialize local partitions (preloading will be stopped): " + exchId,
e);
onDone(e);
if (e instanceof Error) throw (Error) e;
return;
}
if (F.isEmpty(rmtIds)) {
onDone(exchId.topologyVersion());
return;
}
ready.set(true);
initFut.onDone(true);
if (log.isDebugEnabled()) log.debug("Initialized future: " + this);
// If this node is not oldest.
if (!oldestNode.get().id().equals(cctx.localNodeId())) sendPartitions();
else {
boolean allReceived = allReceived();
if (allReceived && replied.compareAndSet(false, true)) {
if (spreadPartitions()) onDone(exchId.topologyVersion());
}
}
scheduleRecheck();
} else assert false : "Skipped init future: " + this;
}
/** {@inheritDoc} */
@Override
public boolean afterExchange(GridDhtPartitionsExchangeFuture exchFut)
throws IgniteCheckedException {
boolean changed = waitForRent();
ClusterNode loc = cctx.localNode();
int num = cctx.affinity().partitions();
AffinityTopologyVersion topVer = exchFut.topologyVersion();
lock.writeLock().lock();
try {
if (stopping) return false;
assert topVer.equals(exchFut.topologyVersion())
: "Invalid topology version [topVer=" + topVer + ", exchId=" + exchFut.exchangeId() + ']';
if (log.isDebugEnabled())
log.debug(
"Partition map before afterExchange [exchId="
+ exchFut.exchangeId()
+ ", fullMap="
+ fullMapString()
+ ']');
long updateSeq = this.updateSeq.incrementAndGet();
for (int p = 0; p < num; p++) {
GridDhtLocalPartition locPart = localPartition(p, topVer, false, false);
if (cctx.affinity().localNode(p, topVer)) {
// This partition will be created during next topology event,
// which obviously has not happened at this point.
if (locPart == null) {
if (log.isDebugEnabled())
log.debug("Skipping local partition afterExchange (will not create): " + p);
continue;
}
GridDhtPartitionState state = locPart.state();
if (state == MOVING) {
if (cctx.rebalanceEnabled()) {
Collection<ClusterNode> owners = owners(p);
// If there are no other owners, then become an owner.
if (F.isEmpty(owners)) {
boolean owned = locPart.own();
assert owned
: "Failed to own partition [cacheName"
+ cctx.name()
+ ", locPart="
+ locPart
+ ']';
updateLocal(p, loc.id(), locPart.state(), updateSeq);
changed = true;
if (cctx.events().isRecordable(EVT_CACHE_REBALANCE_PART_DATA_LOST)) {
DiscoveryEvent discoEvt = exchFut.discoveryEvent();
cctx.events()
.addPreloadEvent(
p,
EVT_CACHE_REBALANCE_PART_DATA_LOST,
discoEvt.eventNode(),
discoEvt.type(),
discoEvt.timestamp());
}
if (log.isDebugEnabled()) log.debug("Owned partition: " + locPart);
} else if (log.isDebugEnabled())
log.debug(
"Will not own partition (there are owners to rebalance from) [locPart="
+ locPart
+ ", owners = "
+ owners
+ ']');
} else updateLocal(p, loc.id(), locPart.state(), updateSeq);
}
} else {
if (locPart != null) {
GridDhtPartitionState state = locPart.state();
if (state == MOVING) {
locPart.rent(false);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
changed = true;
if (log.isDebugEnabled())
log.debug("Evicting moving partition (it does not belong to affinity): " + locPart);
}
}
}
}
consistencyCheck();
} finally {
lock.writeLock().unlock();
}
return changed;
}
/** @return Remaining node IDs. */
Collection<UUID> remaining() {
if (rmtIds == null) return Collections.emptyList();
return F.lose(rmtIds, true, rcvdIds);
}
/**
* @param updateSeq Update sequence.
* @return Checks if any of the local partitions need to be evicted.
*/
private boolean checkEvictions(long updateSeq) {
assert lock.isWriteLockedByCurrentThread();
boolean changed = false;
UUID locId = cctx.nodeId();
for (GridDhtLocalPartition part : locParts.values()) {
GridDhtPartitionState state = part.state();
if (state.active()) {
int p = part.id();
List<ClusterNode> affNodes = cctx.affinity().nodes(p, topVer);
if (!affNodes.contains(cctx.localNode())) {
Collection<UUID> nodeIds = F.nodeIds(nodes(p, topVer, OWNING));
// If all affinity nodes are owners, then evict partition from local node.
if (nodeIds.containsAll(F.nodeIds(affNodes))) {
part.rent(false);
updateLocal(part.id(), locId, part.state(), updateSeq);
changed = true;
if (log.isDebugEnabled())
log.debug("Evicted local partition (all affinity nodes are owners): " + part);
} else {
int ownerCnt = nodeIds.size();
int affCnt = affNodes.size();
if (ownerCnt > affCnt) {
List<ClusterNode> sorted = new ArrayList<>(cctx.discovery().nodes(nodeIds));
// Sort by node orders in ascending order.
Collections.sort(sorted, CU.nodeComparator(true));
int diff = sorted.size() - affCnt;
for (int i = 0; i < diff; i++) {
ClusterNode n = sorted.get(i);
if (locId.equals(n.id())) {
part.rent(false);
updateLocal(part.id(), locId, part.state(), updateSeq);
changed = true;
if (log.isDebugEnabled())
log.debug(
"Evicted local partition (this node is oldest non-affinity node): " + part);
break;
}
}
}
}
}
}
}
return changed;
}
/** Clears values for this partition. */
private void clearAll() {
GridCacheVersion clearVer = cctx.versions().next();
boolean swap = cctx.isSwapOrOffheapEnabled();
boolean rec = cctx.events().isRecordable(EVT_CACHE_REBALANCE_OBJECT_UNLOADED);
Iterator<GridDhtCacheEntry> it = map.values().iterator();
GridCloseableIterator<Map.Entry<byte[], GridCacheSwapEntry>> swapIt = null;
if (swap
&& GridQueryProcessor.isEnabled(cctx.config())) { // Indexing needs to unswap cache values.
Iterator<GridDhtCacheEntry> unswapIt = null;
try {
swapIt = cctx.swap().iterator(id);
unswapIt = unswapIterator(swapIt);
} catch (Exception e) {
U.error(log, "Failed to clear swap for evicted partition: " + this, e);
}
if (unswapIt != null) it = F.concat(it, unswapIt);
}
try {
while (it.hasNext()) {
GridDhtCacheEntry cached = it.next();
try {
if (cached.clearInternal(clearVer, swap)) {
map.remove(cached.key(), cached);
if (!cached.isInternal()) {
mapPubSize.decrement();
if (rec)
cctx.events()
.addEvent(
cached.partition(),
cached.key(),
cctx.localNodeId(),
(IgniteUuid) null,
null,
EVT_CACHE_REBALANCE_OBJECT_UNLOADED,
null,
false,
cached.rawGet(),
cached.hasValue(),
null,
null,
null);
}
}
} catch (IgniteCheckedException e) {
U.error(log, "Failed to clear cache entry for evicted partition: " + cached, e);
}
}
} finally {
U.close(swapIt, log);
}
}
/** {@inheritDoc} */
@SuppressWarnings({"MismatchedQueryAndUpdateOfCollection"})
@Nullable
@Override
public GridDhtPartitionMap update(
@Nullable GridDhtPartitionExchangeId exchId, GridDhtPartitionMap parts) {
if (log.isDebugEnabled())
log.debug(
"Updating single partition map [exchId=" + exchId + ", parts=" + mapString(parts) + ']');
if (!cctx.discovery().alive(parts.nodeId())) {
if (log.isDebugEnabled())
log.debug(
"Received partition update for non-existing node (will ignore) [exchId="
+ exchId
+ ", parts="
+ parts
+ ']');
return null;
}
lock.writeLock().lock();
try {
if (stopping) return null;
if (lastExchangeId != null && exchId != null && lastExchangeId.compareTo(exchId) > 0) {
if (log.isDebugEnabled())
log.debug(
"Stale exchange id for single partition map update (will ignore) [lastExchId="
+ lastExchangeId
+ ", exchId="
+ exchId
+ ']');
return null;
}
if (exchId != null) lastExchangeId = exchId;
if (node2part == null)
// Create invalid partition map.
node2part = new GridDhtPartitionFullMap();
GridDhtPartitionMap cur = node2part.get(parts.nodeId());
if (cur != null && cur.updateSequence() >= parts.updateSequence()) {
if (log.isDebugEnabled())
log.debug(
"Stale update sequence for single partition map update (will ignore) [exchId="
+ exchId
+ ", curSeq="
+ cur.updateSequence()
+ ", newSeq="
+ parts.updateSequence()
+ ']');
return null;
}
long updateSeq = this.updateSeq.incrementAndGet();
node2part = new GridDhtPartitionFullMap(node2part, updateSeq);
boolean changed = false;
if (cur == null || !cur.equals(parts)) changed = true;
node2part.put(parts.nodeId(), parts);
part2node = new HashMap<>(part2node);
// Add new mappings.
for (Integer p : parts.keySet()) {
Set<UUID> ids = part2node.get(p);
if (ids == null)
// Initialize HashSet to size 3 in anticipation that there won't be
// more than 3 nodes per partition.
part2node.put(p, ids = U.newHashSet(3));
changed |= ids.add(parts.nodeId());
}
// Remove obsolete mappings.
if (cur != null) {
for (Integer p : F.view(cur.keySet(), F0.notIn(parts.keySet()))) {
Set<UUID> ids = part2node.get(p);
if (ids != null) changed |= ids.remove(parts.nodeId());
}
}
changed |= checkEvictions(updateSeq);
consistencyCheck();
if (log.isDebugEnabled()) log.debug("Partition map after single update: " + fullMapString());
return changed ? localPartitionMap() : null;
} finally {
lock.writeLock().unlock();
}
}