/**
* Calculates data nodes for replicated caches on unstable topology.
*
* @param cctx Cache context for main space.
* @param extraSpaces Extra spaces.
* @return Collection of all data nodes owning all the caches or {@code null} for retry.
*/
private Collection<ClusterNode> replicatedUnstableDataNodes(
final GridCacheContext<?, ?> cctx, List<String> extraSpaces) {
assert cctx.isReplicated() : cctx.name() + " must be replicated";
Set<ClusterNode> nodes = replicatedUnstableDataNodes(cctx);
if (F.isEmpty(nodes)) return null; // Retry.
if (!F.isEmpty(extraSpaces)) {
for (String extraSpace : extraSpaces) {
GridCacheContext<?, ?> extraCctx = cacheContext(extraSpace);
if (extraCctx.isLocal()) continue;
if (!extraCctx.isReplicated())
throw new CacheException(
"Queries running on replicated cache should not contain JOINs "
+ "with tables in partitioned caches [rCache="
+ cctx.name()
+ ", pCache="
+ extraSpace
+ "]");
Set<ClusterNode> extraOwners = replicatedUnstableDataNodes(extraCctx);
if (F.isEmpty(extraOwners)) return null; // Retry.
nodes.retainAll(extraOwners);
if (nodes.isEmpty()) return null; // Retry.
}
}
return nodes;
}
/**
* @param cctx Context.
* @param id Partition ID.
*/
@SuppressWarnings("ExternalizableWithoutPublicNoArgConstructor")
GridDhtLocalPartition(GridCacheContext cctx, int id) {
assert cctx != null;
this.id = id;
this.cctx = cctx;
log = U.logger(cctx.kernalContext(), logRef, this);
rent =
new GridFutureAdapter<Object>() {
@Override
public String toString() {
return "PartitionRentFuture [part=" + GridDhtLocalPartition.this + ", map=" + map + ']';
}
};
map = new ConcurrentHashMap8<>(cctx.config().getStartSize() / cctx.affinity().partitions());
int delQueueSize =
CU.isSystemCache(cctx.name())
? 100
: Math.max(MAX_DELETE_QUEUE_SIZE / cctx.affinity().partitions(), 20);
rmvQueue = new GridCircularBuffer<>(U.ceilPow2(delQueueSize));
}
/** {@inheritDoc} */
@Override
public GridDhtPartitionFullMap partitionMap(boolean onlyActive) {
lock.readLock().lock();
try {
assert node2part != null && node2part.valid()
: "Invalid node2part [node2part: "
+ node2part
+ ", cache="
+ cctx.name()
+ ", started="
+ cctx.started()
+ ", stopping="
+ stopping
+ ", locNodeId="
+ cctx.localNode().id()
+ ", locName="
+ cctx.gridName()
+ ']';
GridDhtPartitionFullMap m = node2part;
return new GridDhtPartitionFullMap(
m.nodeId(), m.nodeOrder(), m.updateSequence(), m, onlyActive);
} finally {
lock.readLock().unlock();
}
}
/** @param ctx Context. */
public GridNearAtomicCache(GridCacheContext<K, V> ctx) {
super(ctx);
int size =
CU.isSystemCache(ctx.name())
? 100
: Integer.getInteger(IGNITE_ATOMIC_CACHE_DELETE_HISTORY_SIZE, 1_000_000);
rmvQueue = new GridCircularBuffer<>(U.ceilPow2(size / 10));
}
/** {@inheritDoc} */
@Override
public AffinityTopologyVersion topologyVersion() {
lock.readLock().lock();
try {
assert topVer.topologyVersion() > 0
: "Invalid topology version [topVer=" + topVer + ", cacheName=" + cctx.name() + ']';
return topVer;
} finally {
lock.readLock().unlock();
}
}
/**
* Collects all the nodes owning all the partitions for the given replicated cache.
*
* @param cctx Cache context.
* @return Owning nodes or {@code null} if we can't find owners for some partitions.
*/
private Set<ClusterNode> replicatedUnstableDataNodes(GridCacheContext<?, ?> cctx) {
assert cctx.isReplicated() : cctx.name() + " must be replicated";
String space = cctx.name();
Set<ClusterNode> dataNodes = new HashSet<>(dataNodes(space, NONE));
if (dataNodes.isEmpty())
throw new CacheException("Failed to find data nodes for cache: " + space);
// Find all the nodes owning all the partitions for replicated cache.
for (int p = 0, parts = cctx.affinity().partitions(); p < parts; p++) {
List<ClusterNode> owners = cctx.topology().owners(p);
if (F.isEmpty(owners)) return null; // Retry.
dataNodes.retainAll(owners);
if (dataNodes.isEmpty()) return null; // Retry.
}
return dataNodes;
}
/** {@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();
}
}
/** {@inheritDoc} */
@Override
public void printMemoryStats(int threshold) {
X.println(
">>> Cache partition topology stats [grid="
+ cctx.gridName()
+ ", cache="
+ cctx.name()
+ ']');
for (GridDhtLocalPartition part : locParts.values()) {
int size = part.size();
if (size >= threshold)
X.println(">>> Local partition [part=" + part.id() + ", size=" + size + ']');
}
}
/**
* @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();
}
}
/** {@inheritDoc} */
@Override
public boolean onDone(AffinityTopologyVersion res, Throwable err) {
Map<Integer, Boolean> m = null;
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (cacheCtx.config().getTopologyValidator() != null && !CU.isSystemCache(cacheCtx.name())) {
if (m == null) m = new HashMap<>();
m.put(
cacheCtx.cacheId(),
cacheCtx.config().getTopologyValidator().validate(discoEvt.topologyNodes()));
}
}
cacheValidRes = m != null ? m : Collections.<Integer, Boolean>emptyMap();
cctx.cache().onExchangeDone(exchId.topologyVersion(), reqs, err);
cctx.exchange().onExchangeDone(this, err);
if (super.onDone(res, err) && !dummy && !forcePreload) {
if (log.isDebugEnabled())
log.debug(
"Completed partition exchange [localNode="
+ cctx.localNodeId()
+ ", exchange= "
+ this
+ ']');
initFut.onDone(err == null);
GridTimeoutObject timeoutObj = this.timeoutObj;
// Deschedule timeout object.
if (timeoutObj != null) cctx.kernalContext().timeout().removeTimeoutObject(timeoutObj);
if (exchId.isLeft()) {
for (GridCacheContext cacheCtx : cctx.cacheContexts())
cacheCtx.config().getAffinity().removeNode(exchId.nodeId());
}
return true;
}
return dummy;
}
/**
* 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;
}
/**
* Calculates partition mapping for partitioned cache on unstable topology.
*
* @param cctx Cache context for main space.
* @param extraSpaces Extra spaces.
* @return Partition mapping or {@code null} if we can't calculate it due to repartitioning and we
* need to retry.
*/
@SuppressWarnings("unchecked")
private Map<ClusterNode, IntArray> partitionedUnstableDataNodes(
final GridCacheContext<?, ?> cctx, List<String> extraSpaces) {
assert !cctx.isReplicated() && !cctx.isLocal() : cctx.name() + " must be partitioned";
final int partsCnt = cctx.affinity().partitions();
if (extraSpaces != null) { // Check correct number of partitions for partitioned caches.
for (String extraSpace : extraSpaces) {
GridCacheContext<?, ?> extraCctx = cacheContext(extraSpace);
if (extraCctx.isReplicated() || extraCctx.isLocal()) continue;
int parts = extraCctx.affinity().partitions();
if (parts != partsCnt)
throw new CacheException(
"Number of partitions must be the same for correct collocation [cache1="
+ cctx.name()
+ ", parts1="
+ partsCnt
+ ", cache2="
+ extraSpace
+ ", parts2="
+ parts
+ "]");
}
}
Set<ClusterNode>[] partLocs = new Set[partsCnt];
// Fill partition locations for main cache.
for (int p = 0, parts = cctx.affinity().partitions(); p < parts; p++) {
List<ClusterNode> owners = cctx.topology().owners(p);
if (F.isEmpty(owners)) {
if (!F.isEmpty(dataNodes(cctx.name(), NONE))) return null; // Retry.
throw new CacheException(
"Failed to find data nodes [cache=" + cctx.name() + ", part=" + p + "]");
}
partLocs[p] = new HashSet<>(owners);
}
if (extraSpaces != null) {
// Find owner intersections for each participating partitioned cache partition.
// We need this for logical collocation between different partitioned caches with the same
// affinity.
for (String extraSpace : extraSpaces) {
GridCacheContext<?, ?> extraCctx = cacheContext(extraSpace);
if (extraCctx.isReplicated() || extraCctx.isLocal()) continue;
for (int p = 0, parts = extraCctx.affinity().partitions(); p < parts; p++) {
List<ClusterNode> owners = extraCctx.topology().owners(p);
if (F.isEmpty(owners)) {
if (!F.isEmpty(dataNodes(extraSpace, NONE))) return null; // Retry.
throw new CacheException(
"Failed to find data nodes [cache=" + extraSpace + ", part=" + p + "]");
}
if (partLocs[p] == null) partLocs[p] = new HashSet<>(owners);
else {
partLocs[p].retainAll(owners); // Intersection of owners.
if (partLocs[p].isEmpty()) return null; // Intersection is empty -> retry.
}
}
}
// Filter nodes where not all the replicated caches loaded.
for (String extraSpace : extraSpaces) {
GridCacheContext<?, ?> extraCctx = cacheContext(extraSpace);
if (!extraCctx.isReplicated()) continue;
Set<ClusterNode> dataNodes = replicatedUnstableDataNodes(extraCctx);
if (F.isEmpty(dataNodes)) return null; // Retry.
for (Set<ClusterNode> partLoc : partLocs) {
partLoc.retainAll(dataNodes);
if (partLoc.isEmpty()) return null; // Retry.
}
}
}
// Collect the final partitions mapping.
Map<ClusterNode, IntArray> res = new HashMap<>();
// Here partitions in all IntArray's will be sorted in ascending order, this is important.
for (int p = 0; p < partLocs.length; p++) {
Set<ClusterNode> pl = partLocs[p];
assert !F.isEmpty(pl) : pl;
ClusterNode n = pl.size() == 1 ? F.first(pl) : F.rand(pl);
IntArray parts = res.get(n);
if (parts == null) res.put(n, parts = new IntArray());
parts.add(p);
}
return res;
}
/**
* @param topVer Topology version.
* @param cctx Cache context for main space.
* @param extraSpaces Extra spaces.
* @return Data nodes or {@code null} if repartitioning started and we need to retry..
*/
private Collection<ClusterNode> stableDataNodes(
AffinityTopologyVersion topVer, final GridCacheContext<?, ?> cctx, List<String> extraSpaces) {
String space = cctx.name();
Set<ClusterNode> nodes = new HashSet<>(dataNodes(space, topVer));
if (F.isEmpty(nodes)) throw new CacheException("Failed to find data nodes for cache: " + space);
if (!F.isEmpty(extraSpaces)) {
for (String extraSpace : extraSpaces) {
GridCacheContext<?, ?> extraCctx = cacheContext(extraSpace);
if (extraCctx.isLocal()) continue; // No consistency guaranties for local caches.
if (cctx.isReplicated() && !extraCctx.isReplicated())
throw new CacheException(
"Queries running on replicated cache should not contain JOINs "
+ "with partitioned tables [rCache="
+ cctx.name()
+ ", pCache="
+ extraSpace
+ "]");
Collection<ClusterNode> extraNodes = dataNodes(extraSpace, topVer);
if (F.isEmpty(extraNodes))
throw new CacheException("Failed to find data nodes for cache: " + extraSpace);
if (cctx.isReplicated() && extraCctx.isReplicated()) {
nodes.retainAll(extraNodes);
if (nodes.isEmpty()) {
if (isPreloadingActive(cctx, extraSpaces)) return null; // Retry.
else
throw new CacheException(
"Caches have distinct sets of data nodes [cache1="
+ cctx.name()
+ ", cache2="
+ extraSpace
+ "]");
}
} else if (!cctx.isReplicated() && extraCctx.isReplicated()) {
if (!extraNodes.containsAll(nodes))
if (isPreloadingActive(cctx, extraSpaces)) return null; // Retry.
else
throw new CacheException(
"Caches have distinct sets of data nodes [cache1="
+ cctx.name()
+ ", cache2="
+ extraSpace
+ "]");
} else if (!cctx.isReplicated() && !extraCctx.isReplicated()) {
if (extraNodes.size() != nodes.size() || !nodes.containsAll(extraNodes))
if (isPreloadingActive(cctx, extraSpaces)) return null; // Retry.
else
throw new CacheException(
"Caches have distinct sets of data nodes [cache1="
+ cctx.name()
+ ", cache2="
+ extraSpace
+ "]");
} else throw new IllegalStateException();
}
}
return nodes;
}
/** {@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;
}
/** {@inheritDoc} */
@Override
public void beforeExchange(GridDhtPartitionsExchangeFuture exchFut)
throws IgniteCheckedException {
waitForRent();
ClusterNode loc = cctx.localNode();
int num = cctx.affinity().partitions();
lock.writeLock().lock();
try {
GridDhtPartitionExchangeId exchId = exchFut.exchangeId();
if (stopping) return;
assert topVer.equals(exchId.topologyVersion())
: "Invalid topology version [topVer=" + topVer + ", exchId=" + exchId + ']';
if (exchId.isLeft()) removeNode(exchId.nodeId());
// In case if node joins, get topology at the time of joining node.
ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx.shared(), topVer);
assert oldest != null;
if (log.isDebugEnabled())
log.debug(
"Partition map beforeExchange [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
long updateSeq = this.updateSeq.incrementAndGet();
// If this is the oldest node.
if (oldest.id().equals(loc.id())
|| exchFut.isCacheAdded(cctx.cacheId(), exchId.topologyVersion())) {
if (node2part == null) {
node2part = new GridDhtPartitionFullMap(oldest.id(), oldest.order(), updateSeq);
if (log.isDebugEnabled())
log.debug(
"Created brand new full topology map on oldest node [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
} else if (!node2part.valid()) {
node2part =
new GridDhtPartitionFullMap(oldest.id(), oldest.order(), updateSeq, node2part, false);
if (log.isDebugEnabled())
log.debug(
"Created new full topology map on oldest node [exchId="
+ exchId
+ ", fullMap="
+ node2part
+ ']');
} else if (!node2part.nodeId().equals(loc.id())) {
node2part =
new GridDhtPartitionFullMap(oldest.id(), oldest.order(), updateSeq, node2part, false);
if (log.isDebugEnabled())
log.debug(
"Copied old map into new map on oldest node (previous oldest node left) [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
}
}
if (cctx.rebalanceEnabled()) {
for (int p = 0; p < num; p++) {
// If this is the first node in grid.
boolean added = exchFut.isCacheAdded(cctx.cacheId(), exchId.topologyVersion());
if ((oldest.id().equals(loc.id())
&& oldest.id().equals(exchId.nodeId())
&& exchId.isJoined())
|| added) {
assert exchId.isJoined() || added;
try {
GridDhtLocalPartition locPart = localPartition(p, topVer, true, false);
assert locPart != null;
boolean owned = locPart.own();
assert owned
: "Failed to own partition for oldest node [cacheName"
+ cctx.name()
+ ", part="
+ locPart
+ ']';
if (log.isDebugEnabled()) log.debug("Owned partition for oldest node: " + locPart);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition on oldest node (no need to create a partition "
+ "if it no longer belongs to local node: "
+ e.partition());
}
}
// If this is not the first node in grid.
else {
if (node2part != null && node2part.valid()) {
if (cctx.affinity().localNode(p, topVer)) {
try {
// This will make sure that all non-existing partitions
// will be created in MOVING state.
GridDhtLocalPartition locPart = localPartition(p, topVer, true, false);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition (no need to create a partition if it "
+ "no longer belongs to local node: "
+ e.partition());
}
}
}
// If this node's map is empty, we pre-create local partitions,
// so local map will be sent correctly during exchange.
else if (cctx.affinity().localNode(p, topVer)) {
try {
localPartition(p, topVer, true, false);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition (no need to pre-create a partition if it "
+ "no longer belongs to local node: "
+ e.partition());
}
}
}
}
} else {
// If preloader is disabled, then we simply clear out
// the partitions this node is not responsible for.
for (int p = 0; p < num; p++) {
GridDhtLocalPartition locPart = localPartition(p, topVer, false, false);
boolean belongs = cctx.affinity().localNode(p, topVer);
if (locPart != null) {
if (!belongs) {
GridDhtPartitionState state = locPart.state();
if (state.active()) {
locPart.rent(false);
updateLocal(p, loc.id(), locPart.state(), updateSeq);
if (log.isDebugEnabled())
log.debug(
"Evicting partition with rebalancing disabled "
+ "(it does not belong to affinity): "
+ locPart);
}
}
} else if (belongs) {
try {
// Pre-create partitions.
localPartition(p, topVer, true, false);
} catch (GridDhtInvalidPartitionException e) {
if (log.isDebugEnabled())
log.debug(
"Ignoring invalid partition with disabled rebalancer (no need to "
+ "pre-create a partition if it no longer belongs to local node: "
+ e.partition());
}
}
}
}
if (node2part != null && node2part.valid()) checkEvictions(updateSeq);
consistencyCheck();
if (log.isDebugEnabled())
log.debug(
"Partition map after beforeExchange [exchId="
+ exchId
+ ", fullMap="
+ fullMapString()
+ ']');
} finally {
lock.writeLock().unlock();
}
// Wait for evictions.
waitForRent();
}
/** {@inheritDoc} */
@SuppressWarnings("unchecked")
@Override
public void notifyCallback(
UUID nodeId, UUID routineId, Collection<?> objs, GridKernalContext ctx) {
assert nodeId != null;
assert routineId != null;
assert objs != null;
assert ctx != null;
Collection<CacheContinuousQueryEntry> entries = (Collection<CacheContinuousQueryEntry>) objs;
final GridCacheContext cctx = cacheContext(ctx);
for (CacheContinuousQueryEntry e : entries) {
GridCacheDeploymentManager depMgr = cctx.deploy();
ClassLoader ldr = depMgr.globalLoader();
if (ctx.config().isPeerClassLoadingEnabled()) {
GridDeploymentInfo depInfo = e.deployInfo();
if (depInfo != null) {
depMgr.p2pContext(
nodeId,
depInfo.classLoaderId(),
depInfo.userVersion(),
depInfo.deployMode(),
depInfo.participants(),
depInfo.localDeploymentOwner());
}
}
try {
e.unmarshal(cctx, ldr);
} catch (IgniteCheckedException ex) {
U.error(ctx.log(getClass()), "Failed to unmarshal entry.", ex);
}
}
final IgniteCache cache = cctx.kernalContext().cache().jcache(cctx.name());
Collection<CacheContinuousQueryEntry> entries0 = new ArrayList<>();
for (CacheContinuousQueryEntry e : entries) entries0.addAll(handleEvent(ctx, e));
Iterable<CacheEntryEvent<? extends K, ? extends V>> evts =
F.viewReadOnly(
entries0,
new C1<CacheContinuousQueryEntry, CacheEntryEvent<? extends K, ? extends V>>() {
@Override
public CacheEntryEvent<? extends K, ? extends V> apply(CacheContinuousQueryEntry e) {
return new CacheContinuousQueryEvent<>(cache, cctx, e);
}
},
new IgnitePredicate<CacheContinuousQueryEntry>() {
@Override
public boolean apply(CacheContinuousQueryEntry entry) {
return !entry.isFiltered();
}
});
locLsnr.onUpdated(evts);
}
/**
* @param cctx Cache context.
* @param qry Query.
* @param keepPortable Keep portable.
* @return Cursor.
*/
public Iterator<List<?>> query(
GridCacheContext<?, ?> cctx, GridCacheTwoStepQuery qry, boolean keepPortable) {
for (int attempt = 0; ; attempt++) {
if (attempt != 0) {
try {
Thread.sleep(attempt * 10); // Wait for exchange.
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new CacheException("Query was interrupted.", e);
}
}
long qryReqId = reqIdGen.incrementAndGet();
QueryRun r = new QueryRun();
r.pageSize = qry.pageSize() <= 0 ? GridCacheTwoStepQuery.DFLT_PAGE_SIZE : qry.pageSize();
r.idxs = new ArrayList<>(qry.mapQueries().size());
String space = cctx.name();
r.conn = (JdbcConnection) h2.connectionForSpace(space);
AffinityTopologyVersion topVer = h2.readyTopologyVersion();
List<String> extraSpaces = extraSpaces(space, qry.spaces());
Collection<ClusterNode> nodes;
// Explicit partition mapping for unstable topology.
Map<ClusterNode, IntArray> partsMap = null;
if (isPreloadingActive(cctx, extraSpaces)) {
if (cctx.isReplicated()) nodes = replicatedUnstableDataNodes(cctx, extraSpaces);
else {
partsMap = partitionedUnstableDataNodes(cctx, extraSpaces);
nodes = partsMap == null ? null : partsMap.keySet();
}
} else nodes = stableDataNodes(topVer, cctx, extraSpaces);
if (nodes == null) continue; // Retry.
assert !nodes.isEmpty();
if (cctx.isReplicated() || qry.explain()) {
assert qry.explain() || !nodes.contains(ctx.discovery().localNode())
: "We must be on a client node.";
// Select random data node to run query on a replicated data or get EXPLAIN PLAN from a
// single node.
nodes = Collections.singleton(F.rand(nodes));
}
int tblIdx = 0;
final boolean skipMergeTbl = !qry.explain() && qry.skipMergeTable();
for (GridCacheSqlQuery mapQry : qry.mapQueries()) {
GridMergeIndex idx;
if (!skipMergeTbl) {
GridMergeTable tbl;
try {
tbl = createMergeTable(r.conn, mapQry, qry.explain());
} catch (IgniteCheckedException e) {
throw new IgniteException(e);
}
idx = tbl.getScanIndex(null);
fakeTable(r.conn, tblIdx++).setInnerTable(tbl);
} else idx = GridMergeIndexUnsorted.createDummy();
for (ClusterNode node : nodes) idx.addSource(node.id());
r.idxs.add(idx);
}
r.latch = new CountDownLatch(r.idxs.size() * nodes.size());
runs.put(qryReqId, r);
try {
if (ctx.clientDisconnected()) {
throw new CacheException(
"Query was cancelled, client node disconnected.",
new IgniteClientDisconnectedException(
ctx.cluster().clientReconnectFuture(), "Client node disconnected."));
}
Collection<GridCacheSqlQuery> mapQrys = qry.mapQueries();
if (qry.explain()) {
mapQrys = new ArrayList<>(qry.mapQueries().size());
for (GridCacheSqlQuery mapQry : qry.mapQueries())
mapQrys.add(new GridCacheSqlQuery("EXPLAIN " + mapQry.query(), mapQry.parameters()));
}
if (nodes.size() != 1 || !F.first(nodes).isLocal()) { // Marshall params for remotes.
Marshaller m = ctx.config().getMarshaller();
for (GridCacheSqlQuery mapQry : mapQrys) mapQry.marshallParams(m);
}
boolean retry = false;
if (send(
nodes,
new GridQueryRequest(qryReqId, r.pageSize, space, mapQrys, topVer, extraSpaces, null),
partsMap)) {
awaitAllReplies(r, nodes);
Object state = r.state.get();
if (state != null) {
if (state instanceof CacheException) {
CacheException err = (CacheException) state;
if (err.getCause() instanceof IgniteClientDisconnectedException) throw err;
throw new CacheException("Failed to run map query remotely.", err);
}
if (state instanceof AffinityTopologyVersion) {
retry = true;
// If remote node asks us to retry then we have outdated full partition map.
h2.awaitForReadyTopologyVersion((AffinityTopologyVersion) state);
}
}
} else // Send failed.
retry = true;
Iterator<List<?>> resIter = null;
if (!retry) {
if (qry.explain()) return explainPlan(r.conn, space, qry);
if (skipMergeTbl) {
List<List<?>> res = new ArrayList<>();
assert r.idxs.size() == 1 : r.idxs;
GridMergeIndex idx = r.idxs.get(0);
Cursor cur = idx.findInStream(null, null);
while (cur.next()) {
Row row = cur.get();
int cols = row.getColumnCount();
List<Object> resRow = new ArrayList<>(cols);
for (int c = 0; c < cols; c++) resRow.add(row.getValue(c).getObject());
res.add(resRow);
}
resIter = res.iterator();
} else {
GridCacheSqlQuery rdc = qry.reduceQuery();
// Statement caching is prohibited here because we can't guarantee correct merge index
// reuse.
ResultSet res =
h2.executeSqlQueryWithTimer(
space, r.conn, rdc.query(), F.asList(rdc.parameters()), false);
resIter = new Iter(res);
}
}
for (GridMergeIndex idx : r.idxs) {
if (!idx.fetchedAll()) // We have to explicitly cancel queries on remote nodes.
send(nodes, new GridQueryCancelRequest(qryReqId), null);
}
if (retry) {
if (Thread.currentThread().isInterrupted())
throw new IgniteInterruptedCheckedException("Query was interrupted.");
continue;
}
return new GridQueryCacheObjectsIterator(resIter, cctx, keepPortable);
} catch (IgniteCheckedException | RuntimeException e) {
U.closeQuiet(r.conn);
if (e instanceof CacheException) throw (CacheException) e;
Throwable cause = e;
if (e instanceof IgniteCheckedException) {
Throwable disconnectedErr =
((IgniteCheckedException) e).getCause(IgniteClientDisconnectedException.class);
if (disconnectedErr != null) cause = disconnectedErr;
}
throw new CacheException("Failed to run reduce query locally.", cause);
} finally {
if (!runs.remove(qryReqId, r)) U.warn(log, "Query run was already removed: " + qryReqId);
if (!skipMergeTbl) {
for (int i = 0, mapQrys = qry.mapQueries().size(); i < mapQrys; i++)
fakeTable(null, i).setInnerTable(null); // Drop all merge tables.
}
}
}
}
