/**
* @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);
}
/** {@inheritDoc} */
@Override
public boolean onDone(GridCacheTx tx, Throwable err) {
if ((initialized() || err != null) && super.onDone(tx, err)) {
if (error() instanceof GridCacheTxHeuristicException) {
long topVer = this.tx.topologyVersion();
for (GridCacheTxEntry<K, V> e : this.tx.writeMap().values()) {
try {
if (e.op() != NOOP && !cctx.affinity().localNode(e.key(), topVer)) {
GridCacheEntryEx<K, V> cacheEntry = cctx.cache().peekEx(e.key());
if (cacheEntry != null) cacheEntry.invalidate(null, this.tx.xidVersion());
}
} catch (Throwable t) {
U.error(log, "Failed to invalidate entry.", t);
if (t instanceof Error) throw (Error) t;
}
}
}
// Don't forget to clean up.
cctx.mvcc().removeFuture(this);
return true;
}
return false;
}
/**
* @param updateSeq Update sequence.
* @return {@code True} if entry has been transitioned to state EVICTED.
*/
boolean tryEvict(boolean updateSeq) {
if (state.getReference() != RENTING || state.getStamp() != 0 || groupReserved()) return false;
// Attempt to evict partition entries from cache.
clearAll();
if (map.isEmpty() && state.compareAndSet(RENTING, EVICTED, 0, 0)) {
if (log.isDebugEnabled()) log.debug("Evicted partition: " + this);
if (!GridQueryProcessor.isEnabled(cctx.config())) clearSwap();
if (cctx.isDrEnabled()) cctx.dr().partitionEvicted(id);
cctx.dataStructures().onPartitionEvicted(id);
rent.onDone();
((GridDhtPreloader) cctx.preloader()).onPartitionEvicted(this, updateSeq);
clearDeferredDeletes();
return true;
}
return false;
}
/**
* @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();
}
}
/**
* Updates partition map in all caches.
*
* @param msg Partitions single message.
*/
private void updatePartitionSingleMap(GridDhtPartitionsSingleMessage msg) {
for (Map.Entry<Integer, GridDhtPartitionMap> entry : msg.partitions().entrySet()) {
Integer cacheId = entry.getKey();
GridCacheContext cacheCtx = cctx.cacheContext(cacheId);
GridDhtPartitionTopology top =
cacheCtx != null ? cacheCtx.topology() : cctx.exchange().clientTopology(cacheId, this);
top.update(exchId, entry.getValue());
}
}
/**
* @param cacheCtx Cache context.
* @throws IgniteCheckedException If failed.
*/
private void initTopology(GridCacheContext cacheCtx) throws IgniteCheckedException {
if (stopping(cacheCtx.cacheId())) return;
if (canCalculateAffinity(cacheCtx)) {
if (log.isDebugEnabled())
log.debug(
"Will recalculate affinity [locNodeId="
+ cctx.localNodeId()
+ ", exchId="
+ exchId
+ ']');
cacheCtx.affinity().calculateAffinity(exchId.topologyVersion(), discoEvt);
} else {
if (log.isDebugEnabled())
log.debug(
"Will request affinity from remote node [locNodeId="
+ cctx.localNodeId()
+ ", exchId="
+ exchId
+ ']');
// Fetch affinity assignment from remote node.
GridDhtAssignmentFetchFuture fetchFut =
new GridDhtAssignmentFetchFuture(
cacheCtx,
exchId.topologyVersion(),
CU.affinityNodes(cacheCtx, exchId.topologyVersion()));
fetchFut.init();
List<List<ClusterNode>> affAssignment = fetchFut.get();
if (log.isDebugEnabled())
log.debug(
"Fetched affinity from remote node, initializing affinity assignment [locNodeId="
+ cctx.localNodeId()
+ ", topVer="
+ exchId.topologyVersion()
+ ']');
if (affAssignment == null) {
affAssignment = new ArrayList<>(cacheCtx.affinity().partitions());
List<ClusterNode> empty = Collections.emptyList();
for (int i = 0; i < cacheCtx.affinity().partitions(); i++) affAssignment.add(empty);
}
cacheCtx.affinity().initializeAffinity(exchId.topologyVersion(), affAssignment);
}
}
/**
* @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);
}
/**
* Updates partition map in all caches.
*
* @param msg Partitions full messages.
*/
private void updatePartitionFullMap(GridDhtPartitionsFullMessage msg) {
for (Map.Entry<Integer, GridDhtPartitionFullMap> entry : msg.partitions().entrySet()) {
Integer cacheId = entry.getKey();
GridCacheContext cacheCtx = cctx.cacheContext(cacheId);
if (cacheCtx != null) cacheCtx.topology().update(exchId, entry.getValue());
else {
ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx, AffinityTopologyVersion.NONE);
if (oldest != null && oldest.isLocal())
cctx.exchange().clientTopology(cacheId, this).update(exchId, entry.getValue());
}
}
}
/** {@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();
}
}
/**
* @param ctx Cache registry.
* @param implicit Implicit flag.
* @param implicitSingle Implicit with one key flag.
* @param concurrency Concurrency.
* @param isolation Isolation.
* @param timeout Timeout.
* @param invalidate Invalidation policy.
* @param syncCommit Synchronous commit flag.
* @param syncRollback Synchronous rollback flag.
* @param swapEnabled Whether to use swap storage.
* @param storeEnabled Whether to use read/write through.
*/
GridNearTxLocal(
GridCacheContext<K, V> ctx,
boolean implicit,
boolean implicitSingle,
GridCacheTxConcurrency concurrency,
GridCacheTxIsolation isolation,
long timeout,
boolean invalidate,
boolean syncCommit,
boolean syncRollback,
boolean swapEnabled,
boolean storeEnabled) {
super(
ctx,
ctx.versions().next(),
implicit,
implicitSingle,
concurrency,
isolation,
timeout,
invalidate,
swapEnabled,
storeEnabled);
assert ctx != null;
this.syncCommit = syncCommit;
this.syncRollback = syncRollback;
}
/**
* Constructor.
*
* @param name Latch name.
* @param cnt Current count.
* @param initCnt Initial count.
* @param autoDel Auto delete flag.
* @param key Latch key.
* @param latchView Latch projection.
* @param ctx Cache context.
*/
public GridCacheCountDownLatchImpl(
String name,
int cnt,
int initCnt,
boolean autoDel,
GridCacheInternalKey key,
GridCacheProjection<GridCacheInternalKey, GridCacheCountDownLatchValue> latchView,
GridCacheContext ctx) {
assert name != null;
assert cnt >= 0;
assert initCnt >= 0;
assert key != null;
assert latchView != null;
assert ctx != null;
this.name = name;
this.cnt = cnt;
this.initCnt = initCnt;
this.autoDel = autoDel;
this.key = key;
this.latchView = latchView;
this.ctx = ctx;
log = ctx.gridConfig().getGridLogger().getLogger(getClass());
}
/** @param cctx Context. */
GridDhtPartitionTopologyImpl(GridCacheContext<?, ?> cctx) {
assert cctx != null;
this.cctx = cctx;
log = cctx.logger(getClass());
}
/** {@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();
}
}
/** @param m Mapping. */
@SuppressWarnings({"unchecked"})
private void finish(GridDistributedTxMapping<K, V> m) {
GridRichNode n = m.node();
assert !m.empty();
GridNearTxFinishRequest req =
new GridNearTxFinishRequest<K, V>(
futId,
tx.xidVersion(),
tx.commitVersion(),
tx.threadId(),
commit,
tx.isInvalidate(),
m.explicitLock(),
tx.topologyVersion(),
null,
null,
null,
commit && tx.pessimistic() ? m.writes() : null,
tx.syncCommit() && commit || tx.syncRollback() && !commit);
// If this is the primary node for the keys.
if (n.isLocal()) {
req.miniId(GridUuid.randomUuid());
if (CU.DHT_ENABLED) {
GridFuture<GridCacheTx> fut =
commit ? dht().commitTx(n.id(), req) : dht().rollbackTx(n.id(), req);
// Add new future.
add(fut);
} else
// Add done future for testing.
add(new GridFinishedFuture<GridCacheTx>(ctx));
} else {
MiniFuture fut = new MiniFuture(m);
req.miniId(fut.futureId());
add(fut); // Append new future.
try {
cctx.io().send(n, req);
// If we don't wait for result, then mark future as done.
if (!isSync() && !m.explicitLock()) fut.onDone();
} catch (GridTopologyException e) {
// Remove previous mapping.
mappings.remove(m.node().id());
fut.onResult(e);
} catch (GridException e) {
// Fail the whole thing.
fut.onResult(e);
}
}
}
/**
* @param p Partition number.
* @param topVer Topology version.
* @param create Create flag.
* @param updateSeq Update sequence.
* @return Local partition.
*/
private GridDhtLocalPartition localPartition(
int p, AffinityTopologyVersion topVer, boolean create, boolean updateSeq) {
while (true) {
boolean belongs = cctx.affinity().localNode(p, topVer);
GridDhtLocalPartition loc = locParts.get(p);
if (loc != null && loc.state() == EVICTED) {
locParts.remove(p, loc);
if (!create) return null;
if (!belongs)
throw new GridDhtInvalidPartitionException(
p,
"Adding entry to evicted partition [part="
+ p
+ ", topVer="
+ topVer
+ ", this.topVer="
+ this.topVer
+ ']');
continue;
}
if (loc == null && create) {
if (!belongs)
throw new GridDhtInvalidPartitionException(
p,
"Creating partition which does not belong [part="
+ p
+ ", topVer="
+ topVer
+ ", this.topVer="
+ this.topVer
+ ']');
lock.writeLock().lock();
try {
GridDhtLocalPartition old =
locParts.putIfAbsent(p, loc = new GridDhtLocalPartition(cctx, p));
if (old != null) loc = old;
else {
if (updateSeq) this.updateSeq.incrementAndGet();
if (log.isDebugEnabled()) log.debug("Created local partition: " + loc);
}
} finally {
lock.writeLock().unlock();
}
}
return loc;
}
}
/** Clears swap entries for evicted partition. */
private void clearSwap() {
assert state() == EVICTED;
assert !GridQueryProcessor.isEnabled(cctx.config())
: "Indexing needs to have unswapped values.";
try {
GridCloseableIterator<Map.Entry<byte[], GridCacheSwapEntry>> it = cctx.swap().iterator(id);
boolean isLocStore = cctx.store().isLocal();
if (it != null) {
// We can safely remove these values because no entries will be created for evicted
// partition.
while (it.hasNext()) {
Map.Entry<byte[], GridCacheSwapEntry> entry = it.next();
byte[] keyBytes = entry.getKey();
KeyCacheObject key = cctx.toCacheKeyObject(keyBytes);
cctx.swap().remove(key);
if (isLocStore) cctx.store().remove(null, key.value(cctx.cacheObjectContext(), false));
}
}
} catch (IgniteCheckedException e) {
U.error(log, "Failed to clear swap for evicted partition: " + this, e);
}
}
/**
* @param entry Transaction entry.
* @param nodes Nodes.
*/
private void map(GridCacheTxEntry<K, V> entry, Collection<GridRichNode> nodes) {
GridRichNode primary = CU.primary0(cctx.affinity(entry.key(), nodes));
GridDistributedTxMapping<K, V> t = mappings.get(primary.id());
if (t == null) mappings.put(primary.id(), t = new GridDistributedTxMapping<K, V>(primary));
t.add(entry);
}
/** {@inheritDoc} */
@Override
public boolean onDone(GridCacheTx tx, Throwable err) {
if (initialized() && super.onDone(tx, err)) {
// Don't forget to clean up.
cctx.mvcc().removeFuture(this);
return true;
}
return false;
}
/** {@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();
}
}
/**
* @param cacheCtx Cache context.
* @return {@code True} if local node can calculate affinity on it's own for this partition map
* exchange.
*/
private boolean canCalculateAffinity(GridCacheContext cacheCtx) {
AffinityFunction affFunc = cacheCtx.config().getAffinity();
// Do not request affinity from remote nodes if affinity function is not centralized.
if (!U.hasAnnotation(affFunc, AffinityCentralizedFunction.class)) return true;
// If local node did not initiate exchange or local node is the only cache node in grid.
Collection<ClusterNode> affNodes = CU.affinityNodes(cacheCtx, exchId.topologyVersion());
return !exchId.nodeId().equals(cctx.localNodeId())
|| (affNodes.size() == 1 && affNodes.contains(cctx.localNode()));
}
/** {@inheritDoc} */
@Override
public GridFuture<Boolean> awaitAsync(final long timeout, final TimeUnit unit) {
return ctx.closures()
.callLocalSafe(
new Callable<Boolean>() {
@Override
public Boolean call() throws Exception {
return await(timeout, unit);
}
},
true);
}
/**
* @param key Removed key.
* @param ver Removed version.
* @throws IgniteCheckedException If failed.
*/
public void onDeferredDelete(KeyCacheObject key, GridCacheVersion ver)
throws IgniteCheckedException {
try {
T2<KeyCacheObject, GridCacheVersion> evicted = rmvQueue.add(new T2<>(key, ver));
if (evicted != null) cctx.dht().removeVersionedEntry(evicted.get1(), evicted.get2());
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new IgniteInterruptedCheckedException(e);
}
}
/**
* @param updateSeq Update sequence.
* @return Future for evict attempt.
*/
IgniteInternalFuture<Boolean> tryEvictAsync(boolean updateSeq) {
if (map.isEmpty()
&& !GridQueryProcessor.isEnabled(cctx.config())
&& state.compareAndSet(RENTING, EVICTED, 0, 0)) {
if (log.isDebugEnabled()) log.debug("Evicted partition: " + this);
clearSwap();
if (cctx.isDrEnabled()) cctx.dr().partitionEvicted(id);
cctx.dataStructures().onPartitionEvicted(id);
rent.onDone();
((GridDhtPreloader) cctx.preloader()).onPartitionEvicted(this, updateSeq);
clearDeferredDeletes();
return new GridFinishedFuture<>(true);
}
return cctx.closures()
.callLocalSafe(
new GPC<Boolean>() {
@Override
public Boolean call() {
return tryEvict(true);
}
}, /*system pool*/
true);
}
/**
* Default constructor.
*
* @param name Sequence name.
* @param key Sequence key.
* @param seqView Sequence projection.
* @param ctx CacheContext.
* @param locVal Local counter.
* @param upBound Upper bound.
*/
public GridCacheAtomicSequenceImpl(
String name,
GridCacheInternalStorableKey key,
GridCacheProjection<GridCacheInternalStorableKey, GridCacheAtomicSequenceValue> seqView,
GridCacheContext ctx,
long locVal,
long upBound) {
assert key != null;
assert seqView != null;
assert ctx != null;
assert locVal <= upBound;
batchSize = ctx.config().getAtomicSequenceReserveSize();
this.ctx = ctx;
this.key = key;
this.seqView = seqView;
this.upBound = upBound;
this.locVal = locVal;
this.name = name;
log = ctx.gridConfig().getGridLogger().getLogger(getClass());
}
/** {@inheritDoc} */
@Override
public void onRemoved(GridDhtCacheEntry e) {
/*
* Make sure not to acquire any locks here as this method
* may be called from sensitive synchronization blocks.
* ===================================================
*/
GridDhtLocalPartition loc =
localPartition(cctx.affinity().partition(e.key()), topologyVersion(), false);
if (loc != null) loc.onRemoved(e);
}
/** {@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();
}
}
/** @param nodeId Node to remove. */
private void removeNode(UUID nodeId) {
assert nodeId != null;
assert lock.writeLock().isHeldByCurrentThread();
ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx.shared(), topVer);
assert oldest != null;
ClusterNode loc = cctx.localNode();
if (node2part != null) {
if (oldest.equals(loc) && !node2part.nodeId().equals(loc.id())) {
updateSeq.setIfGreater(node2part.updateSequence());
node2part =
new GridDhtPartitionFullMap(
loc.id(), loc.order(), updateSeq.incrementAndGet(), node2part, false);
} else node2part = new GridDhtPartitionFullMap(node2part, node2part.updateSequence());
part2node = new HashMap<>(part2node);
GridDhtPartitionMap parts = node2part.remove(nodeId);
if (parts != null) {
for (Integer p : parts.keySet()) {
Set<UUID> nodeIds = part2node.get(p);
if (nodeIds != null) {
nodeIds.remove(nodeId);
if (nodeIds.isEmpty()) part2node.remove(p);
}
}
}
consistencyCheck();
}
}
/**
* @param node Node.
* @param id ID.
* @throws IgniteCheckedException If failed.
*/
private void sendLocalPartitions(ClusterNode node, @Nullable GridDhtPartitionExchangeId id)
throws IgniteCheckedException {
GridDhtPartitionsSingleMessage m =
new GridDhtPartitionsSingleMessage(
id, cctx.kernalContext().clientNode(), cctx.versions().last());
for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
if (!cacheCtx.isLocal())
m.addLocalPartitionMap(cacheCtx.cacheId(), cacheCtx.topology().localPartitionMap());
}
if (log.isDebugEnabled())
log.debug(
"Sending local partitions [nodeId="
+ node.id()
+ ", exchId="
+ exchId
+ ", msg="
+ m
+ ']');
cctx.io().send(node, m, SYSTEM_POOL);
}
