/** {@inheritDoc} */
@Override
public GridFuture<Boolean> loadMissing(
boolean async, final Collection<? extends K> keys, final GridInClosure2<K, V> closure) {
GridFuture<Map<K, V>> f = cctx.near().txLoadAsync(this, keys, CU.<K, V>empty());
return new GridEmbeddedFuture<Boolean, Map<K, V>>(
cctx.kernalContext(),
f,
new C2<Map<K, V>, Exception, Boolean>() {
@Override
public Boolean apply(Map<K, V> map, Exception e) {
if (e != null) {
setRollbackOnly();
throw new GridClosureException(e);
}
// Must loop through keys, not map entries,
// as map entries may not have all the keys.
for (K key : keys) closure.apply(key, map.get(key));
return true;
}
});
}
/**
* @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 Integer call() throws Exception {
GridCacheTx tx = CU.txStartInternal(ctx, latchView, PESSIMISTIC, REPEATABLE_READ);
try {
GridCacheCountDownLatchValue latchVal = latchView.get(key);
if (latchVal == null) {
if (log.isDebugEnabled())
log.debug("Failed to find count down latch with given name: " + name);
assert cnt == 0;
return cnt;
}
int retVal;
if (val > 0) {
retVal = latchVal.get() - val;
if (retVal < 0) retVal = 0;
} else retVal = 0;
latchVal.set(retVal);
latchView.put(key, latchVal);
tx.commit();
return retVal;
} finally {
tx.end();
}
}
/**
* @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);
}
/**
* @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;
}
/** {@inheritDoc} */
@Override
public void writeExternal(ObjectOutput out) throws IOException {
GridUtils.writeUuid(out, nodeId);
CU.writeVersion(out, ver);
out.writeLong(timeout);
out.writeLong(threadId);
out.writeLong(id);
out.writeShort(flags());
}
/**
* @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()));
}
/**
* @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 reads Read entries.
* @param writes Write entries.
*/
@SuppressWarnings({"unchecked"})
private void prepare(
Iterable<GridCacheTxEntry<K, V>> reads, Iterable<GridCacheTxEntry<K, V>> writes) {
Collection<GridRichNode> nodes = CU.allNodes(cctx);
// Assign keys to primary nodes.
for (GridCacheTxEntry<K, V> read : reads) map(read, nodes);
for (GridCacheTxEntry<K, V> write : writes) map(write, nodes);
// Create mini futures.
for (GridDistributedTxMapping<K, V> m : mappings.values()) finish(m);
}
/**
* @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);
}
/**
* @param cctx Context.
* @param tx Transaction.
* @param failedNodeId ID of failed node started transaction.
*/
@SuppressWarnings("ConstantConditions")
public GridCachePessimisticCheckCommittedTxFuture(
GridCacheContext<K, V> cctx, GridCacheTxEx<K, V> tx, UUID failedNodeId) {
super(cctx.kernalContext(), new SingleReducer<K, V>());
this.cctx = cctx;
this.tx = tx;
this.failedNodeId = failedNodeId;
log = U.logger(ctx, logRef, GridCacheOptimisticCheckPreparedTxFuture.class);
nodes = new GridLeanMap<>();
for (GridNode node : CU.allNodes(cctx, tx.topologyVersion())) nodes.put(node.id(), node);
}
/**
* 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());
}
}
}
/**
* @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;
}
/** @throws GridException If operation failed. */
private void initializeLatch() throws GridException {
if (initGuard.compareAndSet(false, true)) {
try {
internalLatch =
CU.outTx(
new Callable<CountDownLatch>() {
@Override
public CountDownLatch call() throws Exception {
GridCacheTx tx =
CU.txStartInternal(ctx, latchView, PESSIMISTIC, REPEATABLE_READ);
try {
GridCacheCountDownLatchValue val = latchView.get(key);
if (val == null) {
if (log.isDebugEnabled())
log.debug("Failed to find count down latch with given name: " + name);
assert cnt == 0;
return new CountDownLatch(cnt);
}
tx.commit();
return new CountDownLatch(val.get());
} finally {
tx.end();
}
}
},
ctx);
if (log.isDebugEnabled()) log.debug("Initialized internal latch: " + internalLatch);
} finally {
initLatch.countDown();
}
} else {
try {
initLatch.await();
} catch (InterruptedException ignored) {
throw new GridException("Thread has been interrupted.");
}
if (internalLatch == null)
throw new GridException("Internal latch has not been properly initialized.");
}
}
/** {@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;
}
/** {@inheritDoc} */
@Override
public void readExternal(ObjectInput in) throws IOException, ClassNotFoundException {
nodeId = GridUtils.readUuid(in);
ver = CU.readVersion(in);
timeout = in.readLong();
threadId = in.readLong();
id = in.readLong();
short flags = in.readShort();
mask(OWNER, OWNER.get(flags));
mask(USED, USED.get(flags));
mask(TX, TX.get(flags));
ts = U.currentTimeMillis();
}
/**
* @param cctx Registry.
* @param keys Keys to lock.
* @param tx Transaction.
* @param read Read flag.
* @param retval Flag to return value or not.
* @param timeout Lock acquisition timeout.
* @param filter Filter.
*/
public GridNearLockFuture(
GridCacheContext<K, V> cctx,
Collection<? extends K> keys,
@Nullable GridNearTxLocal<K, V> tx,
boolean read,
boolean retval,
long timeout,
GridPredicate<GridCacheEntry<K, V>>[] filter) {
super(cctx.kernalContext(), CU.boolReducer());
assert cctx != null;
assert keys != null;
this.cctx = cctx;
this.keys = keys;
this.tx = tx;
this.read = read;
this.retval = retval;
this.timeout = timeout;
this.filter = filter;
threadId = tx == null ? Thread.currentThread().getId() : tx.threadId();
lockVer = tx != null ? tx.xidVersion() : cctx.versions().next();
futId = GridUuid.randomUuid();
entries = new ArrayList<>(keys.size());
log = U.logger(ctx, logRef, GridNearLockFuture.class);
if (timeout > 0) {
timeoutObj = new LockTimeoutObject();
cctx.time().addTimeoutObject(timeoutObj);
}
valMap = new ConcurrentHashMap8<>(keys.size(), 1f);
}
/** Initializes future. */
public void prepare() {
if (log.isDebugEnabled())
log.debug("Checking if transaction was committed on remote nodes: " + tx);
// Check local node first (local node can be a backup node for some part of this transaction).
long originatingThreadId = tx.threadId();
if (tx instanceof GridCacheTxRemoteEx)
originatingThreadId = ((GridCacheTxRemoteEx) tx).remoteThreadId();
GridCacheCommittedTxInfo<K, V> txInfo =
cctx.tm().txCommitted(tx.nearXidVersion(), tx.eventNodeId(), originatingThreadId);
if (txInfo != null) {
onDone(txInfo);
markInitialized();
return;
}
Collection<GridNode> checkNodes = CU.remoteNodes(cctx, tx.topologyVersion());
if (tx instanceof GridDhtTxRemote) {
// If we got primary node failure and near node has not failed.
if (tx.nodeId().equals(failedNodeId) && !tx.eventNodeId().equals(failedNodeId)) {
nearCheck = true;
GridNode nearNode = cctx.discovery().node(tx.eventNodeId());
if (nearNode == null) {
// Near node failed, separate check prepared future will take care of it.
onDone(
new GridTopologyException(
"Failed to check near transaction state (near node left grid): "
+ tx.eventNodeId()));
return;
}
checkNodes = Collections.singletonList(nearNode);
}
}
for (GridNode rmtNode : checkNodes) {
// Skip left nodes and local node.
if (rmtNode.id().equals(failedNodeId)) continue;
/*
* Send message to all cache nodes in the topology.
*/
MiniFuture fut = new MiniFuture(rmtNode.id());
GridCachePessimisticCheckCommittedTxRequest<K, V> req =
new GridCachePessimisticCheckCommittedTxRequest<>(
tx, originatingThreadId, futureId(), fut.futureId());
add(fut);
try {
cctx.io().send(rmtNode.id(), req);
} catch (GridTopologyException ignored) {
fut.onNodeLeft();
} catch (GridException e) {
fut.onError(e);
break;
}
}
markInitialized();
}
/** {@inheritDoc} */
@Override
public int countDown() throws GridException {
return CU.outTx(new CountDownCallable(1), ctx);
}
/**
* 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 int countDown(int val) throws GridException {
A.ensure(val > 0, "val should be positive");
return CU.outTx(new CountDownCallable(val), ctx);
}
/** {@inheritDoc} */
@Override
public void countDownAll() throws GridException {
CU.outTx(new CountDownCallable(0), ctx);
}
/** @param res Result callback. */
void onResult(GridNearLockResponse<K, V> res) {
if (rcvRes.compareAndSet(false, true)) {
if (res.error() != null) {
if (log.isDebugEnabled())
log.debug(
"Finishing mini future with an error due to error in response [miniFut="
+ this
+ ", res="
+ res
+ ']');
// Fail.
if (res.error() instanceof GridCacheLockTimeoutException) onDone(false);
else onDone(res.error());
return;
}
int i = 0;
long topVer = topSnapshot.get().topologyVersion();
for (K k : keys) {
while (true) {
GridNearCacheEntry<K, V> entry = cctx.near().entryExx(k, topVer);
try {
if (res.dhtVersion(i) == null) {
onDone(
new GridException(
"Failed to receive DHT version from remote node "
+ "(will fail the lock): "
+ res));
return;
}
GridTuple3<GridCacheVersion, V, byte[]> oldValTup = valMap.get(entry.key());
V oldVal = entry.rawGet();
boolean hasOldVal = false;
V newVal = res.value(i);
byte[] newBytes = res.valueBytes(i);
boolean readRecordable = false;
if (retval) {
readRecordable = cctx.events().isRecordable(EVT_CACHE_OBJECT_READ);
if (readRecordable) hasOldVal = entry.hasValue();
}
GridCacheVersion dhtVer = res.dhtVersion(i);
GridCacheVersion mappedVer = res.mappedVersion(i);
if (newVal == null) {
if (oldValTup != null) {
if (oldValTup.get1().equals(dhtVer)) {
newVal = oldValTup.get2();
newBytes = oldValTup.get3();
}
oldVal = oldValTup.get2();
}
}
// Lock is held at this point, so we can set the
// returned value if any.
entry.resetFromPrimary(newVal, newBytes, lockVer, dhtVer, node.id());
if (inTx() && implicitTx() && tx.onePhaseCommit()) {
boolean pass = res.filterResult(i);
tx.entry(k).filters(pass ? CU.<K, V>empty() : CU.<K, V>alwaysFalse());
}
entry.readyNearLock(
lockVer,
mappedVer,
res.committedVersions(),
res.rolledbackVersions(),
res.pending());
if (retval) {
if (readRecordable)
cctx.events()
.addEvent(
entry.partition(),
entry.key(),
tx,
null,
EVT_CACHE_OBJECT_READ,
newVal,
newVal != null || newBytes != null,
oldVal,
hasOldVal,
CU.subjectId(tx, cctx));
cctx.cache().metrics0().onRead(false);
}
if (log.isDebugEnabled())
log.debug("Processed response for entry [res=" + res + ", entry=" + entry + ']');
break; // Inner while loop.
} catch (GridCacheEntryRemovedException ignored) {
if (log.isDebugEnabled())
log.debug("Failed to add candidates because entry was removed (will renew).");
// Replace old entry with new one.
entries.set(i, (GridDistributedCacheEntry<K, V>) cctx.cache().entryEx(entry.key()));
} catch (GridException e) {
onDone(e);
return;
}
}
i++;
}
try {
proceedMapping(mappings);
} catch (GridException e) {
onDone(e);
}
onDone(true);
}
}
/**
* Maps keys to nodes. Note that we can not simply group keys by nodes and send lock request as
* such approach does not preserve order of lock acquisition. Instead, keys are split in
* continuous groups belonging to one primary node and locks for these groups are acquired
* sequentially.
*
* @param keys Keys.
*/
private void map(Iterable<? extends K> keys) {
try {
GridDiscoveryTopologySnapshot snapshot = topSnapshot.get();
assert snapshot != null;
long topVer = snapshot.topologyVersion();
assert topVer > 0;
if (CU.affinityNodes(cctx, topVer).isEmpty()) {
onDone(
new GridTopologyException(
"Failed to map keys for near-only cache (all "
+ "partition nodes left the grid)."));
return;
}
ConcurrentLinkedDeque8<GridNearLockMapping<K, V>> mappings = new ConcurrentLinkedDeque8<>();
// Assign keys to primary nodes.
GridNearLockMapping<K, V> map = null;
for (K key : keys) {
GridNearLockMapping<K, V> updated = map(key, map, topVer);
// If new mapping was created, add to collection.
if (updated != map) mappings.add(updated);
map = updated;
}
if (isDone()) {
if (log.isDebugEnabled()) log.debug("Abandoning (re)map because future is done: " + this);
return;
}
if (log.isDebugEnabled())
log.debug("Starting (re)map for mappings [mappings=" + mappings + ", fut=" + this + ']');
// Create mini futures.
for (Iterator<GridNearLockMapping<K, V>> iter = mappings.iterator(); iter.hasNext(); ) {
GridNearLockMapping<K, V> mapping = iter.next();
GridNode node = mapping.node();
Collection<K> mappedKeys = mapping.mappedKeys();
assert !mappedKeys.isEmpty();
GridNearLockRequest<K, V> req = null;
Collection<K> distributedKeys = new ArrayList<>(mappedKeys.size());
boolean explicit = false;
for (K key : mappedKeys) {
while (true) {
GridNearCacheEntry<K, V> entry = null;
try {
entry = cctx.near().entryExx(key, topVer);
if (!cctx.isAll(entry.wrap(false), filter)) {
if (log.isDebugEnabled())
log.debug("Entry being locked did not pass filter (will not lock): " + entry);
onComplete(false, false);
return;
}
// Removed exception may be thrown here.
GridCacheMvccCandidate<K> cand = addEntry(topVer, entry, node.id());
if (isDone()) {
if (log.isDebugEnabled())
log.debug(
"Abandoning (re)map because future is done after addEntry attempt "
+ "[fut="
+ this
+ ", entry="
+ entry
+ ']');
return;
}
if (cand != null) {
if (tx == null && !cand.reentry())
cctx.mvcc().addExplicitLock(threadId, cand, snapshot);
GridTuple3<GridCacheVersion, V, byte[]> val = entry.versionedValue();
if (val == null) {
GridDhtCacheEntry<K, V> dhtEntry = dht().peekExx(key);
try {
if (dhtEntry != null) val = dhtEntry.versionedValue(topVer);
} catch (GridCacheEntryRemovedException ignored) {
assert dhtEntry.obsolete()
: " Got removed exception for non-obsolete entry: " + dhtEntry;
if (log.isDebugEnabled())
log.debug(
"Got removed exception for DHT entry in map (will ignore): " + dhtEntry);
}
}
GridCacheVersion dhtVer = null;
if (val != null) {
dhtVer = val.get1();
valMap.put(key, val);
}
if (!cand.reentry()) {
if (req == null) {
req =
new GridNearLockRequest<>(
topVer,
cctx.nodeId(),
threadId,
futId,
lockVer,
inTx(),
implicitTx(),
implicitSingleTx(),
read,
isolation(),
isInvalidate(),
timeout,
syncCommit(),
syncRollback(),
mappedKeys.size(),
inTx() ? tx.size() : mappedKeys.size(),
inTx() ? tx.groupLockKey() : null,
inTx() && tx.partitionLock(),
inTx() ? tx.subjectId() : null);
mapping.request(req);
}
distributedKeys.add(key);
GridCacheTxEntry<K, V> writeEntry = tx != null ? tx.writeMap().get(key) : null;
if (tx != null) tx.addKeyMapping(key, mapping.node());
req.addKeyBytes(
key,
node.isLocal() ? null : entry.getOrMarshalKeyBytes(),
retval && dhtVer == null,
dhtVer, // Include DHT version to match remote DHT entry.
writeEntry,
inTx() ? tx.entry(key).drVersion() : null,
cctx);
// Clear transfer required flag since we are sending message.
if (writeEntry != null) writeEntry.transferRequired(false);
}
if (cand.reentry())
explicit = tx != null && !entry.hasLockCandidate(tx.xidVersion());
} else
// Ignore reentries within transactions.
explicit = tx != null && !entry.hasLockCandidate(tx.xidVersion());
if (explicit) tx.addKeyMapping(key, mapping.node());
break;
} catch (GridCacheEntryRemovedException ignored) {
assert entry.obsolete() : "Got removed exception on non-obsolete entry: " + entry;
if (log.isDebugEnabled())
log.debug("Got removed entry in lockAsync(..) method (will retry): " + entry);
}
}
// Mark mapping explicit lock flag.
if (explicit) {
boolean marked = tx != null && tx.markExplicit(node.id());
assert tx == null || marked;
}
}
if (!distributedKeys.isEmpty()) mapping.distributedKeys(distributedKeys);
else {
assert mapping.request() == null;
iter.remove();
}
}
cctx.mvcc().recheckPendingLocks();
proceedMapping(mappings);
} catch (GridException ex) {
onError(ex);
}
}
/**
* Synchronous sequence update operation. Will add given amount to the sequence value.
*
* @param l Increment amount.
* @param updateCall Cache call that will update sequence reservation count in accordance with l.
* @param updated If {@code true}, will return sequence value after update, otherwise will return
* sequence value prior to update.
* @return Sequence value.
* @throws GridException If update failed.
*/
private long internalUpdate(long l, @Nullable Callable<Long> updateCall, boolean updated)
throws GridException {
checkRemoved();
assert l > 0;
lock.lock();
try {
// If reserved range isn't exhausted.
if (locVal + l <= upBound) {
long curVal = locVal;
locVal += l;
return updated ? locVal : curVal;
}
} finally {
lock.unlock();
}
if (updateCall == null) updateCall = internalUpdate(l, updated);
while (true) {
if (updateGuard.compareAndSet(false, true)) {
try {
// This call must be outside lock.
return CU.outTx(updateCall, ctx);
} finally {
lock.lock();
try {
updateGuard.set(false);
cond.signalAll();
} finally {
lock.unlock();
}
}
} else {
lock.lock();
try {
while (locVal >= upBound && updateGuard.get()) {
try {
cond.await(500, MILLISECONDS);
} catch (InterruptedException e) {
throw new GridInterruptedException(e);
}
}
checkRemoved();
// If reserved range isn't exhausted.
if (locVal + l <= upBound) {
long curVal = locVal;
locVal += l;
return updated ? locVal : curVal;
}
} finally {
lock.unlock();
}
}
}
}
