/**
* @param key Key to add to read set.
* @param val Value.
* @param drVer Data center replication version.
* @param skipStore Skip store flag.
* @throws IgniteCheckedException If failed.
* @return {@code True} if entry has been enlisted.
*/
public boolean addEntry(
GridCacheContext cacheCtx,
IgniteTxKey key,
GridCacheOperation op,
CacheObject val,
@Nullable GridCacheVersion drVer,
boolean skipStore)
throws IgniteCheckedException {
checkInternal(key);
GridNearCacheEntry cached = cacheCtx.near().peekExx(key.key());
try {
if (cached == null) {
evicted.add(key);
return false;
} else {
cached.unswap();
CacheObject peek = cached.peek(true, false, false, null);
if (peek == null && cached.evictInternal(false, xidVer, null)) {
cached.context().cache().removeIfObsolete(key.key());
evicted.add(key);
return false;
} else {
IgniteTxEntry txEntry =
new IgniteTxEntry(cacheCtx, this, op, val, -1L, -1L, cached, drVer, skipStore);
writeMap.put(key, txEntry);
return true;
}
}
} catch (GridCacheEntryRemovedException ignore) {
evicted.add(key);
if (log.isDebugEnabled())
log.debug(
"Got removed entry when adding reads to remote transaction (will ignore): " + cached);
return false;
}
}
/**
* @param entry Entry to enlist.
* @throws IgniteCheckedException If failed.
* @return {@code True} if entry was enlisted.
*/
private boolean addEntry(IgniteTxEntry entry) throws IgniteCheckedException {
checkInternal(entry.txKey());
GridCacheContext cacheCtx = entry.context();
if (!cacheCtx.isNear()) cacheCtx = cacheCtx.dht().near().context();
GridNearCacheEntry cached = cacheCtx.near().peekExx(entry.key());
if (cached == null) {
evicted.add(entry.txKey());
return false;
} else {
cached.unswap();
try {
CacheObject val = cached.peek(true, false, false, null);
if (val == null && cached.evictInternal(false, xidVer, null)) {
evicted.add(entry.txKey());
return false;
} else {
// Initialize cache entry.
entry.cached(cached);
writeMap.put(entry.txKey(), entry);
addExplicit(entry);
return true;
}
} catch (GridCacheEntryRemovedException ignore) {
evicted.add(entry.txKey());
if (log.isDebugEnabled())
log.debug("Got removed entry when adding to remote transaction (will ignore): " + cached);
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;
}
/** @return DHT cache. */
private GridDhtCache<K, V> dht() {
return cctx.near().dht();
}
/**
* 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);
}
}