/**
* @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);
}
/**
* Peeks only near cache without looking into DHT cache.
*
* @param key Key.
* @return Peeked value.
*/
@Nullable
public V peekNearOnly(K key) {
try {
return peek0(true, key, SMART, CU.<K, V>empty());
} catch (GridCacheFilterFailedException ignored) {
if (log.isDebugEnabled()) log.debug("Filter validation failed for key: " + key);
return null;
}
}
/**
* @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);
}
/** {@inheritDoc} */
@Override
public void remove() {
if (currEntry == null) throw new IllegalStateException();
assert currIter != null;
currIter.remove();
try {
GridNearCache.this.remove(currEntry.getKey(), CU.<K, V>empty());
} catch (GridException e) {
throw new GridRuntimeException(e);
}
}
/**
* @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);
}
/**
* @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);
}
/**
* Removes locks regardless of whether they are owned or not for given version and keys.
*
* @param ver Lock version.
* @param keys Keys.
*/
@SuppressWarnings({"unchecked"})
public void removeLocks(GridCacheVersion ver, Collection<? extends K> keys) {
if (keys.isEmpty()) return;
try {
Collection<GridRichNode> affNodes = null;
int keyCnt = -1;
Map<GridNode, GridNearUnlockRequest<K, V>> map = null;
for (K key : keys) {
// Send request to remove from remote nodes.
GridNearUnlockRequest<K, V> req = null;
while (true) {
GridDistributedCacheEntry<K, V> entry = peekExx(key);
try {
if (entry != null) {
GridCacheMvccCandidate<K> cand = entry.candidate(ver);
if (cand != null) {
if (affNodes == null) {
affNodes = CU.allNodes(ctx, cand.topologyVersion());
keyCnt = (int) Math.ceil((double) keys.size() / affNodes.size());
map = new HashMap<GridNode, GridNearUnlockRequest<K, V>>(affNodes.size());
}
GridRichNode primary = CU.primary0(ctx.affinity(key, affNodes));
if (!primary.isLocal()) {
req = map.get(primary);
if (req == null) {
map.put(primary, req = new GridNearUnlockRequest<K, V>(keyCnt));
req.version(ver);
}
}
// Remove candidate from local node first.
if (entry.removeLock(cand.version())) {
if (primary.isLocal()) {
dht.removeLocks(primary.id(), ver, F.asList(key), true);
assert req == null;
continue;
}
req.addKey(entry.key(), entry.getOrMarshalKeyBytes(), ctx);
}
}
}
break;
} catch (GridCacheEntryRemovedException ignored) {
if (log.isDebugEnabled())
log.debug(
"Attempted to remove lock from removed entry (will retry) [rmvVer="
+ ver
+ ", entry="
+ entry
+ ']');
}
}
}
if (map == null || map.isEmpty()) return;
Collection<GridCacheVersion> committed = ctx.tm().committedVersions(ver);
Collection<GridCacheVersion> rolledback = ctx.tm().rolledbackVersions(ver);
for (Map.Entry<GridNode, GridNearUnlockRequest<K, V>> mapping : map.entrySet()) {
GridNode n = mapping.getKey();
GridDistributedUnlockRequest<K, V> req = mapping.getValue();
if (!req.keyBytes().isEmpty()) {
req.completedVersions(committed, rolledback);
// We don't wait for reply to this message.
ctx.io().send(n, req);
}
}
} catch (GridException ex) {
U.error(log, "Failed to unlock the lock for keys: " + keys, ex);
}
}
/** {@inheritDoc} */
@Override
public void unlockAll(
Collection<? extends K> keys, GridPredicate<? super GridCacheEntry<K, V>>[] filter) {
if (keys.isEmpty()) return;
try {
GridCacheVersion ver = null;
Collection<GridRichNode> affNodes = null;
int keyCnt = -1;
Map<GridRichNode, GridNearUnlockRequest<K, V>> map = null;
Collection<K> locKeys = new LinkedList<K>();
GridCacheVersion obsoleteVer = ctx.versions().next();
for (K key : keys) {
while (true) {
GridDistributedCacheEntry<K, V> entry = peekExx(key);
if (entry == null || !ctx.isAll(entry.wrap(false), filter)) break; // While.
try {
GridCacheMvccCandidate<K> cand =
entry.candidate(ctx.nodeId(), Thread.currentThread().getId());
if (cand != null) {
ver = cand.version();
if (affNodes == null) {
affNodes = CU.allNodes(ctx, cand.topologyVersion());
keyCnt = (int) Math.ceil((double) keys.size() / affNodes.size());
map = new HashMap<GridRichNode, GridNearUnlockRequest<K, V>>(affNodes.size());
}
// Send request to remove from remote nodes.
GridRichNode primary = CU.primary0(ctx.affinity(key, affNodes));
GridNearUnlockRequest<K, V> req = map.get(primary);
if (req == null) {
map.put(primary, req = new GridNearUnlockRequest<K, V>(keyCnt));
req.version(ver);
}
// Remove candidate from local node first.
GridCacheMvccCandidate<K> rmv = entry.removeLock();
if (rmv != null) {
if (!rmv.reentry()) {
if (ver != null && !ver.equals(rmv.version()))
throw new GridException(
"Failed to unlock (if keys were locked separately, "
+ "then they need to be unlocked separately): "
+ keys);
if (!primary.isLocal()) {
assert req != null;
req.addKey(entry.key(), entry.getOrMarshalKeyBytes(), ctx);
} else locKeys.add(key);
if (log.isDebugEnabled()) log.debug("Removed lock (will distribute): " + rmv);
} else if (log.isDebugEnabled())
log.debug(
"Current thread still owns lock (or there are no other nodes)"
+ " [lock="
+ rmv
+ ", curThreadId="
+ Thread.currentThread().getId()
+ ']');
}
// Try to evict near entry if it's dht-mapped locally.
evictNearEntry(entry, obsoleteVer);
}
break;
} catch (GridCacheEntryRemovedException ignore) {
if (log.isDebugEnabled())
log.debug("Attempted to unlock removed entry (will retry): " + entry);
}
}
}
if (ver == null) return;
for (Map.Entry<GridRichNode, GridNearUnlockRequest<K, V>> mapping : map.entrySet()) {
GridRichNode n = mapping.getKey();
GridDistributedUnlockRequest<K, V> req = mapping.getValue();
if (n.isLocal()) dht.removeLocks(ctx.nodeId(), req.version(), locKeys, true);
else if (!req.keyBytes().isEmpty())
// We don't wait for reply to this message.
ctx.io().send(n, req);
}
} catch (GridException ex) {
U.error(log, "Failed to unlock the lock for keys: " + keys, ex);
}
}
/**
* @param e Transaction entry.
* @return {@code True} if entry is locally mapped as a primary or back up node.
*/
protected boolean isNearLocallyMapped(GridCacheEntryEx<K, V> e) {
return F.contains(ctx.affinity(e.key(), CU.allNodes(ctx)), ctx.localNode());
}
/** {@inheritDoc} */
@Override
public Map<GridRichNode, Collection<K>> mapKeysToNodes(Collection<? extends K> keys) {
return CU.mapKeysToNodes(ctx, keys);
}
/** @return Near entries. */
public Set<GridCacheEntry<K, V>> nearEntries() {
return super.entrySet(CU.<K, V>empty());
}
/**
* 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);
}
}
/** @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);
}
}
/** 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();
}