/**
* @param mapping Mappings.
* @param key Key to map.
* @param topVer Topology version.
* @return Near lock mapping.
* @throws GridException If mapping for key failed.
*/
private GridNearLockMapping<K, V> map(
K key, @Nullable GridNearLockMapping<K, V> mapping, long topVer) throws GridException {
assert mapping == null || mapping.node() != null;
GridNode primary = cctx.affinity().primary(key, topVer);
if (cctx.discovery().node(primary.id()) == null)
// If primary node left the grid before lock acquisition, fail the whole future.
throw newTopologyException(null, primary.id());
if (inTx() && tx.groupLock() && !primary.isLocal())
throw new GridException(
"Failed to start group lock transaction (local node is not primary for "
+ " key) [key="
+ key
+ ", primaryNodeId="
+ primary.id()
+ ']');
if (mapping == null || !primary.id().equals(mapping.node().id()))
mapping = new GridNearLockMapping<>(primary, key);
else mapping.addKey(key);
return mapping;
}
/**
* @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);
}
/**
* Acquires topology future and checks it completeness under the read lock. If it is not complete,
* will asynchronously wait for it's completeness and then try again.
*/
void mapOnTopology() {
// We must acquire topology snapshot from the topology version future.
try {
cctx.topology().readLock();
try {
GridDhtTopologyFuture fut = cctx.topologyVersionFuture();
if (fut.isDone()) {
GridDiscoveryTopologySnapshot snapshot = fut.topologySnapshot();
if (tx != null) {
tx.topologyVersion(snapshot.topologyVersion());
tx.topologySnapshot(snapshot);
}
topSnapshot.compareAndSet(null, snapshot);
map(keys);
markInitialized();
} else {
fut.listenAsync(
new CI1<GridFuture<Long>>() {
@Override
public void apply(GridFuture<Long> t) {
mapOnTopology();
}
});
}
} finally {
cctx.topology().readUnlock();
}
} catch (GridException e) {
onDone(e);
}
}
/**
* Undoes all locks.
*
* @param dist If {@code true}, then remove locks from remote nodes as well.
*/
private void undoLocks(boolean dist) {
// Transactions will undo during rollback.
if (dist && tx == null) cctx.nearTx().removeLocks(lockVer, keys);
else {
if (tx != null) {
if (tx.setRollbackOnly()) {
if (log.isDebugEnabled())
log.debug(
"Marked transaction as rollback only because locks could not be acquired: " + tx);
} else if (log.isDebugEnabled())
log.debug(
"Transaction was not marked rollback-only while locks were not acquired: " + tx);
}
for (GridCacheEntryEx<K, V> e : entriesCopy()) {
try {
e.removeLock(lockVer);
} catch (GridCacheEntryRemovedException ignored) {
while (true) {
try {
e = cctx.cache().peekEx(e.key());
if (e != null) e.removeLock(lockVer);
break;
} catch (GridCacheEntryRemovedException ignore) {
if (log.isDebugEnabled())
log.debug(
"Attempted to remove lock on removed entry (will retry) [ver="
+ lockVer
+ ", entry="
+ e
+ ']');
}
}
}
}
}
cctx.mvcc().recheckPendingLocks();
}
/**
* Adds entry to future.
*
* @param topVer Topology version.
* @param entry Entry to add.
* @param dhtNodeId DHT node ID.
* @return Lock candidate.
* @throws GridCacheEntryRemovedException If entry was removed.
*/
@Nullable
private GridCacheMvccCandidate<K> addEntry(
long topVer, GridNearCacheEntry<K, V> entry, UUID dhtNodeId)
throws GridCacheEntryRemovedException {
// Check if lock acquisition is timed out.
if (timedOut) return null;
// Add local lock first, as it may throw GridCacheEntryRemovedException.
GridCacheMvccCandidate<K> c =
entry.addNearLocal(
dhtNodeId, threadId, lockVer, timeout, !inTx(), inTx(), implicitSingleTx());
if (inTx()) {
GridCacheTxEntry<K, V> txEntry = tx.entry(entry.key());
txEntry.cached(entry, txEntry.keyBytes());
}
if (c != null) c.topologyVersion(topVer);
synchronized (mux) {
entries.add(entry);
}
if (c == null && timeout < 0) {
if (log.isDebugEnabled()) log.debug("Failed to acquire lock with negative timeout: " + entry);
onFailed(false);
return null;
}
// Double check if lock acquisition has already timed out.
if (timedOut) {
entry.removeLock(lockVer);
return null;
}
return c;
}
/**
* Basically, future mapping consists from two parts. First, we must determine the topology
* version this future will map on. Locking is performed within a user transaction, we must
* continue to map keys on the same topology version as it started. If topology version is
* undefined, we get current topology future and wait until it completes so the topology is ready
* to use.
*
* <p>During the second part we map keys to primary nodes using topology snapshot we obtained
* during the first part. Note that if primary node leaves grid, the future will fail and
* transaction will be rolled back.
*/
void map() {
// Obtain the topology version to use.
GridDiscoveryTopologySnapshot snapshot =
tx != null
? tx.topologySnapshot()
: cctx.mvcc().lastExplicitLockTopologySnapshot(Thread.currentThread().getId());
if (snapshot != null) {
// Continue mapping on the same topology version as it was before.
topSnapshot.compareAndSet(null, snapshot);
map(keys);
markInitialized();
return;
}
// Must get topology snapshot and map on that version.
mapOnTopology();
}
/**
* Gets next near lock mapping and either acquires dht locks locally or sends near lock request to
* remote primary node.
*
* @param mappings Queue of mappings.
* @throws GridException If mapping can not be completed.
*/
private void proceedMapping(final ConcurrentLinkedDeque8<GridNearLockMapping<K, V>> mappings)
throws GridException {
GridNearLockMapping<K, V> map = mappings.poll();
// If there are no more mappings to process, complete the future.
if (map == null) return;
final GridNearLockRequest<K, V> req = map.request();
final Collection<K> mappedKeys = map.distributedKeys();
final GridNode node = map.node();
if (filter != null && filter.length != 0) req.filter(filter, cctx);
if (node.isLocal()) {
req.miniId(GridUuid.randomUuid());
if (log.isDebugEnabled()) log.debug("Before locally locking near request: " + req);
GridFuture<GridNearLockResponse<K, V>> fut;
if (CU.DHT_ENABLED) fut = dht().lockAllAsync(cctx.localNode(), req, filter);
else {
// Create dummy values for testing.
GridNearLockResponse<K, V> res =
new GridNearLockResponse<>(lockVer, futId, null, false, 1, null);
res.addValueBytes(null, null, true, lockVer, lockVer, cctx);
fut = new GridFinishedFuture<>(ctx, res);
}
// Add new future.
add(
new GridEmbeddedFuture<>(
cctx.kernalContext(),
fut,
new C2<GridNearLockResponse<K, V>, Exception, Boolean>() {
@Override
public Boolean apply(GridNearLockResponse<K, V> res, Exception e) {
if (CU.isLockTimeoutOrCancelled(e)
|| (res != null && CU.isLockTimeoutOrCancelled(res.error()))) return false;
if (e != null) {
onError(e);
return false;
}
if (res == null) {
onError(new GridException("Lock response is null for future: " + this));
return false;
}
if (res.error() != null) {
onError(res.error());
return false;
}
if (log.isDebugEnabled())
log.debug(
"Acquired lock for local DHT mapping [locId="
+ cctx.nodeId()
+ ", mappedKeys="
+ mappedKeys
+ ", fut="
+ GridNearLockFuture.this
+ ']');
try {
int i = 0;
for (K k : mappedKeys) {
while (true) {
GridNearCacheEntry<K, V> entry =
cctx.near().entryExx(k, req.topologyVersion());
try {
GridTuple3<GridCacheVersion, V, byte[]> oldValTup =
valMap.get(entry.key());
boolean hasBytes = entry.hasValue();
V oldVal = entry.rawGet();
V newVal = res.value(i);
byte[] newBytes = res.valueBytes(i);
GridCacheVersion dhtVer = res.dhtVersion(i);
GridCacheVersion mappedVer = res.mappedVersion(i);
// On local node don't record twice if DHT cache already recorded.
boolean record =
retval && oldValTup != null && oldValTup.get1().equals(dhtVer);
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());
entry.readyNearLock(
lockVer,
mappedVer,
res.committedVersions(),
res.rolledbackVersions(),
res.pending());
if (inTx() && implicitTx() && tx.onePhaseCommit()) {
boolean pass = res.filterResult(i);
tx.entry(k).filters(pass ? CU.<K, V>empty() : CU.<K, V>alwaysFalse());
}
if (record) {
if (cctx.events().isRecordable(EVT_CACHE_OBJECT_READ))
cctx.events()
.addEvent(
entry.partition(),
entry.key(),
tx,
null,
EVT_CACHE_OBJECT_READ,
newVal,
newVal != null,
oldVal,
hasBytes,
CU.subjectId(tx, cctx));
cctx.cache().metrics0().onRead(oldVal != null);
}
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()));
}
}
i++; // Increment outside of while loop.
}
// Proceed and add new future (if any) before completing embedded future.
proceedMapping(mappings);
} catch (GridException ex) {
onError(ex);
return false;
}
return true;
}
}));
} else {
final MiniFuture fut = new MiniFuture(node, mappedKeys, mappings);
req.miniId(fut.futureId());
add(fut); // Append new future.
GridFuture<?> txSync = null;
if (inTx()) txSync = cctx.tm().awaitFinishAckAsync(node.id(), tx.threadId());
if (txSync == null || txSync.isDone()) {
try {
if (log.isDebugEnabled())
log.debug("Sending near lock request [node=" + node.id() + ", req=" + req + ']');
cctx.io().send(node, req);
} catch (GridTopologyException ex) {
assert fut != null;
fut.onResult(ex);
}
} else {
txSync.listenAsync(
new CI1<GridFuture<?>>() {
@Override
public void apply(GridFuture<?> t) {
try {
if (log.isDebugEnabled())
log.debug(
"Sending near lock request [node=" + node.id() + ", req=" + req + ']');
cctx.io().send(node, req);
} catch (GridTopologyException ex) {
assert fut != null;
fut.onResult(ex);
} catch (GridException e) {
onError(e);
}
}
});
}
}
}
/**
* 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);
}
}
/** @return {@code true} if related transaction is implicit. */
private boolean implicitTx() {
return tx != null && tx.implicit();
}
/** @return Transaction isolation or {@code null} if no transaction. */
@Nullable
private GridCacheTxIsolation isolation() {
return tx == null ? null : tx.isolation();
}
/** @return {@code True} if rollback is synchronous. */
private boolean syncRollback() {
return tx != null && tx.syncRollback();
}
/** @return {@code True} if commit is synchronous. */
private boolean syncCommit() {
return tx != null && tx.syncCommit();
}
/** @return {@code True} if transaction is not {@code null} and has invalidate flag set. */
private boolean isInvalidate() {
return tx != null && tx.isInvalidate();
}
/** @return {@code True} if implicit-single-tx flag is set. */
private boolean implicitSingleTx() {
return tx != null && tx.implicitSingle();
}