/** * Message received callback. * * @param src Sender node ID. * @param msg Received message. * @return {@code True}. */ public boolean onMessageReceived(UUID src, GridHadoopMessage msg) { if (msg instanceof GridHadoopShuffleMessage) { GridHadoopShuffleMessage m = (GridHadoopShuffleMessage) msg; try { job(m.jobId()).onShuffleMessage(m); } catch (GridException e) { U.error(log, "Message handling failed.", e); } try { // Reply with ack. send0(src, new GridHadoopShuffleAck(m.id(), m.jobId())); } catch (GridException e) { U.error( log, "Failed to reply back to shuffle message sender [snd=" + src + ", msg=" + msg + ']', e); } } else if (msg instanceof GridHadoopShuffleAck) { GridHadoopShuffleAck m = (GridHadoopShuffleAck) msg; try { job(m.jobId()).onShuffleAck(m); } catch (GridException e) { U.error(log, "Message handling failed.", e); } } else throw new IllegalStateException( "Unknown message type received to Hadoop shuffle [src=" + src + ", msg=" + msg + ']'); return true; }
/** @param e Error. */ void onError(Throwable e) { tx.commitError(e); if (err.compareAndSet(null, e)) { boolean marked = tx.setRollbackOnly(); if (e instanceof GridCacheTxRollbackException) { if (marked) { try { tx.rollback(); } catch (GridException ex) { U.error(log, "Failed to automatically rollback transaction: " + tx, ex); } } } else if (tx.implicit() && tx.isSystemInvalidate()) { // Finish implicit transaction on heuristic error. try { tx.close(); } catch (GridException ex) { U.error(log, "Failed to invalidate transaction: " + tx, ex); } } onComplete(); } }
/** {@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 nodeId Sender node ID. * @param msg Response to prepare request. */ private void processPrepareResponse(UUID nodeId, GridDistributedTxPrepareResponse<K, V> msg) { assert nodeId != null; assert msg != null; GridReplicatedTxLocal<K, V> tx = ctx.tm().tx(msg.version()); if (tx == null) { if (log.isDebugEnabled()) log.debug( "Received prepare response for non-existing transaction [senderNodeId=" + nodeId + ", res=" + msg + ']'); return; } GridReplicatedTxPrepareFuture<K, V> future = (GridReplicatedTxPrepareFuture<K, V>) tx.future(); if (future != null) future.onResult(nodeId, msg); else U.error( log, "Received prepare response for transaction with no future [res=" + msg + ", tx=" + tx + ']'); }
/** Initializes future. */ @SuppressWarnings({"unchecked"}) void finish() { if (mappings != null) { finish(mappings.values()); markInitialized(); if (!isSync()) { boolean complete = true; for (GridFuture<?> f : pending()) // Mini-future in non-sync mode gets done when message gets sent. if (isMini(f) && !f.isDone()) complete = false; if (complete) onComplete(); } } else { assert !commit; try { tx.rollback(); } catch (GridException e) { U.error(log, "Failed to rollback empty transaction: " + tx, e); } markInitialized(); } }
/** * Processes unlock request. * * @param nodeId Sender node ID. * @param req Unlock request. */ @SuppressWarnings({"unchecked"}) private void processUnlockRequest(UUID nodeId, GridDistributedUnlockRequest req) { assert nodeId != null; try { ClassLoader ldr = ctx.deploy().globalLoader(); List<byte[]> keys = req.keyBytes(); for (byte[] keyBytes : keys) { K key = (K) U.unmarshal(ctx.marshaller(), new ByteArrayInputStream(keyBytes), ldr); while (true) { boolean created = false; GridDistributedCacheEntry<K, V> entry = peekexx(key); if (entry == null) { entry = entryexx(key); created = true; } try { entry.doneRemote( req.version(), req.version(), req.committedVersions(), req.rolledbackVersions()); // Note that we don't reorder completed versions here, // as there is no point to reorder relative to the version // we are about to remove. if (entry.removeLock(req.version())) { if (log.isDebugEnabled()) log.debug("Removed lock [lockId=" + req.version() + ", key=" + key + ']'); if (created && entry.markObsolete(req.version())) removeIfObsolete(entry.key()); } else if (log.isDebugEnabled()) log.debug( "Received unlock request for unknown candidate " + "(added to cancelled locks set): " + req); break; } catch (GridCacheEntryRemovedException ignored) { if (log.isDebugEnabled()) log.debug( "Received remove lock request for removed entry (will retry) [entry=" + entry + ", req=" + req + ']'); } } } } catch (GridException e) { U.error(log, "Failed to unmarshal unlock key (unlock will not be performed): " + req, e); } }
/** * 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; Collection<GridRichNode> nodes = ctx.remoteNodes(keys); try { // Send request to remove from remote nodes. GridDistributedUnlockRequest<K, V> req = new GridDistributedUnlockRequest<K, V>(keys.size()); req.version(ver); for (K key : keys) { while (true) { GridDistributedCacheEntry<K, V> entry = peekexx(key); try { if (entry != null) { GridCacheMvccCandidate<K> cand = entry.candidate(ver); if (cand != null) { // Remove candidate from local node first. if (entry.removeLock(cand.version())) { // If there is only local node in this lock's topology, // then there is no reason to distribute the request. if (nodes.isEmpty()) 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 (nodes.isEmpty()) return; req.completedVersions(ctx.tm().committedVersions(ver), ctx.tm().rolledbackVersions(ver)); if (!req.keyBytes().isEmpty()) // We don't wait for reply to this message. ctx.io().safeSend(nodes, req, null); } catch (GridException ex) { U.error(log, "Failed to unlock the lock for keys: " + keys, ex); } }
/** @param jobId Job id. */ public void jobFinished(GridHadoopJobId jobId) { GridHadoopShuffleJob job = jobs.remove(jobId); if (job != null) { try { job.close(); } catch (GridException e) { U.error(log, "Failed to close job: " + jobId, e); } } }
/** * Stops shuffle. * * @param cancel If should cancel all ongoing activities. */ @Override public void stop(boolean cancel) { for (GridHadoopShuffleJob job : jobs.values()) { try { job.close(); } catch (GridException e) { U.error(log, "Failed to close job.", e); } } jobs.clear(); }
void cancelAll() { GridException err = new GridException("Data loader has been cancelled: " + GridDataLoaderImpl.this); for (GridFuture<?> f : locFuts) { try { f.cancel(); } catch (GridException e) { U.error(log, "Failed to cancel mini-future.", e); } } for (GridFutureAdapter<?> f : reqs.values()) f.onDone(err); }
/** * Notifies single listener. * * @param lsnr Listener. */ private void notifyListener(GridInClosure<? super GridFuture<R>> lsnr) { assert lsnr != null; try { lsnr.apply(this); } catch (IllegalStateException ignore) { U.warn( null, "Failed to notify listener (grid is stopped) [grid=" + ctx.gridName() + ", lsnr=" + lsnr + ']'); } catch (RuntimeException e) { U.error(log, "Failed to notify listener: " + lsnr, e); throw e; } catch (Error e) { U.error(log, "Failed to notify listener: " + lsnr, e); throw e; } }
/** @param e Error. */ void onError(Throwable e) { tx.commitError(e); if (err.compareAndSet(null, e)) { boolean marked = tx.setRollbackOnly(); if (e instanceof GridCacheTxRollbackException) if (marked) { try { tx.rollback(); } catch (GridException ex) { U.error(log, "Failed to automatically rollback transaction: " + tx, ex); } } onComplete(); } }
/** * 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 GridDeployment getDeployment(GridDeploymentMetadata meta) { assert meta != null; assert ctx.config().isPeerClassLoadingEnabled(); // Validate metadata. assert meta.classLoaderId() != null; assert meta.senderNodeId() != null; assert meta.sequenceNumber() >= -1; assert meta.parentLoader() == null; if (log.isDebugEnabled()) log.debug("Starting to peer-load class based on deployment metadata: " + meta); while (true) { List<SharedDeployment> depsToCheck = null; SharedDeployment dep = null; synchronized (mux) { // Check obsolete request. if (isDeadClassLoader(meta)) return null; List<SharedDeployment> deps = cache.get(meta.userVersion()); if (deps != null) { assert !deps.isEmpty(); for (SharedDeployment d : deps) { if (d.hasParticipant(meta.senderNodeId(), meta.classLoaderId()) || meta.senderNodeId().equals(ctx.localNodeId())) { // Done. dep = d; break; } } if (dep == null) { GridTuple2<Boolean, SharedDeployment> redeployCheck = checkRedeploy(meta); if (!redeployCheck.get1()) { // Checking for redeployment encountered invalid state. if (log.isDebugEnabled()) log.debug("Checking for redeployment encountered invalid state: " + meta); return null; } dep = redeployCheck.get2(); if (dep == null) { // Find existing deployments that need to be checked // whether they should be reused for this request. for (SharedDeployment d : deps) { if (!d.isPendingUndeploy() && !d.isUndeployed()) { if (depsToCheck == null) depsToCheck = new LinkedList<SharedDeployment>(); if (log.isDebugEnabled()) log.debug("Adding deployment to check: " + d); depsToCheck.add(d); } } // If no deployment can be reused, create a new one. if (depsToCheck == null) { dep = createNewDeployment(meta, false); deps.add(dep); } } } } else { GridTuple2<Boolean, SharedDeployment> redeployCheck = checkRedeploy(meta); if (!redeployCheck.get1()) { // Checking for redeployment encountered invalid state. if (log.isDebugEnabled()) log.debug("Checking for redeployment encountered invalid state: " + meta); return null; } dep = redeployCheck.get2(); if (dep == null) // Create peer class loader. dep = createNewDeployment(meta, true); } } if (dep != null) { if (log.isDebugEnabled()) log.debug("Found SHARED or CONTINUOUS deployment after first check: " + dep); // Cache the deployed class. Class<?> cls = dep.deployedClass(meta.className(), meta.alias()); if (cls == null) { U.warn( log, "Failed to load peer class (ignore if class got undeployed during preloading) [alias=" + meta.alias() + ", dep=" + dep + ']'); return null; } return dep; } assert meta.parentLoader() == null; assert depsToCheck != null; assert !depsToCheck.isEmpty(); /* * Logic below must be performed outside of synchronization * because it involves network calls. */ // Check if class can be loaded from existing nodes. // In most cases this loop will find something. for (SharedDeployment d : depsToCheck) { // Load class. Note, that remote node will not load this class. // The class will only be loaded on this node. Class<?> cls = d.deployedClass(meta.className(), meta.alias()); if (cls != null) { synchronized (mux) { if (!d.isUndeployed() && !d.isPendingUndeploy()) { if (!addParticipant(d, meta)) return null; if (log.isDebugEnabled()) log.debug( "Acquired deployment after verifying it's availability on " + "existing nodes [depCls=" + cls + ", dep=" + d + ", meta=" + meta + ']'); return d; } } } else if (log.isDebugEnabled()) { log.debug( "Deployment cannot be reused (class does not exist on participating nodes) [dep=" + d + ", meta=" + meta + ']'); } } // We are here either because all participant nodes failed // or the class indeed should have a separate deployment. for (SharedDeployment d : depsToCheck) { // Temporary class loader. ClassLoader temp = new GridDeploymentClassLoader( GridUuid.randomUuid(), meta.userVersion(), meta.deploymentMode(), true, ctx, ctxLdr, meta.classLoaderId(), meta.senderNodeId(), meta.sequenceNumber(), comm, ctx.config().getNetworkTimeout(), log, ctx.config().getPeerClassLoadingClassPathExclude(), 0, false); String path = U.classNameToResourceName(d.sampleClassName()); // We check if any random class from existing deployment can be // loaded from sender node. If it can, then we reuse existing // deployment. InputStream rsrcIn = temp.getResourceAsStream(path); if (rsrcIn != null) { // We don't need the actual stream. U.closeQuiet(rsrcIn); synchronized (mux) { if (d.isUndeployed() || d.isPendingUndeploy()) continue; // Add new node prior to loading the class, so we attempt // to load the class from the latest node. if (!addParticipant(d, meta)) { if (log.isDebugEnabled()) log.debug( "Failed to add participant to deployment " + "[meta=" + meta + ", dep=" + dep + ']'); return null; } } Class<?> depCls = d.deployedClass(meta.className(), meta.alias()); if (depCls == null) { U.error( log, "Failed to peer load class after loading it as a resource [alias=" + meta.alias() + ", dep=" + dep + ']'); return null; } if (log.isDebugEnabled()) log.debug( "Acquired deployment class after verifying other class " + "availability on sender node [depCls=" + depCls + ", rndCls=" + d.sampleClass() + ", sampleClsName=" + d.sampleClassName() + ", meta=" + meta + ']'); return d; } else if (log.isDebugEnabled()) log.debug( "Deployment cannot be reused (random class could not be loaded from sender node) [dep=" + d + ", meta=" + meta + ']'); } synchronized (mux) { if (log.isDebugEnabled()) log.debug( "None of the existing class-loaders fit (will try to create a new one): " + meta); // Check obsolete request. if (isDeadClassLoader(meta)) return null; // Check that deployment picture has not changed. List<SharedDeployment> deps = cache.get(meta.userVersion()); if (deps != null) { assert !deps.isEmpty(); boolean retry = false; for (SharedDeployment d : deps) { // Double check if sender was already added. if (d.hasParticipant(meta.senderNodeId(), meta.classLoaderId())) { dep = d; retry = false; break; } // New deployment was added while outside of synchronization. // Need to recheck it again. if (!d.isPendingUndeploy() && !d.isUndeployed() && !depsToCheck.contains(d)) retry = true; } if (retry) { if (log.isDebugEnabled()) log.debug("Retrying due to concurrency issues: " + meta); // Outer while loop. continue; } if (dep == null) { // No new deployments were added, so we can safely add ours. dep = createNewDeployment(meta, false); deps.add(dep); if (log.isDebugEnabled()) log.debug( "Adding new deployment within second check [dep=" + dep + ", meta=" + meta + ']'); } } else { dep = createNewDeployment(meta, true); if (log.isDebugEnabled()) log.debug( "Created new deployment within second check [dep=" + dep + ", meta=" + meta + ']'); } } if (dep != null) { // Cache the deployed class. Class<?> cls = dep.deployedClass(meta.className(), meta.alias()); if (cls == null) { U.warn( log, "Failed to load peer class (ignore if class got undeployed during preloading) [alias=" + meta.alias() + ", dep=" + dep + ']'); return null; } } return dep; } }
/** * @param entries Entries to submit. * @param curFut Current future. * @throws GridInterruptedException If interrupted. */ private void submit(final List<Map.Entry<K, V>> entries, final GridFutureAdapter<Object> curFut) throws GridInterruptedException { assert entries != null; assert !entries.isEmpty(); assert curFut != null; incrementActiveTasks(); GridFuture<Object> fut; if (isLocNode) { fut = ctx.closure() .callLocalSafe( new GridDataLoadUpdateJob<>(ctx, log, cacheName, entries, false, updater), false); locFuts.add(fut); fut.listenAsync( new GridInClosure<GridFuture<Object>>() { @Override public void apply(GridFuture<Object> t) { try { boolean rmv = locFuts.remove(t); assert rmv; curFut.onDone(t.get()); } catch (GridException e) { curFut.onDone(e); } } }); } else { byte[] entriesBytes; try { entriesBytes = ctx.config().getMarshaller().marshal(entries); if (updaterBytes == null) { assert updater != null; updaterBytes = ctx.config().getMarshaller().marshal(updater); } if (topicBytes == null) topicBytes = ctx.config().getMarshaller().marshal(topic); } catch (GridException e) { U.error(log, "Failed to marshal (request will not be sent).", e); return; } GridDeployment dep = null; GridPeerDeployAware jobPda0 = null; if (ctx.deploy().enabled()) { try { jobPda0 = jobPda; assert jobPda0 != null; dep = ctx.deploy().deploy(jobPda0.deployClass(), jobPda0.classLoader()); } catch (GridException e) { U.error( log, "Failed to deploy class (request will not be sent): " + jobPda0.deployClass(), e); return; } if (dep == null) U.warn(log, "Failed to deploy class (request will be sent): " + jobPda0.deployClass()); } long reqId = idGen.incrementAndGet(); fut = curFut; reqs.put(reqId, (GridFutureAdapter<Object>) fut); GridDataLoadRequest<Object, Object> req = new GridDataLoadRequest<>( reqId, topicBytes, cacheName, updaterBytes, entriesBytes, true, dep != null ? dep.deployMode() : null, dep != null ? jobPda0.deployClass().getName() : null, dep != null ? dep.userVersion() : null, dep != null ? dep.participants() : null, dep != null ? dep.classLoaderId() : null, dep == null); try { ctx.io().send(node, TOPIC_DATALOAD, req, PUBLIC_POOL); if (log.isDebugEnabled()) log.debug("Sent request to node [nodeId=" + node.id() + ", req=" + req + ']'); } catch (GridException e) { if (ctx.discovery().alive(node) && ctx.discovery().pingNode(node.id())) ((GridFutureAdapter<Object>) fut).onDone(e); else ((GridFutureAdapter<Object>) fut) .onDone( new GridTopologyException( "Failed to send " + "request (node has left): " + node.id())); } } }
/** {@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); } }
/** {@inheritDoc} */ @SuppressWarnings({"unchecked"}) @Override public void unlockAll( Collection<? extends K> keys, GridPredicate<? super GridCacheEntry<K, V>>[] filter) { if (keys == null || keys.isEmpty()) return; Collection<? extends GridNode> nodes = ctx.remoteNodes(keys); try { GridDistributedUnlockRequest<K, V> req = new GridDistributedUnlockRequest<K, V>(keys.size()); for (K key : keys) { GridDistributedCacheEntry<K, V> entry = entryexx(key); if (!ctx.isAll(entry.wrap(false), filter)) continue; // Unlock local lock first. GridCacheMvccCandidate<K> rmv = entry.removeLock(); if (rmv != null && !nodes.isEmpty()) { if (!rmv.reentry()) { req.addKey(entry.key(), entry.getOrMarshalKeyBytes(), ctx); // We are assuming that lock ID is the same for all keys. req.version(rmv.version()); if (log.isDebugEnabled()) log.debug("Removed lock (will distribute): " + rmv); } else { if (log.isDebugEnabled()) log.debug( "Locally unlocked lock reentry without distributing to other nodes [removed=" + rmv + ", entry=" + entry + ']'); } } else { if (log.isDebugEnabled()) log.debug( "Current thread still owns lock (or there are no other nodes) [lock=" + rmv + ", curThreadId=" + Thread.currentThread().getId() + ']'); } } // Don't proceed of no keys to unlock. if (req.keyBytes().isEmpty()) { if (log.isDebugEnabled()) log.debug("No keys to unlock locally (was it reentry unlock?): " + keys); return; } // We don't wait for reply to this message. Receiving side will have // to make sure that unlock requests don't come before lock requests. ctx.io().safeSend(nodes, req, null); } catch (GridException e) { U.error(log, "Failed to unlock keys: " + keys, e); } }
/** * Processes lock request. * * @param nodeId Sender node ID. * @param msg Lock request. */ @SuppressWarnings({"unchecked", "ThrowableInstanceNeverThrown"}) private void processLockRequest(UUID nodeId, GridDistributedLockRequest<K, V> msg) { assert !nodeId.equals(locNodeId); List<byte[]> keys = msg.keyBytes(); int cnt = keys.size(); GridReplicatedTxRemote<K, V> tx = null; GridDistributedLockResponse res; ClassLoader ldr = null; try { ldr = ctx.deploy().globalLoader(); if (ldr != null) { res = new GridDistributedLockResponse(msg.version(), msg.futureId(), cnt); for (int i = 0; i < keys.size(); i++) { byte[] bytes = keys.get(i); K key = msg.keys().get(i); Collection<GridCacheMvccCandidate<K>> cands = msg.candidatesByIndex(i); if (bytes == null) continue; if (log.isDebugEnabled()) log.debug("Unmarshalled key: " + key); GridDistributedCacheEntry<K, V> entry = null; while (true) { try { entry = entryexx(key); // Handle implicit locks for pessimistic transactions. if (msg.inTx()) { tx = ctx.tm().tx(msg.version()); if (tx != null) { if (msg.txRead()) tx.addRead(key, bytes); else tx.addWrite(key, bytes); } else { tx = new GridReplicatedTxRemote<K, V>( nodeId, msg.threadId(), msg.version(), null, PESSIMISTIC, msg.isolation(), msg.isInvalidate(), msg.timeout(), key, bytes, msg.txRead(), ctx); tx = ctx.tm().onCreated(tx); if (tx == null || !ctx.tm().onStarted(tx)) throw new GridCacheTxRollbackException( "Failed to acquire lock " + "(transaction has been completed): " + msg.version()); } } // Add remote candidate before reordering. entry.addRemote( msg.nodeId(), null, msg.threadId(), msg.version(), msg.timeout(), tx != null && tx.ec(), tx != null, tx != null && tx.implicitSingle()); // Remote candidates for ordered lock queuing. entry.addRemoteCandidates( cands, msg.version(), msg.committedVersions(), msg.rolledbackVersions()); // Double-check in case if sender node left the grid. if (ctx.discovery().node(msg.nodeId()) == null) { if (log.isDebugEnabled()) log.debug( "Node requesting lock left grid (lock request will be ignored): " + msg); if (tx != null) tx.rollback(); return; } res.setCandidates( i, entry.localCandidates(), ctx.tm().committedVersions(msg.version()), ctx.tm().rolledbackVersions(msg.version())); res.addValueBytes( entry.rawGet(), msg.returnValue(i) ? entry.valueBytes(null) : null, ctx); // Entry is legit. break; } catch (GridCacheEntryRemovedException ignored) { assert entry.obsoleteVersion() != null : "Obsolete flag not set on removed entry: " + entry; if (log.isDebugEnabled()) log.debug( "Received entry removed exception (will retry on renewed entry): " + entry); if (tx != null) { tx.clearEntry(entry.key()); if (log.isDebugEnabled()) log.debug( "Cleared removed entry from remote transaction (will retry) [entry=" + entry + ", tx=" + tx + ']'); } } } } } else { String err = "Failed to acquire deployment class for message: " + msg; U.warn(log, err); res = new GridDistributedLockResponse(msg.version(), msg.futureId(), new GridException(err)); } } catch (GridCacheTxRollbackException e) { if (log.isDebugEnabled()) log.debug("Received lock request for completed transaction (will ignore): " + e); res = new GridDistributedLockResponse(msg.version(), msg.futureId(), e); } catch (GridException e) { String err = "Failed to unmarshal at least one of the keys for lock request message: " + msg; log.error(err, e); res = new GridDistributedLockResponse(msg.version(), msg.futureId(), new GridException(err, e)); if (tx != null) tx.rollback(); } catch (GridDistributedLockCancelledException ignored) { // Received lock request for cancelled lock. if (log.isDebugEnabled()) log.debug("Received lock request for canceled lock (will ignore): " + msg); if (tx != null) tx.rollback(); // Don't send response back. return; } GridNode node = ctx.discovery().node(msg.nodeId()); boolean releaseAll = false; if (node != null) { try { // Reply back to sender. ctx.io().send(node, res); } catch (GridException e) { U.error(log, "Failed to send message to node (did the node leave grid?): " + node.id(), e); releaseAll = ldr != null; } } // If sender left grid, release all locks acquired so far. else releaseAll = ldr != null; // Release all locks because sender node left grid. if (releaseAll) { for (K key : msg.keys()) { while (true) { GridDistributedCacheEntry<K, V> entry = peekexx(key); try { if (entry != null) entry.removeExplicitNodeLocks(msg.nodeId()); break; } catch (GridCacheEntryRemovedException ignore) { if (log.isDebugEnabled()) log.debug( "Attempted to remove lock on removed entity during failure " + "of replicated lock request handling (will retry): " + entry); } } } U.warn( log, "Sender node left grid in the midst of lock acquisition (locks will be released)."); } }
/** {@inheritDoc} */ @SuppressWarnings({"unchecked", "ThrowableInstanceNeverThrown"}) @Override protected GridFuture<Boolean> lockAllAsync( Collection<? extends K> keys, long timeout, GridCacheTxLocalEx<K, V> tx, boolean isInvalidate, boolean isRead, boolean retval, GridCacheTxIsolation isolation, GridPredicate<? super GridCacheEntry<K, V>>[] filter) { if (keys.isEmpty()) return new GridFinishedFuture<Boolean>(ctx.kernalContext(), true); Collection<GridRichNode> nodes = ctx.remoteNodes(keys); final GridReplicatedLockFuture<K, V> fut = new GridReplicatedLockFuture<K, V>(ctx, keys, tx, this, nodes, timeout, filter); GridDistributedLockRequest<K, V> req = new GridDistributedLockRequest<K, V>( locNodeId, Thread.currentThread().getId(), fut.futureId(), fut.version(), tx != null, isRead, isolation, isInvalidate, timeout, keys.size()); try { // Must add future before redying locks. if (!ctx.mvcc().addFuture(fut)) throw new IllegalStateException("Duplicate future ID: " + fut); boolean distribute = false; for (K key : keys) { while (true) { GridDistributedCacheEntry<K, V> entry = null; try { entry = entryexx(key); if (!ctx.isAll(entry.wrap(false), filter)) { if (log.isDebugEnabled()) log.debug("Entry being locked did not pass filter (will not lock): " + entry); fut.onDone(false); return fut; } // Removed exception may be thrown here. GridCacheMvccCandidate<K> cand = fut.addEntry(entry); if (cand != null) { req.addKeyBytes( key, cand.reentry() ? null : entry.getOrMarshalKeyBytes(), retval, entry.localCandidates(fut.version()), ctx); req.completedVersions( ctx.tm().committedVersions(fut.version()), ctx.tm().rolledbackVersions(fut.version())); distribute = !cand.reentry(); } else if (fut.isDone()) return fut; break; } catch (GridCacheEntryRemovedException ignored) { if (log.isDebugEnabled()) log.debug("Got removed entry in lockAsync(..) method (will retry): " + entry); } } } // If nothing to distribute at this point, // then all locks are reentries. if (!distribute) fut.complete(true); if (nodes.isEmpty()) fut.readyLocks(); // No reason to send request if all locks are locally re-entered, // or if timeout is negative and local locks could not be acquired. if (fut.isDone()) return fut; try { ctx.io() .safeSend( fut.nodes(), req, new P1<GridNode>() { @Override public boolean apply(GridNode node) { fut.onNodeLeft(node.id()); return !fut.isDone(); } }); } catch (GridException e) { U.error( log, "Failed to send lock request to node [nodes=" + U.toShortString(nodes) + ", req=" + req + ']', e); fut.onError(e); } return fut; } catch (GridException e) { Throwable err = new GridException("Failed to acquire asynchronous lock for keys: " + keys, e); // Clean-up. fut.onError(err); ctx.mvcc().removeFuture(fut); return fut; } }
/** * This method is called in case when method {@link #receive(UUID, Object)} threw an exception. * Insides of this method the implementation should call any of the {@code respond}, {@code stop} * or {@code skip} methods. If overriding method does nothing - than return value of method {@link * #receive(UUID, Object)} is undefined. * * <p>Default implementation simply calls method {@link #stop()}. * * @param e Exception thrown from method {@link #receive(UUID, Object)}. */ protected void onError(Throwable e) { U.error(log, "Listener operation failed.", e); stop(); }
/** {@inheritDoc} */ @Override public GridFuture<GridCacheTxEx<K, V>> prepareAsync() { GridNearTxPrepareFuture<K, V> fut = prepFut.get(); if (fut == null) { // Future must be created before any exception can be thrown. if (!prepFut.compareAndSet(null, fut = new GridNearTxPrepareFuture<K, V>(cctx, this))) return prepFut.get(); } else // Prepare was called explicitly. return fut; if (!state(PREPARING)) { if (setRollbackOnly()) { if (timedOut()) fut.onError( new GridCacheTxTimeoutException( "Transaction timed out and was rolled back: " + this)); else fut.onError( new GridException( "Invalid transaction state for prepare [state=" + state() + ", tx=" + this + ']')); } else fut.onError( new GridCacheTxRollbackException( "Invalid transaction state for prepare [state=" + state() + ", tx=" + this + ']')); return fut; } // For pessimistic mode we don't distribute prepare request. if (pessimistic()) { try { userPrepare(); if (!state(PREPARED)) { setRollbackOnly(); fut.onError( new GridException( "Invalid transaction state for commit [state=" + state() + ", tx=" + this + ']')); return fut; } fut.complete(); return fut; } catch (GridException e) { fut.onError(e); return fut; } } try { cctx.topology().readLock(); try { topologyVersion(cctx.topology().topologyVersion()); userPrepare(); } finally { cctx.topology().readUnlock(); } // This will attempt to locally commit // EVENTUALLY CONSISTENT transactions. fut.onPreparedEC(); // Make sure to add future before calling prepare. cctx.mvcc().addFuture(fut); fut.prepare(); } catch (GridCacheTxTimeoutException e) { fut.onError(e); } catch (GridCacheTxOptimisticException e) { fut.onError(e); } catch (GridException e) { setRollbackOnly(); String msg = "Failed to prepare transaction (will attempt rollback): " + this; log.error(msg, e); try { rollback(); } catch (GridException e1) { U.error(log, "Failed to rollback transaction: " + this, e1); } fut.onError(new GridCacheTxRollbackException(msg, e)); } return fut; }
/** * @param nodeId Sender node ID. * @param msg Prepare request. */ @SuppressWarnings({"InstanceofCatchParameter"}) private void processPrepareRequest(UUID nodeId, GridDistributedTxPrepareRequest<K, V> msg) { assert nodeId != null; assert msg != null; GridReplicatedTxRemote<K, V> tx = null; GridDistributedTxPrepareResponse<K, V> res; try { tx = new GridReplicatedTxRemote<K, V>( ctx.deploy().globalLoader(), nodeId, msg.threadId(), msg.version(), msg.commitVersion(), msg.concurrency(), msg.isolation(), msg.isInvalidate(), msg.timeout(), msg.reads(), msg.writes(), ctx); tx = ctx.tm().onCreated(tx); if (tx == null || !ctx.tm().onStarted(tx)) throw new GridCacheTxRollbackException("Attempt to start a completed transaction: " + tx); // Prepare prior to reordering, so the pending locks added // in prepare phase will get properly ordered as well. tx.prepare(); // Add remote candidates and reorder completed and uncompleted versions. tx.addRemoteCandidates( msg.candidatesByKey(), msg.committedVersions(), msg.rolledbackVersions()); if (msg.concurrency() == EVENTUALLY_CONSISTENT) { if (log.isDebugEnabled()) log.debug("Committing transaction during remote prepare: " + tx); tx.commit(); if (log.isDebugEnabled()) log.debug("Committed transaction during remote prepare: " + tx); // Don't send response. return; } res = new GridDistributedTxPrepareResponse<K, V>(msg.version()); Map<K, Collection<GridCacheMvccCandidate<K>>> cands = tx.localCandidates(); // Add local candidates (completed version must be set below). res.candidates(cands); } catch (GridException e) { if (e instanceof GridCacheTxRollbackException) { if (log.isDebugEnabled()) log.debug("Transaction was rolled back before prepare completed: " + tx); } else if (e instanceof GridCacheTxOptimisticException) { if (log.isDebugEnabled()) log.debug("Optimistic failure for remote transaction (will rollback): " + tx); } else { U.error(log, "Failed to process prepare request: " + msg, e); } if (tx != null) // Automatically rollback remote transactions. tx.rollback(); // Don't send response. if (msg.concurrency() == EVENTUALLY_CONSISTENT) return; res = new GridDistributedTxPrepareResponse<K, V>(msg.version()); res.error(e); } // Add completed versions. res.completedVersions( ctx.tm().committedVersions(msg.version()), ctx.tm().rolledbackVersions(msg.version())); assert msg.concurrency() != EVENTUALLY_CONSISTENT; GridNode node = ctx.discovery().node(nodeId); if (node != null) { try { // Reply back to sender. ctx.io().send(node, res); } catch (GridException e) { U.error( log, "Failed to send tx response to node (did the node leave grid?) [node=" + node.id() + ", msg=" + res + ']', e); if (tx != null) tx.rollback(); } } }
/** * Removes obsolete deployments in case of redeploy. * * @param meta Request metadata. * @return List of shares deployment. */ private GridTuple2<Boolean, SharedDeployment> checkRedeploy(GridDeploymentMetadata meta) { assert Thread.holdsLock(mux); SharedDeployment newDep = null; for (List<SharedDeployment> deps : cache.values()) { for (SharedDeployment dep : deps) { if (!dep.isUndeployed() && !dep.isPendingUndeploy()) { long undeployTimeout = ctx.config().getNetworkTimeout(); SharedDeployment doomed = null; // Only check deployments with no participants. if (!dep.hasParticipants()) { // In case of SHARED deployment it is possible to get hear if // unmarshalling happens during undeploy. In this case, we // simply don't do anything. if (dep.deployMode() == CONTINUOUS) { if (dep.existingDeployedClass(meta.className()) != null) { // Change from shared deploy to shared undeploy or user version change. // Simply remove all deployments with no participating nodes. if (meta.deploymentMode() == SHARED || !meta.userVersion().equals(dep.userVersion())) doomed = dep; } } } // If there are participants, we undeploy if class loader ID on some node changed. else if (dep.existingDeployedClass(meta.className()) != null) { GridTuple2<GridUuid, Long> ldr = dep.getClassLoaderId(meta.senderNodeId()); if (ldr != null) { if (!ldr.get1().equals(meta.classLoaderId())) { // If deployed sequence number is less, then schedule for undeployment. if (ldr.get2() < meta.sequenceNumber()) { if (log.isDebugEnabled()) log.debug( "Received request for a class with newer sequence number " + "(will schedule current class for undeployment) [newSeq=" + meta.sequenceNumber() + ", oldSeq=" + ldr.get2() + ", senderNodeId=" + meta.senderNodeId() + ", newClsLdrId=" + meta.classLoaderId() + ", oldClsLdrId=" + ldr.get1() + ']'); doomed = dep; } else if (ldr.get2() > meta.sequenceNumber()) { long time = System.currentTimeMillis() - dep.timestamp(); if (newDep == null && time < ctx.config().getNetworkTimeout()) { // Set undeployTimeout, so the class will be scheduled // for undeployment. undeployTimeout = ctx.config().getNetworkTimeout() - time; if (log.isDebugEnabled()) log.debug( "Received execution request for a stale class (will deploy and " + "schedule undeployment in " + undeployTimeout + "ms) " + "[curSeq=" + ldr.get2() + ", staleSeq=" + meta.sequenceNumber() + ", cls=" + meta.className() + ", senderNodeId=" + meta.senderNodeId() + ", curLdrId=" + ldr.get1() + ", staleLdrId=" + meta.classLoaderId() + ']'); // We got the redeployed class before the old one. // Simply create a temporary deployment for the sender node, // and schedule undeploy for it. newDep = createNewDeployment(meta, false); doomed = newDep; } else { U.warn( log, "Received execution request for a class that has been redeployed " + "(will ignore): " + meta.alias()); if (log.isDebugEnabled()) log.debug( "Received execution request for a class that has been redeployed " + "(will ignore) [alias=" + meta.alias() + ", dep=" + dep + ']'); return F.t(false, null); } } else { U.error( log, "Sequence number does not correspond to class loader ID [seqNum=" + meta.sequenceNumber() + ", dep=" + dep + ']'); return F.t(false, null); } } } } if (doomed != null) { doomed.onUndeployScheduled(); if (log.isDebugEnabled()) log.debug("Deployment was scheduled for undeploy: " + doomed); // Lifespan time. final long endTime = System.currentTimeMillis() + undeployTimeout; // Deployment to undeploy. final SharedDeployment undep = doomed; ctx.timeout() .addTimeoutObject( new GridTimeoutObject() { @Override public GridUuid timeoutId() { return undep.classLoaderId(); } @Override public long endTime() { return endTime < 0 ? Long.MAX_VALUE : endTime; } @Override public void onTimeout() { boolean removed = false; // Hot redeployment. synchronized (mux) { assert undep.isPendingUndeploy(); if (!undep.isUndeployed()) { undep.undeploy(); undep.onRemoved(); removed = true; Collection<SharedDeployment> deps = cache.get(undep.userVersion()); if (deps != null) { for (Iterator<SharedDeployment> i = deps.iterator(); i.hasNext(); ) if (i.next() == undep) i.remove(); if (deps.isEmpty()) cache.remove(undep.userVersion()); } if (log.isInfoEnabled()) log.info( "Undeployed class loader due to deployment mode change, " + "user version change, or hot redeployment: " + undep); } } // Outside synchronization. if (removed) undep.recordUndeployed(null); } }); } } } } if (newDep != null) { List<SharedDeployment> list = F.addIfAbsent(cache, meta.userVersion(), F.<SharedDeployment>newList()); assert list != null; list.add(newDep); } return F.t(true, newDep); }
/** * Process HTTP request. * * @param act Action. * @param req Http request. * @param res Http response. */ private void processRequest(String act, HttpServletRequest req, HttpServletResponse res) { res.setContentType("application/json"); res.setCharacterEncoding("UTF-8"); GridRestCommand cmd = command(req); if (cmd == null) { res.setStatus(HttpServletResponse.SC_BAD_REQUEST); return; } if (!authChecker.apply(req.getHeader("X-Signature"))) { res.setStatus(HttpServletResponse.SC_UNAUTHORIZED); return; } GridRestResponse cmdRes; Map<String, Object> params = parameters(req); try { GridRestRequest cmdReq = createRequest(cmd, params, req); if (log.isDebugEnabled()) log.debug("Initialized command request: " + cmdReq); cmdRes = hnd.handle(cmdReq); if (cmdRes == null) throw new IllegalStateException("Received null result from handler: " + hnd); byte[] sesTok = cmdRes.sessionTokenBytes(); if (sesTok != null) cmdRes.setSessionToken(U.byteArray2HexString(sesTok)); res.setStatus(HttpServletResponse.SC_OK); } catch (Exception e) { res.setStatus(HttpServletResponse.SC_OK); U.error(log, "Failed to process HTTP request [action=" + act + ", req=" + req + ']', e); cmdRes = new GridRestResponse(STATUS_FAILED, e.getMessage()); } catch (Throwable e) { U.error(log, "Failed to process HTTP request [action=" + act + ", req=" + req + ']', e); throw e; } JsonConfig cfg = new GridJettyJsonConfig(); // Workaround for not needed transformation of string into JSON object. if (cmdRes.getResponse() instanceof String) cfg.registerJsonValueProcessor(cmdRes.getClass(), "response", SKIP_STR_VAL_PROC); if (cmdRes.getResponse() instanceof GridClientTaskResultBean && ((GridClientTaskResultBean) cmdRes.getResponse()).getResult() instanceof String) cfg.registerJsonValueProcessor(cmdRes.getResponse().getClass(), "result", SKIP_STR_VAL_PROC); JSON json; try { json = JSONSerializer.toJSON(cmdRes, cfg); } catch (JSONException e) { U.error(log, "Failed to convert response to JSON: " + cmdRes, e); json = JSONSerializer.toJSON(new GridRestResponse(STATUS_FAILED, e.getMessage()), cfg); } try { if (log.isDebugEnabled()) log.debug("Parsed command response into JSON object: " + json.toString(2)); res.getWriter().write(json.toString()); if (log.isDebugEnabled()) log.debug( "Processed HTTP request [action=" + act + ", jsonRes=" + cmdRes + ", req=" + req + ']'); } catch (IOException e) { U.error(log, "Failed to send HTTP response: " + json.toString(2), e); } }
/** * @param nodeId Sender node ID. * @param req Finish transaction message. */ @SuppressWarnings({"CatchGenericClass"}) private void processFinishRequest(UUID nodeId, GridDistributedTxFinishRequest<K, V> req) { assert nodeId != null; assert req != null; GridReplicatedTxRemote<K, V> tx = ctx.tm().tx(req.version()); try { ClassLoader ldr = ctx.deploy().globalLoader(); if (req.commit()) { // If lock was acquired explicitly. if (tx == null) { // Create transaction and add entries. tx = ctx.tm() .onCreated( new GridReplicatedTxRemote<K, V>( ldr, nodeId, req.threadId(), req.version(), req.commitVersion(), PESSIMISTIC, READ_COMMITTED, req.isInvalidate(), /*timeout */ 0, /*read entries*/ null, req.writes(), ctx)); if (tx == null || !ctx.tm().onStarted(tx)) throw new GridCacheTxRollbackException( "Attempt to start a completed " + "transaction: " + req); } else { boolean set = tx.commitVersion(req.commitVersion()); assert set; } Collection<GridCacheTxEntry<K, V>> writeEntries = req.writes(); if (!F.isEmpty(writeEntries)) { // In OPTIMISTIC mode, we get the values at PREPARE stage. assert tx.concurrency() == PESSIMISTIC; for (GridCacheTxEntry<K, V> entry : writeEntries) { // Unmarshal write entries. entry.unmarshal(ctx, ldr); if (log.isDebugEnabled()) log.debug( "Unmarshalled transaction entry from pessimistic transaction [key=" + entry.key() + ", value=" + entry.value() + ", tx=" + tx + ']'); if (!tx.setWriteValue(entry)) U.warn( log, "Received entry to commit that was not present in transaction [entry=" + entry + ", tx=" + tx + ']'); } } // Add completed versions. tx.doneRemote(req.baseVersion(), req.committedVersions(), req.rolledbackVersions()); if (tx.pessimistic()) tx.prepare(); tx.commit(); } else if (tx != null) { tx.doneRemote(req.baseVersion(), req.committedVersions(), req.rolledbackVersions()); tx.rollback(); } if (req.replyRequired()) { GridCacheMessage<K, V> res = new GridDistributedTxFinishResponse<K, V>(req.version(), req.futureId()); try { ctx.io().send(nodeId, res); } catch (Throwable e) { // Double-check. if (ctx.discovery().node(nodeId) == null) { if (log.isDebugEnabled()) log.debug( "Node left while sending finish response [nodeId=" + nodeId + ", res=" + res + ']'); } else U.error( log, "Failed to send finish response to node [nodeId=" + nodeId + ", res=" + res + ']', e); } } } catch (GridCacheTxRollbackException e) { if (log.isDebugEnabled()) log.debug("Attempted to start a completed transaction (will ignore): " + e); } catch (Throwable e) { U.error( log, "Failed completing transaction [commit=" + req.commit() + ", tx=" + CU.txString(tx) + ']', e); if (tx != null) tx.rollback(); } }