protected void handleTaskSubmittedRequest(
Runnable runnable, Address source, long requestId, long threadId) {
// We store in our map so that when that task is
// finished so that we can send back to the owner
// with the results
_running.put(runnable, new Owner(source, requestId));
// We give the task to the thread that is now waiting for it to be returned
// If we can't offer then we have to respond back to
// caller that we can't handle it. They must have
// gotten our address when we had a consumer, but
// they went away between then and now.
boolean received;
try {
_tasks.put(threadId, runnable);
CyclicBarrier barrier = _taskBarriers.remove(threadId);
if (received = (barrier != null)) {
// Only wait 10 milliseconds, in case if the consumer was
// stopped between when we were told it was available and now
barrier.await(10, TimeUnit.MILLISECONDS);
}
} catch (InterruptedException e) {
if (log.isDebugEnabled()) log.debug("Interrupted while handing off task");
Thread.currentThread().interrupt();
received = false;
} catch (BrokenBarrierException e) {
if (log.isDebugEnabled())
log.debug(
"Consumer " + threadId + " has been interrupted, " + "must retry to submit elsewhere");
received = false;
} catch (TimeoutException e) {
if (log.isDebugEnabled())
log.debug("Timeout waiting to hand off to barrier, consumer " + threadId + " must be slow");
// This should only happen if the consumer put the latch then got
// interrupted but hadn't yet removed the latch, should almost never
// happen
received = false;
}
if (!received) {
// Clean up the tasks request
_tasks.remove(threadId);
if (log.isDebugEnabled()) log.debug("Run rejected not able to pass off to consumer");
// If we couldn't hand off the task we have to tell the client
// and also reupdate the coordinator that our consumer is ready
sendRequest(source, Type.RUN_REJECTED, requestId, null);
_running.remove(runnable);
}
}
/**
* @param sesId Session ID.
* @param taskNodeId Task node ID.
* @param taskName Task name.
* @param dep Deployment.
* @param taskClsName Task class name.
* @param top Topology.
* @param startTime Execution start time.
* @param endTime Execution end time.
* @param siblings Collection of siblings.
* @param attrs Map of attributes.
* @param fullSup {@code True} to enable distributed session attributes and checkpoints.
* @return New session if one did not exist, or existing one.
*/
public GridTaskSessionImpl createTaskSession(
GridUuid sesId,
UUID taskNodeId,
String taskName,
@Nullable GridDeployment dep,
String taskClsName,
@Nullable Collection<UUID> top,
long startTime,
long endTime,
Collection<GridComputeJobSibling> siblings,
Map<Object, Object> attrs,
boolean fullSup) {
if (!fullSup) {
return new GridTaskSessionImpl(
taskNodeId,
taskName,
dep,
taskClsName,
sesId,
top,
startTime,
endTime,
siblings,
attrs,
ctx,
fullSup);
}
while (true) {
GridTaskSessionImpl ses = sesMap.get(sesId);
if (ses == null) {
GridTaskSessionImpl old =
sesMap.putIfAbsent(
sesId,
ses =
new GridTaskSessionImpl(
taskNodeId,
taskName,
dep,
taskClsName,
sesId,
top,
startTime,
endTime,
siblings,
attrs,
ctx,
fullSup));
if (old != null) ses = old;
else
// Return without acquire.
return ses;
}
if (ses.acquire()) return ses;
else sesMap.remove(sesId, ses);
}
}
/**
* @param p Partition number.
* @param topVer Topology version.
* @param create Create flag.
* @param updateSeq Update sequence.
* @return Local partition.
*/
private GridDhtLocalPartition localPartition(
int p, AffinityTopologyVersion topVer, boolean create, boolean updateSeq) {
while (true) {
boolean belongs = cctx.affinity().localNode(p, topVer);
GridDhtLocalPartition loc = locParts.get(p);
if (loc != null && loc.state() == EVICTED) {
locParts.remove(p, loc);
if (!create) return null;
if (!belongs)
throw new GridDhtInvalidPartitionException(
p,
"Adding entry to evicted partition [part="
+ p
+ ", topVer="
+ topVer
+ ", this.topVer="
+ this.topVer
+ ']');
continue;
}
if (loc == null && create) {
if (!belongs)
throw new GridDhtInvalidPartitionException(
p,
"Creating partition which does not belong [part="
+ p
+ ", topVer="
+ topVer
+ ", this.topVer="
+ this.topVer
+ ']');
lock.writeLock().lock();
try {
GridDhtLocalPartition old =
locParts.putIfAbsent(p, loc = new GridDhtLocalPartition(cctx, p));
if (old != null) loc = old;
else {
if (updateSeq) this.updateSeq.incrementAndGet();
if (log.isDebugEnabled()) log.debug("Created local partition: " + loc);
}
} finally {
lock.writeLock().unlock();
}
}
return loc;
}
}
/**
* Removes the metric with the given name.
*
* @param name the name of the metric
*/
public void removeMetric(MetricName name) {
final Metric metric = metrics.remove(name);
if (metric != null) {
if (metric instanceof Stoppable) {
((Stoppable) metric).stop();
}
notifyMetricRemoved(name);
}
}
protected void handleInterruptRequest(Address source, long requestId) {
Owner owner = new Owner(source, requestId);
Runnable runnable = removeKeyForValue(_running, owner);
Thread thread = null;
if (runnable != null) {
thread = _runnableThreads.remove(runnable);
}
if (thread != null) {
thread.interrupt();
} else if (log.isTraceEnabled())
log.trace("Message could not be interrupted due to it already returned");
}
/**
* Removes session for a given session ID.
*
* @param sesId ID of session to remove.
* @return {@code True} if session was removed.
*/
public boolean removeSession(GridUuid sesId) {
GridTaskSessionImpl ses = sesMap.get(sesId);
assert ses == null || ses.isFullSupport();
if (ses != null && ses.release()) {
sesMap.remove(sesId, ses);
return true;
}
return false;
}
/** @param entry Entry to remove. */
@SuppressWarnings("SynchronizationOnLocalVariableOrMethodParameter")
void onRemoved(GridDhtCacheEntry entry) {
assert entry.obsolete();
// Make sure to remove exactly this entry.
synchronized (entry) {
map.remove(entry.key(), entry);
if (!entry.isInternal() && !entry.deleted()) mapPubSize.decrement();
}
// Attempt to evict.
tryEvict(true);
}
private void processResponse(DataInputStream in) throws IOException {
long requestId = in.readLong();
Request r = requests.remove(requestId);
if (r == null) {
throw new IllegalStateException(
"Request " + requestId + " is unknown (last request generated was " + nextRequest.get());
}
Object o = null;
if (in.readBoolean()) {
o = serializerFor(classForName(in.readUTF()), r.getResultDeclaredType()).deserialize(in);
}
r.set(o);
}
protected void handleValueResponse(Address source, long requestId, Object value) {
Runnable runnable = _awaitingReturn.remove(new Owner(source, requestId));
if (runnable != null) {
_requestId.remove(runnable);
}
// We can only notify of success if it was a future
if (runnable instanceof RunnableFuture<?>) {
RunnableFuture<?> future = (RunnableFuture<?>) runnable;
ExecutorNotification notifier = notifiers.remove(future);
if (notifier != null) {
notifier.resultReturned(value);
}
} else {
log.warn("Runnable was not found in awaiting");
}
}
/**
* Processes cache query request.
*
* @param sndId Sender node id.
* @param req Query request.
*/
@SuppressWarnings("unchecked")
@Override
void processQueryRequest(UUID sndId, GridCacheQueryRequest req) {
if (req.cancel()) {
cancelIds.add(new CancelMessageId(req.id(), sndId));
if (req.fields()) removeFieldsQueryResult(sndId, req.id());
else removeQueryResult(sndId, req.id());
} else {
if (!cancelIds.contains(new CancelMessageId(req.id(), sndId))) {
if (!F.eq(req.cacheName(), cctx.name())) {
GridCacheQueryResponse res =
new GridCacheQueryResponse(
cctx.cacheId(),
req.id(),
new IgniteCheckedException(
"Received request for incorrect cache [expected="
+ cctx.name()
+ ", actual="
+ req.cacheName()));
sendQueryResponse(sndId, res, 0);
} else {
threads.put(req.id(), Thread.currentThread());
try {
GridCacheQueryInfo info = distributedQueryInfo(sndId, req);
if (info == null) return;
if (req.fields()) runFieldsQuery(info);
else runQuery(info);
} catch (Throwable e) {
U.error(log(), "Failed to run query.", e);
sendQueryResponse(
sndId, new GridCacheQueryResponse(cctx.cacheId(), req.id(), e.getCause()), 0);
if (e instanceof Error) throw (Error) e;
} finally {
threads.remove(req.id());
}
}
}
}
}
public void onDown(Host host) {
loadBalancer.onDown(host);
HostConnectionPool pool = pools.remove(host);
// This should not be necessary but it's harmless
if (pool != null) pool.shutdown();
// If we've remove a host, the loadBalancer is allowed to change his mind on host distances.
for (Host h : cluster.getMetadata().allHosts()) {
if (!h.getMonitor().isUp()) continue;
HostDistance dist = loadBalancer.distance(h);
if (dist != HostDistance.IGNORED) {
HostConnectionPool p = pools.get(h);
if (p == null) addHost(host);
else p.hostDistance = dist;
}
}
}
protected void handleExceptionResponse(Address source, long requestId, Throwable throwable) {
Runnable runnable = _awaitingReturn.remove(new Owner(source, requestId));
if (runnable != null) {
_requestId.remove(runnable);
}
// We can only notify of exception if it was a future
if (runnable instanceof RunnableFuture<?>) {
RunnableFuture<?> future = (RunnableFuture<?>) runnable;
ExecutorNotification notifier = notifiers.remove(future);
if (notifier != null) {
notifier.throwableEncountered(throwable);
}
} else {
// All we can do is log the error since their is no
// way to return this to the user since they don't
// have a future object.
log.error("Runtime Error encountered from cluster execute(Runnable) method", throwable);
}
}
/**
* @param entries Entries.
* @param resFut Result future.
* @param activeKeys Active keys.
* @param remaps Remaps count.
*/
private void load0(
Collection<? extends Map.Entry<K, V>> entries,
final GridFutureAdapter<Object> resFut,
final Collection<K> activeKeys,
final int remaps) {
assert entries != null;
if (remaps >= MAX_REMAP_CNT) {
resFut.onDone(new GridException("Failed to finish operation (too many remaps): " + remaps));
return;
}
Map<GridNode, Collection<Map.Entry<K, V>>> mappings = new HashMap<>();
boolean initPda = ctx.deploy().enabled() && jobPda == null;
for (Map.Entry<K, V> entry : entries) {
GridNode node;
try {
K key = entry.getKey();
assert key != null;
if (initPda) {
jobPda = new DataLoaderPda(key, entry.getValue(), updater);
initPda = false;
}
node = ctx.affinity().mapKeyToNode(cacheName, key);
} catch (GridException e) {
resFut.onDone(e);
return;
}
if (node == null) {
resFut.onDone(
new GridTopologyException(
"Failed to map key to node "
+ "(no nodes with cache found in topology) [infos="
+ entries.size()
+ ", cacheName="
+ cacheName
+ ']'));
return;
}
Collection<Map.Entry<K, V>> col = mappings.get(node);
if (col == null) mappings.put(node, col = new ArrayList<>());
col.add(entry);
}
for (final Map.Entry<GridNode, Collection<Map.Entry<K, V>>> e : mappings.entrySet()) {
final UUID nodeId = e.getKey().id();
Buffer buf = bufMappings.get(nodeId);
if (buf == null) {
Buffer old = bufMappings.putIfAbsent(nodeId, buf = new Buffer(e.getKey()));
if (old != null) buf = old;
}
final Collection<Map.Entry<K, V>> entriesForNode = e.getValue();
GridInClosure<GridFuture<?>> lsnr =
new GridInClosure<GridFuture<?>>() {
@Override
public void apply(GridFuture<?> t) {
try {
t.get();
for (Map.Entry<K, V> e : entriesForNode) activeKeys.remove(e.getKey());
if (activeKeys.isEmpty()) resFut.onDone();
} catch (GridException e1) {
if (log.isDebugEnabled())
log.debug("Future finished with error [nodeId=" + nodeId + ", err=" + e1 + ']');
if (cancelled) {
resFut.onDone(
new GridException(
"Data loader has been cancelled: " + GridDataLoaderImpl.this, e1));
} else load0(entriesForNode, resFut, activeKeys, remaps + 1);
}
}
};
GridFutureAdapter<?> f;
try {
f = buf.update(entriesForNode, lsnr);
} catch (GridInterruptedException e1) {
resFut.onDone(e1);
return;
}
if (ctx.discovery().node(nodeId) == null) {
if (bufMappings.remove(nodeId, buf)) buf.onNodeLeft();
if (f != null)
f.onDone(
new GridTopologyException(
"Failed to wait for request completion " + "(node has left): " + nodeId));
}
}
}
/** Clears values for this partition. */
private void clearAll() {
GridCacheVersion clearVer = cctx.versions().next();
boolean swap = cctx.isSwapOrOffheapEnabled();
boolean rec = cctx.events().isRecordable(EVT_CACHE_REBALANCE_OBJECT_UNLOADED);
Iterator<GridDhtCacheEntry> it = map.values().iterator();
GridCloseableIterator<Map.Entry<byte[], GridCacheSwapEntry>> swapIt = null;
if (swap
&& GridQueryProcessor.isEnabled(cctx.config())) { // Indexing needs to unswap cache values.
Iterator<GridDhtCacheEntry> unswapIt = null;
try {
swapIt = cctx.swap().iterator(id);
unswapIt = unswapIterator(swapIt);
} catch (Exception e) {
U.error(log, "Failed to clear swap for evicted partition: " + this, e);
}
if (unswapIt != null) it = F.concat(it, unswapIt);
}
try {
while (it.hasNext()) {
GridDhtCacheEntry cached = it.next();
try {
if (cached.clearInternal(clearVer, swap)) {
map.remove(cached.key(), cached);
if (!cached.isInternal()) {
mapPubSize.decrement();
if (rec)
cctx.events()
.addEvent(
cached.partition(),
cached.key(),
cctx.localNodeId(),
(IgniteUuid) null,
null,
EVT_CACHE_REBALANCE_OBJECT_UNLOADED,
null,
false,
cached.rawGet(),
cached.hasValue(),
null,
null,
null);
}
}
} catch (IgniteCheckedException e) {
U.error(log, "Failed to clear cache entry for evicted partition: " + cached, e);
}
}
} finally {
U.close(swapIt, log);
}
}
ClientEndpoint removeEndpoint(final Connection connection, boolean closeImmediately) {
final ClientEndpoint endpoint = endpoints.remove(connection);
destroyEndpoint(endpoint, closeImmediately);
return endpoint;
}
public Object down(Event evt) {
switch (evt.getType()) {
case ExecutorEvent.TASK_SUBMIT:
Runnable runnable = evt.getArg();
// We are limited to a number of concurrent request id's
// equal to 2^63-1. This is quite large and if it
// overflows it will still be positive
long requestId = Math.abs(counter.getAndIncrement());
if (requestId == Long.MIN_VALUE) {
// TODO: need to fix this it isn't safe for concurrent modifications
counter.set(0);
requestId = Math.abs(counter.getAndIncrement());
}
// Need to make sure to put the requestId in our map before
// adding the runnable to awaiting consumer in case if
// coordinator sent a consumer found and their original task
// is no longer around
// see https://issues.jboss.org/browse/JGRP-1744
_requestId.put(runnable, requestId);
_awaitingConsumer.add(runnable);
sendToCoordinator(RUN_REQUEST, requestId, local_addr);
break;
case ExecutorEvent.CONSUMER_READY:
Thread currentThread = Thread.currentThread();
long threadId = currentThread.getId();
_consumerId.put(threadId, PRESENT);
try {
for (; ; ) {
CyclicBarrier barrier = new CyclicBarrier(2);
_taskBarriers.put(threadId, barrier);
// We only send to the coordinator that we are ready after
// making the barrier, wait for request to come and let
// us free
sendToCoordinator(Type.CONSUMER_READY, threadId, local_addr);
try {
barrier.await();
break;
} catch (BrokenBarrierException e) {
if (log.isDebugEnabled())
log.debug(
"Producer timed out before we picked up"
+ " the task, have to tell coordinator"
+ " we are still good.");
}
}
// This should always be non nullable since the latch
// was freed
runnable = _tasks.remove(threadId);
_runnableThreads.put(runnable, currentThread);
return runnable;
} catch (InterruptedException e) {
if (log.isDebugEnabled()) log.debug("Consumer " + threadId + " stopped via interrupt");
sendToCoordinator(Type.CONSUMER_UNREADY, threadId, local_addr);
Thread.currentThread().interrupt();
} finally {
// Make sure the barriers are cleaned up as well
_taskBarriers.remove(threadId);
_consumerId.remove(threadId);
}
break;
case ExecutorEvent.TASK_COMPLETE:
Object arg = evt.getArg();
Throwable throwable = null;
if (arg instanceof Object[]) {
Object[] array = (Object[]) arg;
runnable = (Runnable) array[0];
throwable = (Throwable) array[1];
} else {
runnable = (Runnable) arg;
}
Owner owner = _running.remove(runnable);
// This won't remove anything if owner doesn't come back
_runnableThreads.remove(runnable);
Object value = null;
boolean exception = false;
if (throwable != null) {
// InterruptedException is special telling us that
// we interrupted the thread while waiting but still got
// a task therefore we have to reject it.
if (throwable instanceof InterruptedException) {
if (log.isDebugEnabled())
log.debug("Run rejected due to interrupted exception returned");
sendRequest(owner.address, Type.RUN_REJECTED, owner.requestId, null);
break;
}
value = throwable;
exception = true;
} else if (runnable instanceof RunnableFuture<?>) {
RunnableFuture<?> future = (RunnableFuture<?>) runnable;
boolean interrupted = false;
boolean gotValue = false;
// We have the value, before we interrupt at least get it!
while (!gotValue) {
try {
value = future.get();
gotValue = true;
} catch (InterruptedException e) {
interrupted = true;
} catch (ExecutionException e) {
value = e.getCause();
exception = true;
gotValue = true;
}
}
if (interrupted) {
Thread.currentThread().interrupt();
}
}
if (owner != null) {
final Type type;
final Object valueToSend;
if (value == null) {
type = Type.RESULT_SUCCESS;
valueToSend = value;
}
// Both serializable values and exceptions would go in here
else if (value instanceof Serializable
|| value instanceof Externalizable
|| value instanceof Streamable) {
type = exception ? Type.RESULT_EXCEPTION : Type.RESULT_SUCCESS;
valueToSend = value;
}
// This would happen if the value wasn't serializable,
// so we have to send back to the client that the class
// wasn't serializable
else {
type = Type.RESULT_EXCEPTION;
valueToSend = new NotSerializableException(value.getClass().getName());
}
if (local_addr.equals(owner.getAddress())) {
if (log.isTraceEnabled())
log.trace(
"[redirect] <--> ["
+ local_addr
+ "] "
+ type.name()
+ " ["
+ value
+ (owner.requestId != -1 ? " request id: " + owner.requestId : "")
+ "]");
if (type == Type.RESULT_SUCCESS) {
handleValueResponse(local_addr, owner.requestId, valueToSend);
} else if (type == Type.RESULT_EXCEPTION) {
handleExceptionResponse(local_addr, owner.requestId, (Throwable) valueToSend);
}
} else {
sendRequest(owner.getAddress(), type, owner.requestId, valueToSend);
}
} else {
if (log.isTraceEnabled()) {
log.trace("Could not return result - most likely because it was interrupted");
}
}
break;
case ExecutorEvent.TASK_CANCEL:
Object[] array = evt.getArg();
runnable = (Runnable) array[0];
if (_awaitingConsumer.remove(runnable)) {
_requestId.remove(runnable);
ExecutorNotification notification = notifiers.remove(runnable);
if (notification != null) {
notification.interrupted(runnable);
}
if (log.isTraceEnabled())
log.trace("Cancelled task " + runnable + " before it was picked up");
return Boolean.TRUE;
}
// This is guaranteed to not be null so don't take cost of auto unboxing
else if (array[1] == Boolean.TRUE) {
owner = removeKeyForValue(_awaitingReturn, runnable);
if (owner != null) {
Long requestIdValue = _requestId.remove(runnable);
// We only cancel if the requestId is still available
// this means the result hasn't been returned yet and
// we still have a chance to interrupt
if (requestIdValue != null) {
if (requestIdValue != owner.getRequestId()) {
log.warn("Cancelling requestId didn't match waiting");
}
sendRequest(owner.getAddress(), Type.INTERRUPT_RUN, owner.getRequestId(), null);
}
} else {
if (log.isTraceEnabled()) log.warn("Couldn't interrupt server task: " + runnable);
}
ExecutorNotification notification = notifiers.remove(runnable);
if (notification != null) {
notification.interrupted(runnable);
}
return Boolean.TRUE;
} else {
return Boolean.FALSE;
}
case ExecutorEvent.ALL_TASK_CANCEL:
array = evt.getArg();
// This is a RunnableFuture<?> so this cast is okay
@SuppressWarnings("unchecked")
Set<Runnable> runnables = (Set<Runnable>) array[0];
Boolean booleanValue = (Boolean) array[1];
List<Runnable> notRan = new ArrayList<>();
for (Runnable cancelRunnable : runnables) {
// Removed from the consumer
if (!_awaitingConsumer.remove(cancelRunnable) && booleanValue == Boolean.TRUE) {
synchronized (_awaitingReturn) {
owner = removeKeyForValue(_awaitingReturn, cancelRunnable);
if (owner != null) {
Long requestIdValue = _requestId.remove(cancelRunnable);
if (requestIdValue != owner.getRequestId()) {
log.warn("Cancelling requestId didn't match waiting");
}
sendRequest(owner.getAddress(), Type.INTERRUPT_RUN, owner.getRequestId(), null);
}
ExecutorNotification notification = notifiers.remove(cancelRunnable);
if (notification != null) {
log.trace("Notifying listener");
notification.interrupted(cancelRunnable);
}
}
} else {
_requestId.remove(cancelRunnable);
notRan.add(cancelRunnable);
}
}
return notRan;
case Event.SET_LOCAL_ADDRESS:
local_addr = evt.getArg();
break;
case Event.VIEW_CHANGE:
handleView(evt.getArg());
break;
}
return down_prot.down(evt);
}
public void onRemove(Host host) {
loadBalancer.onRemove(host);
HostConnectionPool pool = pools.remove(host);
if (pool != null) pool.shutdown();
}
private void removePool(Host host) {
HostConnectionPool pool = pools.remove(host);
if (pool != null) pool.shutdown();
}