use of io.trino.execution.TaskStatus in project trino by trinodb.
the class HttpRemoteTask method cancel.
@Override
public synchronized void cancel() {
try (SetThreadName ignored = new SetThreadName("HttpRemoteTask-%s", taskId)) {
TaskStatus taskStatus = getTaskStatus();
if (taskStatus.getState().isDone()) {
return;
}
// send cancel to task and ignore response
scheduleAsyncCleanupRequest(new Backoff(maxErrorDuration), "cancel", false);
}
}
use of io.trino.execution.TaskStatus in project trino by trinodb.
the class FaultTolerantStageScheduler method schedule.
public synchronized void schedule() throws Exception {
if (failure != null) {
propagateIfPossible(failure, Exception.class);
throw new RuntimeException(failure);
}
if (closed) {
return;
}
if (isFinished()) {
return;
}
if (!blocked.isDone()) {
return;
}
if (taskSource == null) {
Map<PlanFragmentId, ListenableFuture<List<ExchangeSourceHandle>>> sourceHandles = sourceExchanges.entrySet().stream().collect(toImmutableMap(Map.Entry::getKey, entry -> toListenableFuture(entry.getValue().getSourceHandles())));
List<ListenableFuture<List<ExchangeSourceHandle>>> blockedFutures = sourceHandles.values().stream().filter(future -> !future.isDone()).collect(toImmutableList());
if (!blockedFutures.isEmpty()) {
blocked = asVoid(allAsList(blockedFutures));
return;
}
Multimap<PlanFragmentId, ExchangeSourceHandle> exchangeSources = sourceHandles.entrySet().stream().collect(flatteningToImmutableListMultimap(Map.Entry::getKey, entry -> getFutureValue(entry.getValue()).stream()));
taskSource = taskSourceFactory.create(session, stage.getFragment(), sourceExchanges, exchangeSources, stage::recordGetSplitTime, sourceBucketToPartitionMap, sourceBucketNodeMap);
}
while (!queuedPartitions.isEmpty() || !taskSource.isFinished()) {
while (queuedPartitions.isEmpty() && !taskSource.isFinished()) {
List<TaskDescriptor> tasks = taskSource.getMoreTasks();
for (TaskDescriptor task : tasks) {
queuedPartitions.add(task.getPartitionId());
allPartitions.add(task.getPartitionId());
taskDescriptorStorage.put(stage.getStageId(), task);
sinkExchange.ifPresent(exchange -> {
ExchangeSinkHandle exchangeSinkHandle = exchange.addSink(task.getPartitionId());
partitionToExchangeSinkHandleMap.put(task.getPartitionId(), exchangeSinkHandle);
});
}
if (taskSource.isFinished()) {
sinkExchange.ifPresent(Exchange::noMoreSinks);
}
}
if (queuedPartitions.isEmpty()) {
break;
}
int partition = queuedPartitions.peek();
Optional<TaskDescriptor> taskDescriptorOptional = taskDescriptorStorage.get(stage.getStageId(), partition);
if (taskDescriptorOptional.isEmpty()) {
// query has been terminated
return;
}
TaskDescriptor taskDescriptor = taskDescriptorOptional.get();
MemoryRequirements memoryRequirements = partitionMemoryRequirements.computeIfAbsent(partition, ignored -> partitionMemoryEstimator.getInitialMemoryRequirements(session, taskDescriptor.getNodeRequirements().getMemory()));
if (nodeLease == null) {
NodeRequirements nodeRequirements = taskDescriptor.getNodeRequirements();
nodeRequirements = nodeRequirements.withMemory(memoryRequirements.getRequiredMemory());
nodeLease = nodeAllocator.acquire(nodeRequirements);
}
if (!nodeLease.getNode().isDone()) {
blocked = asVoid(nodeLease.getNode());
return;
}
NodeInfo node = getFutureValue(nodeLease.getNode());
queuedPartitions.poll();
Multimap<PlanNodeId, Split> tableScanSplits = taskDescriptor.getSplits();
Multimap<PlanNodeId, Split> remoteSplits = createRemoteSplits(taskDescriptor.getExchangeSourceHandles());
Multimap<PlanNodeId, Split> taskSplits = ImmutableListMultimap.<PlanNodeId, Split>builder().putAll(tableScanSplits).putAll(remoteSplits).build();
int attemptId = getNextAttemptIdForPartition(partition);
OutputBuffers outputBuffers;
Optional<ExchangeSinkInstanceHandle> exchangeSinkInstanceHandle;
if (sinkExchange.isPresent()) {
ExchangeSinkHandle sinkHandle = partitionToExchangeSinkHandleMap.get(partition);
exchangeSinkInstanceHandle = Optional.of(sinkExchange.get().instantiateSink(sinkHandle, attemptId));
outputBuffers = createSpoolingExchangeOutputBuffers(exchangeSinkInstanceHandle.get());
} else {
exchangeSinkInstanceHandle = Optional.empty();
// stage will be consumed by the coordinator using direct exchange
outputBuffers = createInitialEmptyOutputBuffers(PARTITIONED).withBuffer(new OutputBuffers.OutputBufferId(0), 0).withNoMoreBufferIds();
}
Set<PlanNodeId> allSourcePlanNodeIds = ImmutableSet.<PlanNodeId>builder().addAll(stage.getFragment().getPartitionedSources()).addAll(stage.getFragment().getRemoteSourceNodes().stream().map(RemoteSourceNode::getId).iterator()).build();
RemoteTask task = stage.createTask(node.getNode(), partition, attemptId, sinkBucketToPartitionMap, outputBuffers, taskSplits, allSourcePlanNodeIds.stream().collect(toImmutableListMultimap(Function.identity(), planNodeId -> Lifespan.taskWide())), allSourcePlanNodeIds).orElseThrow(() -> new VerifyException("stage execution is expected to be active"));
partitionToRemoteTaskMap.put(partition, task);
runningTasks.put(task.getTaskId(), task);
runningNodes.put(task.getTaskId(), nodeLease);
nodeLease = null;
if (taskFinishedFuture == null) {
taskFinishedFuture = SettableFuture.create();
}
taskLifecycleListener.taskCreated(stage.getFragment().getId(), task);
task.addStateChangeListener(taskStatus -> updateTaskStatus(taskStatus, exchangeSinkInstanceHandle));
task.start();
}
if (taskFinishedFuture != null && !taskFinishedFuture.isDone()) {
blocked = taskFinishedFuture;
}
}
use of io.trino.execution.TaskStatus in project trino by trinodb.
the class TestHttpRemoteTask method createHttpRemoteTaskFactory.
private static HttpRemoteTaskFactory createHttpRemoteTaskFactory(TestingTaskResource testingTaskResource, DynamicFilterService dynamicFilterService) {
Bootstrap app = new Bootstrap(new JsonModule(), new HandleJsonModule(), new Module() {
@Override
public void configure(Binder binder) {
binder.bind(JsonMapper.class).in(SINGLETON);
binder.bind(Metadata.class).toInstance(createTestMetadataManager());
jsonBinder(binder).addDeserializerBinding(Type.class).to(TypeDeserializer.class);
jsonCodecBinder(binder).bindJsonCodec(TaskStatus.class);
jsonCodecBinder(binder).bindJsonCodec(VersionedDynamicFilterDomains.class);
jsonBinder(binder).addSerializerBinding(Block.class).to(BlockJsonSerde.Serializer.class);
jsonBinder(binder).addDeserializerBinding(Block.class).to(BlockJsonSerde.Deserializer.class);
jsonCodecBinder(binder).bindJsonCodec(TaskInfo.class);
jsonCodecBinder(binder).bindJsonCodec(TaskUpdateRequest.class);
jsonCodecBinder(binder).bindJsonCodec(FailTaskRequest.class);
binder.bind(TypeManager.class).toInstance(TESTING_TYPE_MANAGER);
binder.bind(BlockEncodingManager.class).in(SINGLETON);
binder.bind(BlockEncodingSerde.class).to(InternalBlockEncodingSerde.class).in(SINGLETON);
}
@Provides
private HttpRemoteTaskFactory createHttpRemoteTaskFactory(JsonMapper jsonMapper, JsonCodec<TaskStatus> taskStatusCodec, JsonCodec<VersionedDynamicFilterDomains> dynamicFilterDomainsCodec, JsonCodec<TaskInfo> taskInfoCodec, JsonCodec<TaskUpdateRequest> taskUpdateRequestCodec, JsonCodec<FailTaskRequest> failTaskRequestCodec) {
JaxrsTestingHttpProcessor jaxrsTestingHttpProcessor = new JaxrsTestingHttpProcessor(URI.create("http://fake.invalid/"), testingTaskResource, jsonMapper);
TestingHttpClient testingHttpClient = new TestingHttpClient(jaxrsTestingHttpProcessor.setTrace(TRACE_HTTP));
testingTaskResource.setHttpClient(testingHttpClient);
return new HttpRemoteTaskFactory(new QueryManagerConfig(), TASK_MANAGER_CONFIG, testingHttpClient, new TestSqlTaskManager.MockLocationFactory(), taskStatusCodec, dynamicFilterDomainsCodec, taskInfoCodec, taskUpdateRequestCodec, failTaskRequestCodec, new RemoteTaskStats(), dynamicFilterService);
}
});
Injector injector = app.doNotInitializeLogging().quiet().initialize();
return injector.getInstance(HttpRemoteTaskFactory.class);
}
use of io.trino.execution.TaskStatus in project trino by trinodb.
the class TaskResource method getTaskStatus.
@ResourceSecurity(INTERNAL_ONLY)
@GET
@Path("{taskId}/status")
@Produces(MediaType.APPLICATION_JSON)
public void getTaskStatus(@PathParam("taskId") TaskId taskId, @HeaderParam(TRINO_CURRENT_VERSION) Long currentVersion, @HeaderParam(TRINO_MAX_WAIT) Duration maxWait, @Context UriInfo uriInfo, @Suspended AsyncResponse asyncResponse) {
requireNonNull(taskId, "taskId is null");
if (injectFailure(taskManager.getTraceToken(taskId), taskId, RequestType.GET_TASK_STATUS, asyncResponse)) {
return;
}
if (currentVersion == null || maxWait == null) {
TaskStatus taskStatus = taskManager.getTaskStatus(taskId);
asyncResponse.resume(taskStatus);
return;
}
Duration waitTime = randomizeWaitTime(maxWait);
// TODO: With current implementation, a newly completed driver group won't trigger immediate HTTP response,
// leading to a slight delay of approx 1 second, which is not a major issue for any query that are heavy weight enough
// to justify group-by-group execution. In order to fix this, REST endpoint /v1/{task}/status will need change.
ListenableFuture<TaskStatus> futureTaskStatus = addTimeout(taskManager.getTaskStatus(taskId, currentVersion), () -> taskManager.getTaskStatus(taskId), waitTime, timeoutExecutor);
// For hard timeout, add an additional time to max wait for thread scheduling contention and GC
Duration timeout = new Duration(waitTime.toMillis() + ADDITIONAL_WAIT_TIME.toMillis(), MILLISECONDS);
bindAsyncResponse(asyncResponse, futureTaskStatus, responseExecutor).withTimeout(timeout);
}
use of io.trino.execution.TaskStatus in project trino by trinodb.
the class HttpRemoteTask method fail.
/**
* Move the task directly to the failed state if there was a failure in this task
*/
@Override
public synchronized void fail(Throwable cause) {
TaskStatus taskStatus = getTaskStatus();
if (!taskStatus.getState().isDone()) {
log.debug(cause, "Remote task %s failed with %s", taskStatus.getSelf(), cause);
}
TaskStatus status = failWith(getTaskStatus(), FAILED, ImmutableList.of(toFailure(cause)));
taskStatusFetcher.updateTaskStatus(status);
try (SetThreadName ignored = new SetThreadName("HttpRemoteTask-%s", taskId)) {
if (cause instanceof TrinoTransportException) {
// task is unreachable
cleanUpLocally();
} else {
// send abort to task
scheduleAsyncCleanupRequest(new Backoff(maxErrorDuration), "abort", true);
}
}
}
Aggregations