@Override
public void onSuccess(TaskRunner2Result result) {
knownTasks.remove(taskWrapper.getRequestId());
taskWrapper.setIsInPreemptableQueue(false);
taskWrapper.maybeUnregisterForFinishedStateNotifications();
taskWrapper.getTaskRunnerCallable().getCallback().onSuccess(result);
updatePreemptionListAndNotify(result.getEndReason());
}
// Errors are handled on the way over. FAIL/SUCCESS is informed via regular heartbeats. Killed
// via a kill message when a task kill is requested by the daemon.
@Override
public void onSuccess(TaskRunner2Result result) {
isCompleted.set(true);
switch (result.getEndReason()) {
// Only the KILLED case requires a message to be sent out to the AM.
case SUCCESS:
LOG.debug("Successfully finished {}", requestId);
metrics.incrExecutorTotalSuccess();
break;
case CONTAINER_STOP_REQUESTED:
LOG.info("Received container stop request (AM preemption) for {}", requestId);
break;
case KILL_REQUESTED:
LOG.info("Killed task {}", requestId);
if (killtimerWatch.isRunning()) {
killtimerWatch.stop();
long elapsed = killtimerWatch.elapsedMillis();
LOG.info("Time to die for task {}", elapsed);
}
metrics.incrPreemptionTimeLost(runtimeWatch.elapsedMillis());
metrics.incrExecutorTotalKilled();
break;
case COMMUNICATION_FAILURE:
LOG.info("Failed to run {} due to communication failure", requestId);
metrics.incrExecutorTotalExecutionFailed();
break;
case TASK_ERROR:
LOG.info("Failed to run {} due to task error", requestId);
metrics.incrExecutorTotalExecutionFailed();
break;
}
fragmentCompletionHanler.fragmentComplete(fragmentInfo);
taskRunnerCallable.shutdown();
HistoryLogger.logFragmentEnd(
request.getApplicationIdString(),
request.getContainerIdString(),
executionContext.getHostName(),
request.getFragmentSpec().getDagName(),
request.getFragmentSpec().getVertexName(),
request.getFragmentSpec().getFragmentNumber(),
request.getFragmentSpec().getAttemptNumber(),
taskRunnerCallable.threadName,
taskRunnerCallable.startTime,
true);
metrics.decrExecutorNumQueuedRequests();
}
@Override
protected TaskRunner2Result callInternal() throws Exception {
isStarted.set(true);
this.startTime = System.currentTimeMillis();
this.threadName = Thread.currentThread().getName();
if (LOG.isDebugEnabled()) {
LOG.debug("canFinish: " + taskSpec.getTaskAttemptID() + ": " + canFinish());
}
// Unregister from the AMReporter, since the task is now running.
this.amReporter.unregisterTask(request.getAmHost(), request.getAmPort());
synchronized (this) {
if (!shouldRunTask) {
LOG.info("Not starting task {} since it was killed earlier", taskSpec.getTaskAttemptID());
return new TaskRunner2Result(EndReason.KILL_REQUESTED, null, false);
}
}
// TODO This executor seems unnecessary. Here and TezChild
ExecutorService executorReal =
Executors.newFixedThreadPool(
1, new ThreadFactoryBuilder().setDaemon(true).setNameFormat("TezTaskRunner").build());
executor = MoreExecutors.listeningDecorator(executorReal);
// TODO Consolidate this code with TezChild.
runtimeWatch.start();
UserGroupInformation taskUgi = UserGroupInformation.createRemoteUser(request.getUser());
taskUgi.addCredentials(credentials);
Map<String, ByteBuffer> serviceConsumerMetadata = new HashMap<>();
serviceConsumerMetadata.put(
TezConstants.TEZ_SHUFFLE_HANDLER_SERVICE_ID,
TezCommonUtils.convertJobTokenToBytes(jobToken));
Multimap<String, String> startedInputsMap = createStartedInputMap(request.getFragmentSpec());
UserGroupInformation taskOwner =
UserGroupInformation.createRemoteUser(request.getTokenIdentifier());
final InetSocketAddress address =
NetUtils.createSocketAddrForHost(request.getAmHost(), request.getAmPort());
SecurityUtil.setTokenService(jobToken, address);
taskOwner.addToken(jobToken);
umbilical =
taskOwner.doAs(
new PrivilegedExceptionAction<LlapTaskUmbilicalProtocol>() {
@Override
public LlapTaskUmbilicalProtocol run() throws Exception {
return RPC.getProxy(
LlapTaskUmbilicalProtocol.class,
LlapTaskUmbilicalProtocol.versionID,
address,
conf);
}
});
taskReporter =
new LlapTaskReporter(
umbilical,
confParams.amHeartbeatIntervalMsMax,
confParams.amCounterHeartbeatInterval,
confParams.amMaxEventsPerHeartbeat,
new AtomicLong(0),
request.getContainerIdString());
String attemptId = fragmentInfo.getFragmentIdentifierString();
IOContextMap.setThreadAttemptId(attemptId);
try {
synchronized (this) {
if (shouldRunTask) {
taskRunner =
new TezTaskRunner2(
conf,
taskUgi,
fragmentInfo.getLocalDirs(),
taskSpec,
request.getAppAttemptNumber(),
serviceConsumerMetadata,
envMap,
startedInputsMap,
taskReporter,
executor,
objectRegistry,
pid,
executionContext,
memoryAvailable,
false);
}
}
if (taskRunner == null) {
LOG.info("Not starting task {} since it was killed earlier", taskSpec.getTaskAttemptID());
return new TaskRunner2Result(EndReason.KILL_REQUESTED, null, false);
}
try {
TaskRunner2Result result = taskRunner.run();
if (result.isContainerShutdownRequested()) {
LOG.warn("Unexpected container shutdown requested while running task. Ignoring");
}
isCompleted.set(true);
return result;
} finally {
FileSystem.closeAllForUGI(taskUgi);
LOG.info(
"ExecutionTime for Container: "
+ request.getContainerIdString()
+ "="
+ runtimeWatch.stop().elapsedMillis());
if (LOG.isDebugEnabled()) {
LOG.debug(
"canFinish post completion: " + taskSpec.getTaskAttemptID() + ": " + canFinish());
}
}
} finally {
IOContextMap.clearThreadAttempt(attemptId);
}
}
