use of org.apache.hadoop.mapreduce.v2.app.job.TaskAttemptStateInternal in project hadoop by apache.
the class TaskAttemptImpl method recover.
@SuppressWarnings("unchecked")
public TaskAttemptStateInternal recover(TaskAttemptInfo taInfo, OutputCommitter committer, boolean recoverOutput) {
ContainerId containerId = taInfo.getContainerId();
NodeId containerNodeId = NodeId.fromString(taInfo.getHostname() + ":" + taInfo.getPort());
String nodeHttpAddress = StringInterner.weakIntern(taInfo.getHostname() + ":" + taInfo.getHttpPort());
// Resource/Priority/Tokens are only needed while launching the container on
// an NM, these are already completed tasks, so setting them to null
container = Container.newInstance(containerId, containerNodeId, nodeHttpAddress, null, null, null);
computeRackAndLocality();
launchTime = taInfo.getStartTime();
finishTime = (taInfo.getFinishTime() != -1) ? taInfo.getFinishTime() : clock.getTime();
shufflePort = taInfo.getShufflePort();
trackerName = taInfo.getHostname();
httpPort = taInfo.getHttpPort();
sendLaunchedEvents();
reportedStatus.id = attemptId;
reportedStatus.progress = 1.0f;
reportedStatus.counters = taInfo.getCounters();
reportedStatus.stateString = taInfo.getState();
reportedStatus.phase = Phase.CLEANUP;
reportedStatus.mapFinishTime = taInfo.getMapFinishTime();
reportedStatus.shuffleFinishTime = taInfo.getShuffleFinishTime();
reportedStatus.sortFinishTime = taInfo.getSortFinishTime();
addDiagnosticInfo(taInfo.getError());
boolean needToClean = false;
String recoveredState = taInfo.getTaskStatus();
if (recoverOutput && TaskAttemptState.SUCCEEDED.toString().equals(recoveredState)) {
TaskAttemptContext tac = new TaskAttemptContextImpl(conf, TypeConverter.fromYarn(attemptId));
try {
committer.recoverTask(tac);
LOG.info("Recovered output from task attempt " + attemptId);
} catch (Exception e) {
LOG.error("Unable to recover task attempt " + attemptId, e);
LOG.info("Task attempt " + attemptId + " will be recovered as KILLED");
recoveredState = TaskAttemptState.KILLED.toString();
needToClean = true;
}
}
TaskAttemptStateInternal attemptState;
if (TaskAttemptState.SUCCEEDED.toString().equals(recoveredState)) {
attemptState = TaskAttemptStateInternal.SUCCEEDED;
reportedStatus.taskState = TaskAttemptState.SUCCEEDED;
eventHandler.handle(createJobCounterUpdateEventTASucceeded(this));
logAttemptFinishedEvent(attemptState);
} else if (TaskAttemptState.FAILED.toString().equals(recoveredState)) {
attemptState = TaskAttemptStateInternal.FAILED;
reportedStatus.taskState = TaskAttemptState.FAILED;
eventHandler.handle(createJobCounterUpdateEventTAFailed(this, false));
TaskAttemptUnsuccessfulCompletionEvent tauce = createTaskAttemptUnsuccessfulCompletionEvent(this, TaskAttemptStateInternal.FAILED);
eventHandler.handle(new JobHistoryEvent(attemptId.getTaskId().getJobId(), tauce));
} else {
if (!TaskAttemptState.KILLED.toString().equals(recoveredState)) {
if (String.valueOf(recoveredState).isEmpty()) {
LOG.info("TaskAttempt" + attemptId + " had not completed, recovering as KILLED");
} else {
LOG.warn("TaskAttempt " + attemptId + " found in unexpected state " + recoveredState + ", recovering as KILLED");
}
addDiagnosticInfo("Killed during application recovery");
needToClean = true;
}
attemptState = TaskAttemptStateInternal.KILLED;
reportedStatus.taskState = TaskAttemptState.KILLED;
eventHandler.handle(createJobCounterUpdateEventTAKilled(this, false));
TaskAttemptUnsuccessfulCompletionEvent tauce = createTaskAttemptUnsuccessfulCompletionEvent(this, TaskAttemptStateInternal.KILLED);
eventHandler.handle(new JobHistoryEvent(attemptId.getTaskId().getJobId(), tauce));
}
if (needToClean) {
TaskAttemptContext tac = new TaskAttemptContextImpl(conf, TypeConverter.fromYarn(attemptId));
try {
committer.abortTask(tac);
} catch (Exception e) {
LOG.warn("Task cleanup failed for attempt " + attemptId, e);
}
}
return attemptState;
}
use of org.apache.hadoop.mapreduce.v2.app.job.TaskAttemptStateInternal in project hadoop by apache.
the class TaskAttemptImpl method handle.
@SuppressWarnings("unchecked")
@Override
public void handle(TaskAttemptEvent event) {
if (LOG.isDebugEnabled()) {
LOG.debug("Processing " + event.getTaskAttemptID() + " of type " + event.getType());
}
writeLock.lock();
try {
final TaskAttemptStateInternal oldState = getInternalState();
try {
stateMachine.doTransition(event.getType(), event);
} catch (InvalidStateTransitionException e) {
LOG.error("Can't handle this event at current state for " + this.attemptId, e);
eventHandler.handle(new JobDiagnosticsUpdateEvent(this.attemptId.getTaskId().getJobId(), "Invalid event " + event.getType() + " on TaskAttempt " + this.attemptId));
eventHandler.handle(new JobEvent(this.attemptId.getTaskId().getJobId(), JobEventType.INTERNAL_ERROR));
}
if (oldState != getInternalState()) {
if (getInternalState() == TaskAttemptStateInternal.FAILED) {
String nodeId = null == this.container ? "Not-assigned" : this.container.getNodeId().toString();
LOG.info(attemptId + " transitioned from state " + oldState + " to " + getInternalState() + ", event type is " + event.getType() + " and nodeId=" + nodeId);
} else {
LOG.info(attemptId + " TaskAttempt Transitioned from " + oldState + " to " + getInternalState());
}
}
} finally {
writeLock.unlock();
}
}
use of org.apache.hadoop.mapreduce.v2.app.job.TaskAttemptStateInternal in project hadoop by apache.
the class MRApp method waitForInternalState.
public void waitForInternalState(TaskAttemptImpl attempt, TaskAttemptStateInternal finalState) throws Exception {
int timeoutSecs = 0;
TaskAttemptReport report = attempt.getReport();
TaskAttemptStateInternal iState = attempt.getInternalState();
while (!finalState.equals(iState) && timeoutSecs++ < 20) {
System.out.println("TaskAttempt Internal State is : " + iState + " Waiting for Internal state : " + finalState + " progress : " + report.getProgress());
Thread.sleep(500);
report = attempt.getReport();
iState = attempt.getInternalState();
}
System.out.println("TaskAttempt Internal State is : " + iState);
Assert.assertEquals("TaskAttempt Internal state is not correct (timedout)", finalState, iState);
}
Aggregations