use of org.apache.hadoop.mapreduce.v2.api.records.TaskAttemptState in project hadoop by apache.
the class MockJobs method newTaskAttempt.
public static TaskAttempt newTaskAttempt(TaskId tid, int i) {
final TaskAttemptId taid = Records.newRecord(TaskAttemptId.class);
taid.setTaskId(tid);
taid.setId(i);
final TaskAttemptReport report = newTaskAttemptReport(taid);
return new TaskAttempt() {
@Override
public NodeId getNodeId() throws UnsupportedOperationException {
throw new UnsupportedOperationException();
}
@Override
public TaskAttemptId getID() {
return taid;
}
@Override
public TaskAttemptReport getReport() {
return report;
}
@Override
public long getLaunchTime() {
return report.getStartTime();
}
@Override
public long getFinishTime() {
return report.getFinishTime();
}
@Override
public int getShufflePort() {
return ShuffleHandler.DEFAULT_SHUFFLE_PORT;
}
@Override
public Counters getCounters() {
if (report != null && report.getCounters() != null) {
return new Counters(TypeConverter.fromYarn(report.getCounters()));
}
return null;
}
@Override
public float getProgress() {
return report.getProgress();
}
@Override
public Phase getPhase() {
return report.getPhase();
}
@Override
public TaskAttemptState getState() {
return report.getTaskAttemptState();
}
@Override
public boolean isFinished() {
switch(report.getTaskAttemptState()) {
case SUCCEEDED:
case FAILED:
case KILLED:
return true;
}
return false;
}
@Override
public ContainerId getAssignedContainerID() {
ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(taid.getTaskId().getJobId().getAppId(), 0);
ContainerId id = ContainerId.newContainerId(appAttemptId, 0);
return id;
}
@Override
public String getNodeHttpAddress() {
return "localhost:8042";
}
@Override
public List<String> getDiagnostics() {
return Lists.newArrayList(report.getDiagnosticInfo());
}
@Override
public String getAssignedContainerMgrAddress() {
return "localhost:9998";
}
@Override
public long getShuffleFinishTime() {
return report.getShuffleFinishTime();
}
@Override
public long getSortFinishTime() {
return report.getSortFinishTime();
}
@Override
public String getNodeRackName() {
return "/default-rack";
}
};
}
use of org.apache.hadoop.mapreduce.v2.api.records.TaskAttemptState in project hadoop by apache.
the class TestRecovery method testRecoverySuccessAttempt.
@Test
public void testRecoverySuccessAttempt() {
LOG.info("--- START: testRecoverySuccessAttempt ---");
long clusterTimestamp = System.currentTimeMillis();
EventHandler mockEventHandler = mock(EventHandler.class);
MapTaskImpl recoverMapTask = getMockMapTask(clusterTimestamp, mockEventHandler);
TaskId taskId = recoverMapTask.getID();
JobID jobID = new JobID(Long.toString(clusterTimestamp), 1);
TaskID taskID = new TaskID(jobID, org.apache.hadoop.mapreduce.TaskType.MAP, taskId.getId());
//Mock up the TaskAttempts
Map<TaskAttemptID, TaskAttemptInfo> mockTaskAttempts = new HashMap<TaskAttemptID, TaskAttemptInfo>();
TaskAttemptID taId1 = new TaskAttemptID(taskID, 2);
TaskAttemptInfo mockTAinfo1 = getMockTaskAttemptInfo(taId1, TaskAttemptState.SUCCEEDED);
mockTaskAttempts.put(taId1, mockTAinfo1);
TaskAttemptID taId2 = new TaskAttemptID(taskID, 1);
TaskAttemptInfo mockTAinfo2 = getMockTaskAttemptInfo(taId2, TaskAttemptState.FAILED);
mockTaskAttempts.put(taId2, mockTAinfo2);
OutputCommitter mockCommitter = mock(OutputCommitter.class);
TaskInfo mockTaskInfo = mock(TaskInfo.class);
when(mockTaskInfo.getTaskStatus()).thenReturn("SUCCEEDED");
when(mockTaskInfo.getTaskId()).thenReturn(taskID);
when(mockTaskInfo.getAllTaskAttempts()).thenReturn(mockTaskAttempts);
recoverMapTask.handle(new TaskRecoverEvent(taskId, mockTaskInfo, mockCommitter, true));
ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
verify(mockEventHandler, atLeast(1)).handle((org.apache.hadoop.yarn.event.Event) arg.capture());
Map<TaskAttemptID, TaskAttemptState> finalAttemptStates = new HashMap<TaskAttemptID, TaskAttemptState>();
finalAttemptStates.put(taId1, TaskAttemptState.SUCCEEDED);
finalAttemptStates.put(taId2, TaskAttemptState.FAILED);
List<EventType> jobHistoryEvents = new ArrayList<EventType>();
jobHistoryEvents.add(EventType.TASK_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_FINISHED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_FAILED);
jobHistoryEvents.add(EventType.TASK_FINISHED);
recoveryChecker(recoverMapTask, TaskState.SUCCEEDED, finalAttemptStates, arg, jobHistoryEvents, 2L, 1L);
}
use of org.apache.hadoop.mapreduce.v2.api.records.TaskAttemptState in project hadoop by apache.
the class TestRecovery method testRecoveryTaskSuccessAllAttemptsFail.
@Test
public void testRecoveryTaskSuccessAllAttemptsFail() {
LOG.info("--- START: testRecoveryTaskSuccessAllAttemptsFail ---");
long clusterTimestamp = System.currentTimeMillis();
EventHandler mockEventHandler = mock(EventHandler.class);
MapTaskImpl recoverMapTask = getMockMapTask(clusterTimestamp, mockEventHandler);
TaskId taskId = recoverMapTask.getID();
JobID jobID = new JobID(Long.toString(clusterTimestamp), 1);
TaskID taskID = new TaskID(jobID, org.apache.hadoop.mapreduce.TaskType.MAP, taskId.getId());
//Mock up the TaskAttempts
Map<TaskAttemptID, TaskAttemptInfo> mockTaskAttempts = new HashMap<TaskAttemptID, TaskAttemptInfo>();
TaskAttemptID taId1 = new TaskAttemptID(taskID, 2);
TaskAttemptInfo mockTAinfo1 = getMockTaskAttemptInfo(taId1, TaskAttemptState.FAILED);
mockTaskAttempts.put(taId1, mockTAinfo1);
TaskAttemptID taId2 = new TaskAttemptID(taskID, 1);
TaskAttemptInfo mockTAinfo2 = getMockTaskAttemptInfo(taId2, TaskAttemptState.FAILED);
mockTaskAttempts.put(taId2, mockTAinfo2);
OutputCommitter mockCommitter = mock(OutputCommitter.class);
TaskInfo mockTaskInfo = mock(TaskInfo.class);
when(mockTaskInfo.getTaskStatus()).thenReturn("SUCCEEDED");
when(mockTaskInfo.getTaskId()).thenReturn(taskID);
when(mockTaskInfo.getAllTaskAttempts()).thenReturn(mockTaskAttempts);
recoverMapTask.handle(new TaskRecoverEvent(taskId, mockTaskInfo, mockCommitter, true));
ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
verify(mockEventHandler, atLeast(1)).handle((org.apache.hadoop.yarn.event.Event) arg.capture());
Map<TaskAttemptID, TaskAttemptState> finalAttemptStates = new HashMap<TaskAttemptID, TaskAttemptState>();
finalAttemptStates.put(taId1, TaskAttemptState.FAILED);
finalAttemptStates.put(taId2, TaskAttemptState.FAILED);
// check for one new attempt launched since successful attempt not found
TaskAttemptID taId3 = new TaskAttemptID(taskID, 2000);
finalAttemptStates.put(taId3, TaskAttemptState.NEW);
List<EventType> jobHistoryEvents = new ArrayList<EventType>();
jobHistoryEvents.add(EventType.TASK_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_FAILED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_FAILED);
recoveryChecker(recoverMapTask, TaskState.RUNNING, finalAttemptStates, arg, jobHistoryEvents, 2L, 2L);
}
use of org.apache.hadoop.mapreduce.v2.api.records.TaskAttemptState in project hadoop by apache.
the class TaskImpl method recover.
/**
* Recover a completed task from a previous application attempt
* @param taskInfo recovered info about the task
* @param recoverTaskOutput whether to recover task outputs
* @return state of the task after recovery
*/
private TaskStateInternal recover(TaskInfo taskInfo, OutputCommitter committer, boolean recoverTaskOutput) {
LOG.info("Recovering task " + taskId + " from prior app attempt, status was " + taskInfo.getTaskStatus());
scheduledTime = taskInfo.getStartTime();
sendTaskStartedEvent();
Collection<TaskAttemptInfo> attemptInfos = taskInfo.getAllTaskAttempts().values();
if (attemptInfos.size() > 0) {
metrics.launchedTask(this);
}
// recover the attempts for this task in the order they finished
// so task attempt completion events are ordered properly
int savedNextAttemptNumber = nextAttemptNumber;
ArrayList<TaskAttemptInfo> taInfos = new ArrayList<TaskAttemptInfo>(taskInfo.getAllTaskAttempts().values());
Collections.sort(taInfos, TA_INFO_COMPARATOR);
for (TaskAttemptInfo taInfo : taInfos) {
nextAttemptNumber = taInfo.getAttemptId().getId();
TaskAttemptImpl attempt = addAttempt(Avataar.VIRGIN);
// handle the recovery inline so attempts complete before task does
attempt.handle(new TaskAttemptRecoverEvent(attempt.getID(), taInfo, committer, recoverTaskOutput));
finishedAttempts.add(attempt.getID());
TaskAttemptCompletionEventStatus taces = null;
TaskAttemptState attemptState = attempt.getState();
switch(attemptState) {
case FAILED:
taces = TaskAttemptCompletionEventStatus.FAILED;
break;
case KILLED:
taces = TaskAttemptCompletionEventStatus.KILLED;
break;
case SUCCEEDED:
taces = TaskAttemptCompletionEventStatus.SUCCEEDED;
break;
default:
throw new IllegalStateException("Unexpected attempt state during recovery: " + attemptState);
}
if (attemptState == TaskAttemptState.FAILED) {
failedAttempts.add(attempt.getID());
if (failedAttempts.size() >= maxAttempts) {
taces = TaskAttemptCompletionEventStatus.TIPFAILED;
}
}
// TODO: this shouldn't be necessary after MAPREDUCE-4330
if (successfulAttempt == null) {
handleTaskAttemptCompletion(attempt.getID(), taces);
if (attemptState == TaskAttemptState.SUCCEEDED) {
successfulAttempt = attempt.getID();
}
}
}
nextAttemptNumber = savedNextAttemptNumber;
TaskStateInternal taskState = TaskStateInternal.valueOf(taskInfo.getTaskStatus());
switch(taskState) {
case SUCCEEDED:
if (successfulAttempt != null) {
sendTaskSucceededEvents();
} else {
LOG.info("Missing successful attempt for task " + taskId + ", recovering as RUNNING");
// there must have been a fetch failure and the retry wasn't complete
taskState = TaskStateInternal.RUNNING;
metrics.runningTask(this);
addAndScheduleAttempt(Avataar.VIRGIN);
}
break;
case FAILED:
case KILLED:
{
if (taskState == TaskStateInternal.KILLED && attemptInfos.size() == 0) {
metrics.endWaitingTask(this);
}
TaskFailedEvent tfe = new TaskFailedEvent(taskInfo.getTaskId(), taskInfo.getFinishTime(), taskInfo.getTaskType(), taskInfo.getError(), taskInfo.getTaskStatus(), taskInfo.getFailedDueToAttemptId(), taskInfo.getCounters());
eventHandler.handle(new JobHistoryEvent(taskId.getJobId(), tfe));
eventHandler.handle(new JobTaskEvent(taskId, getExternalState(taskState)));
break;
}
default:
throw new java.lang.AssertionError("Unexpected recovered task state: " + taskState);
}
return taskState;
}
use of org.apache.hadoop.mapreduce.v2.api.records.TaskAttemptState in project hadoop by apache.
the class TestRecovery method testRecoveryTaskSuccessAllAttemptsSucceed.
@Test
public void testRecoveryTaskSuccessAllAttemptsSucceed() {
LOG.info("--- START: testRecoveryTaskSuccessAllAttemptsFail ---");
long clusterTimestamp = System.currentTimeMillis();
EventHandler mockEventHandler = mock(EventHandler.class);
MapTaskImpl recoverMapTask = getMockMapTask(clusterTimestamp, mockEventHandler);
TaskId taskId = recoverMapTask.getID();
JobID jobID = new JobID(Long.toString(clusterTimestamp), 1);
TaskID taskID = new TaskID(jobID, org.apache.hadoop.mapreduce.TaskType.MAP, taskId.getId());
//Mock up the TaskAttempts
Map<TaskAttemptID, TaskAttemptInfo> mockTaskAttempts = new HashMap<TaskAttemptID, TaskAttemptInfo>();
TaskAttemptID taId1 = new TaskAttemptID(taskID, 2);
TaskAttemptInfo mockTAinfo1 = getMockTaskAttemptInfo(taId1, TaskAttemptState.SUCCEEDED);
mockTaskAttempts.put(taId1, mockTAinfo1);
TaskAttemptID taId2 = new TaskAttemptID(taskID, 1);
TaskAttemptInfo mockTAinfo2 = getMockTaskAttemptInfo(taId2, TaskAttemptState.SUCCEEDED);
mockTaskAttempts.put(taId2, mockTAinfo2);
OutputCommitter mockCommitter = mock(OutputCommitter.class);
TaskInfo mockTaskInfo = mock(TaskInfo.class);
when(mockTaskInfo.getTaskStatus()).thenReturn("SUCCEEDED");
when(mockTaskInfo.getTaskId()).thenReturn(taskID);
when(mockTaskInfo.getAllTaskAttempts()).thenReturn(mockTaskAttempts);
recoverMapTask.handle(new TaskRecoverEvent(taskId, mockTaskInfo, mockCommitter, true));
ArgumentCaptor<Event> arg = ArgumentCaptor.forClass(Event.class);
verify(mockEventHandler, atLeast(1)).handle((org.apache.hadoop.yarn.event.Event) arg.capture());
Map<TaskAttemptID, TaskAttemptState> finalAttemptStates = new HashMap<TaskAttemptID, TaskAttemptState>();
finalAttemptStates.put(taId1, TaskAttemptState.SUCCEEDED);
finalAttemptStates.put(taId2, TaskAttemptState.SUCCEEDED);
List<EventType> jobHistoryEvents = new ArrayList<EventType>();
jobHistoryEvents.add(EventType.TASK_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_FINISHED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_STARTED);
jobHistoryEvents.add(EventType.MAP_ATTEMPT_FINISHED);
jobHistoryEvents.add(EventType.TASK_FINISHED);
recoveryChecker(recoverMapTask, TaskState.SUCCEEDED, finalAttemptStates, arg, jobHistoryEvents, 2L, 0L);
}
Aggregations