use of org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator in project hadoop by apache.
the class TestStagingCleanup method testDeletionofStaging.
@Test
public void testDeletionofStaging() throws IOException {
conf.set(MRJobConfig.MAPREDUCE_JOB_DIR, stagingJobDir);
fs = mock(FileSystem.class);
when(fs.delete(any(Path.class), anyBoolean())).thenReturn(true);
//Staging Dir exists
String user = UserGroupInformation.getCurrentUser().getShortUserName();
Path stagingDir = MRApps.getStagingAreaDir(conf, user);
when(fs.exists(stagingDir)).thenReturn(true);
ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 0);
ApplicationAttemptId attemptId = ApplicationAttemptId.newInstance(appId, 1);
JobId jobid = recordFactory.newRecordInstance(JobId.class);
jobid.setAppId(appId);
ContainerAllocator mockAlloc = mock(ContainerAllocator.class);
Assert.assertTrue(MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS > 1);
MRAppMaster appMaster = new TestMRApp(attemptId, mockAlloc, JobStateInternal.RUNNING, MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS);
appMaster.init(conf);
appMaster.start();
appMaster.shutDownJob();
//test whether notifyIsLastAMRetry called
Assert.assertEquals(true, ((TestMRApp) appMaster).getTestIsLastAMRetry());
verify(fs).delete(stagingJobPath, true);
}
use of org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator in project hadoop by apache.
the class TestLocalContainerAllocator method testAllocatedContainerResourceIsNotNull.
@Test
public void testAllocatedContainerResourceIsNotNull() {
ArgumentCaptor<TaskAttemptContainerAssignedEvent> containerAssignedCaptor = ArgumentCaptor.forClass(TaskAttemptContainerAssignedEvent.class);
@SuppressWarnings("unchecked") EventHandler<Event> eventHandler = mock(EventHandler.class);
AppContext context = mock(AppContext.class);
when(context.getEventHandler()).thenReturn(eventHandler);
ContainerId containerId = ContainerId.fromString("container_1427562107907_0002_01_000001");
LocalContainerAllocator containerAllocator = new LocalContainerAllocator(mock(ClientService.class), context, "localhost", -1, -1, containerId);
ContainerAllocatorEvent containerAllocatorEvent = createContainerRequestEvent();
containerAllocator.handle(containerAllocatorEvent);
verify(eventHandler, times(1)).handle(containerAssignedCaptor.capture());
Container container = containerAssignedCaptor.getValue().getContainer();
Resource containerResource = container.getResource();
Assert.assertNotNull(containerResource);
Assert.assertEquals(containerResource.getMemorySize(), 0);
Assert.assertEquals(containerResource.getVirtualCores(), 0);
}
use of org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator in project hadoop by apache.
the class TestRMContainerAllocator method testUnregistrationOnlyIfRegistered.
@Test
public void testUnregistrationOnlyIfRegistered() throws Exception {
Configuration conf = new Configuration();
final MyResourceManager rm = new MyResourceManager(conf);
rm.start();
DrainDispatcher rmDispatcher = (DrainDispatcher) rm.getRMContext().getDispatcher();
// Submit the application
RMApp rmApp = rm.submitApp(1024);
rmDispatcher.await();
MockNM amNodeManager = rm.registerNode("127.0.0.1:1234", 11264);
amNodeManager.nodeHeartbeat(true);
rmDispatcher.await();
final ApplicationAttemptId appAttemptId = rmApp.getCurrentAppAttempt().getAppAttemptId();
rm.sendAMLaunched(appAttemptId);
rmDispatcher.await();
MRApp mrApp = new MRApp(appAttemptId, ContainerId.newContainerId(appAttemptId, 0), 10, 0, false, this.getClass().getName(), true, 1) {
@Override
protected Dispatcher createDispatcher() {
return new DrainDispatcher();
}
protected ContainerAllocator createContainerAllocator(ClientService clientService, AppContext context) {
return new MyContainerAllocator(rm, appAttemptId, context);
}
;
};
mrApp.submit(conf);
DrainDispatcher amDispatcher = (DrainDispatcher) mrApp.getDispatcher();
MyContainerAllocator allocator = (MyContainerAllocator) mrApp.getContainerAllocator();
amDispatcher.await();
Assert.assertTrue(allocator.isApplicationMasterRegistered());
mrApp.stop();
Assert.assertTrue(allocator.isUnregistered());
}
use of org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator in project hadoop by apache.
the class TestStagingCleanup method testPreservePatternMatchedStaging.
@Test
public void testPreservePatternMatchedStaging() throws IOException {
conf.set(MRJobConfig.MAPREDUCE_JOB_DIR, stagingJobDir);
// The staging files that are matched to the pattern
// should not be deleted
conf.set(MRJobConfig.PRESERVE_FILES_PATTERN, "JobDir");
fs = mock(FileSystem.class);
when(fs.delete(any(Path.class), anyBoolean())).thenReturn(true);
//Staging Dir exists
String user = UserGroupInformation.getCurrentUser().getShortUserName();
Path stagingDir = MRApps.getStagingAreaDir(conf, user);
when(fs.exists(stagingDir)).thenReturn(true);
ApplicationId appId = ApplicationId.newInstance(System.currentTimeMillis(), 0);
ApplicationAttemptId attemptId = ApplicationAttemptId.newInstance(appId, 1);
JobId jobid = recordFactory.newRecordInstance(JobId.class);
jobid.setAppId(appId);
ContainerAllocator mockAlloc = mock(ContainerAllocator.class);
Assert.assertTrue(MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS > 1);
MRAppMaster appMaster = new TestMRApp(attemptId, mockAlloc, JobStateInternal.RUNNING, MRJobConfig.DEFAULT_MR_AM_MAX_ATTEMPTS);
appMaster.init(conf);
appMaster.start();
appMaster.shutDownJob();
//test whether notifyIsLastAMRetry called
Assert.assertEquals(true, ((TestMRApp) appMaster).getTestIsLastAMRetry());
verify(fs, times(0)).delete(stagingJobPath, true);
}
use of org.apache.hadoop.mapreduce.v2.app.rm.ContainerAllocator in project hadoop by apache.
the class MRAppMaster method serviceInit.
@Override
protected void serviceInit(final Configuration conf) throws Exception {
// create the job classloader if enabled
createJobClassLoader(conf);
conf.setBoolean(Dispatcher.DISPATCHER_EXIT_ON_ERROR_KEY, true);
initJobCredentialsAndUGI(conf);
dispatcher = createDispatcher();
addIfService(dispatcher);
taskAttemptFinishingMonitor = createTaskAttemptFinishingMonitor(dispatcher.getEventHandler());
addIfService(taskAttemptFinishingMonitor);
context = new RunningAppContext(conf, taskAttemptFinishingMonitor);
// Job name is the same as the app name util we support DAG of jobs
// for an app later
appName = conf.get(MRJobConfig.JOB_NAME, "<missing app name>");
conf.setInt(MRJobConfig.APPLICATION_ATTEMPT_ID, appAttemptID.getAttemptId());
newApiCommitter = false;
jobId = MRBuilderUtils.newJobId(appAttemptID.getApplicationId(), appAttemptID.getApplicationId().getId());
int numReduceTasks = conf.getInt(MRJobConfig.NUM_REDUCES, 0);
if ((numReduceTasks > 0 && conf.getBoolean("mapred.reducer.new-api", false)) || (numReduceTasks == 0 && conf.getBoolean("mapred.mapper.new-api", false))) {
newApiCommitter = true;
LOG.info("Using mapred newApiCommitter.");
}
boolean copyHistory = false;
committer = createOutputCommitter(conf);
try {
String user = UserGroupInformation.getCurrentUser().getShortUserName();
Path stagingDir = MRApps.getStagingAreaDir(conf, user);
FileSystem fs = getFileSystem(conf);
boolean stagingExists = fs.exists(stagingDir);
Path startCommitFile = MRApps.getStartJobCommitFile(conf, user, jobId);
boolean commitStarted = fs.exists(startCommitFile);
Path endCommitSuccessFile = MRApps.getEndJobCommitSuccessFile(conf, user, jobId);
boolean commitSuccess = fs.exists(endCommitSuccessFile);
Path endCommitFailureFile = MRApps.getEndJobCommitFailureFile(conf, user, jobId);
boolean commitFailure = fs.exists(endCommitFailureFile);
if (!stagingExists) {
isLastAMRetry = true;
LOG.info("Attempt num: " + appAttemptID.getAttemptId() + " is last retry: " + isLastAMRetry + " because the staging dir doesn't exist.");
errorHappenedShutDown = true;
forcedState = JobStateInternal.ERROR;
shutDownMessage = "Staging dir does not exist " + stagingDir;
LOG.fatal(shutDownMessage);
} else if (commitStarted) {
//A commit was started so this is the last time, we just need to know
// what result we will use to notify, and how we will unregister
errorHappenedShutDown = true;
isLastAMRetry = true;
LOG.info("Attempt num: " + appAttemptID.getAttemptId() + " is last retry: " + isLastAMRetry + " because a commit was started.");
copyHistory = true;
if (commitSuccess) {
shutDownMessage = "Job commit succeeded in a prior MRAppMaster attempt " + "before it crashed. Recovering.";
forcedState = JobStateInternal.SUCCEEDED;
} else if (commitFailure) {
shutDownMessage = "Job commit failed in a prior MRAppMaster attempt " + "before it crashed. Not retrying.";
forcedState = JobStateInternal.FAILED;
} else {
if (isCommitJobRepeatable()) {
// cleanup previous half done commits if committer supports
// repeatable job commit.
errorHappenedShutDown = false;
cleanupInterruptedCommit(conf, fs, startCommitFile);
} else {
//The commit is still pending, commit error
shutDownMessage = "Job commit from a prior MRAppMaster attempt is " + "potentially in progress. Preventing multiple commit executions";
forcedState = JobStateInternal.ERROR;
}
}
}
} catch (IOException e) {
throw new YarnRuntimeException("Error while initializing", e);
}
if (errorHappenedShutDown) {
NoopEventHandler eater = new NoopEventHandler();
//We do not have a JobEventDispatcher in this path
dispatcher.register(JobEventType.class, eater);
EventHandler<JobHistoryEvent> historyService = null;
if (copyHistory) {
historyService = createJobHistoryHandler(context);
dispatcher.register(org.apache.hadoop.mapreduce.jobhistory.EventType.class, historyService);
} else {
dispatcher.register(org.apache.hadoop.mapreduce.jobhistory.EventType.class, eater);
}
if (copyHistory) {
// Now that there's a FINISHING state for application on RM to give AMs
// plenty of time to clean up after unregister it's safe to clean staging
// directory after unregistering with RM. So, we start the staging-dir
// cleaner BEFORE the ContainerAllocator so that on shut-down,
// ContainerAllocator unregisters first and then the staging-dir cleaner
// deletes staging directory.
addService(createStagingDirCleaningService());
}
// service to allocate containers from RM (if non-uber) or to fake it (uber)
containerAllocator = createContainerAllocator(null, context);
addIfService(containerAllocator);
dispatcher.register(ContainerAllocator.EventType.class, containerAllocator);
if (copyHistory) {
// Add the JobHistoryEventHandler last so that it is properly stopped first.
// This will guarantee that all history-events are flushed before AM goes
// ahead with shutdown.
// Note: Even though JobHistoryEventHandler is started last, if any
// component creates a JobHistoryEvent in the meanwhile, it will be just be
// queued inside the JobHistoryEventHandler
addIfService(historyService);
JobHistoryCopyService cpHist = new JobHistoryCopyService(appAttemptID, dispatcher.getEventHandler());
addIfService(cpHist);
}
} else {
//service to handle requests from JobClient
clientService = createClientService(context);
// Init ClientService separately so that we stop it separately, since this
// service needs to wait some time before it stops so clients can know the
// final states
clientService.init(conf);
containerAllocator = createContainerAllocator(clientService, context);
//service to handle the output committer
committerEventHandler = createCommitterEventHandler(context, committer);
addIfService(committerEventHandler);
//policy handling preemption requests from RM
callWithJobClassLoader(conf, new Action<Void>() {
public Void call(Configuration conf) {
preemptionPolicy = createPreemptionPolicy(conf);
preemptionPolicy.init(context);
return null;
}
});
//service to handle requests to TaskUmbilicalProtocol
taskAttemptListener = createTaskAttemptListener(context, preemptionPolicy);
addIfService(taskAttemptListener);
//service to log job history events
EventHandler<JobHistoryEvent> historyService = createJobHistoryHandler(context);
dispatcher.register(org.apache.hadoop.mapreduce.jobhistory.EventType.class, historyService);
this.jobEventDispatcher = new JobEventDispatcher();
//register the event dispatchers
dispatcher.register(JobEventType.class, jobEventDispatcher);
dispatcher.register(TaskEventType.class, new TaskEventDispatcher());
dispatcher.register(TaskAttemptEventType.class, new TaskAttemptEventDispatcher());
dispatcher.register(CommitterEventType.class, committerEventHandler);
if (conf.getBoolean(MRJobConfig.MAP_SPECULATIVE, false) || conf.getBoolean(MRJobConfig.REDUCE_SPECULATIVE, false)) {
//optional service to speculate on task attempts' progress
speculator = createSpeculator(conf, context);
addIfService(speculator);
}
speculatorEventDispatcher = new SpeculatorEventDispatcher(conf);
dispatcher.register(Speculator.EventType.class, speculatorEventDispatcher);
// Now that there's a FINISHING state for application on RM to give AMs
// plenty of time to clean up after unregister it's safe to clean staging
// directory after unregistering with RM. So, we start the staging-dir
// cleaner BEFORE the ContainerAllocator so that on shut-down,
// ContainerAllocator unregisters first and then the staging-dir cleaner
// deletes staging directory.
addService(createStagingDirCleaningService());
// service to allocate containers from RM (if non-uber) or to fake it (uber)
addIfService(containerAllocator);
dispatcher.register(ContainerAllocator.EventType.class, containerAllocator);
// corresponding service to launch allocated containers via NodeManager
containerLauncher = createContainerLauncher(context);
addIfService(containerLauncher);
dispatcher.register(ContainerLauncher.EventType.class, containerLauncher);
// Add the JobHistoryEventHandler last so that it is properly stopped first.
// This will guarantee that all history-events are flushed before AM goes
// ahead with shutdown.
// Note: Even though JobHistoryEventHandler is started last, if any
// component creates a JobHistoryEvent in the meanwhile, it will be just be
// queued inside the JobHistoryEventHandler
addIfService(historyService);
}
super.serviceInit(conf);
}
Aggregations