use of org.apache.hadoop.mapreduce.v2.api.records.AMInfo in project hadoop by apache.
the class MRAppMaster method readJustAMInfos.
private List<AMInfo> readJustAMInfos() {
List<AMInfo> amInfos = new ArrayList<AMInfo>();
FSDataInputStream inputStream = null;
try {
inputStream = getPreviousJobHistoryStream(getConfig(), appAttemptID);
EventReader jobHistoryEventReader = new EventReader(inputStream);
// All AMInfos are contiguous. Track when the first AMStartedEvent
// appears.
boolean amStartedEventsBegan = false;
HistoryEvent event;
while ((event = jobHistoryEventReader.getNextEvent()) != null) {
if (event.getEventType() == EventType.AM_STARTED) {
if (!amStartedEventsBegan) {
// First AMStartedEvent.
amStartedEventsBegan = true;
}
AMStartedEvent amStartedEvent = (AMStartedEvent) event;
amInfos.add(MRBuilderUtils.newAMInfo(amStartedEvent.getAppAttemptId(), amStartedEvent.getStartTime(), amStartedEvent.getContainerId(), StringInterner.weakIntern(amStartedEvent.getNodeManagerHost()), amStartedEvent.getNodeManagerPort(), amStartedEvent.getNodeManagerHttpPort()));
} else if (amStartedEventsBegan) {
// No need to continue reading all the other events.
break;
}
}
} catch (IOException e) {
LOG.warn("Could not parse the old history file. " + "Will not have old AMinfos ", e);
} finally {
if (inputStream != null) {
IOUtils.closeQuietly(inputStream);
}
}
return amInfos;
}
use of org.apache.hadoop.mapreduce.v2.api.records.AMInfo in project hadoop by apache.
the class MRAppMaster method serviceStart.
@SuppressWarnings("unchecked")
@Override
protected void serviceStart() throws Exception {
amInfos = new LinkedList<AMInfo>();
completedTasksFromPreviousRun = new HashMap<TaskId, TaskInfo>();
processRecovery();
// Current an AMInfo for the current AM generation.
AMInfo amInfo = MRBuilderUtils.newAMInfo(appAttemptID, startTime, containerID, nmHost, nmPort, nmHttpPort);
// /////////////////// Create the job itself.
job = createJob(getConfig(), forcedState, shutDownMessage);
// Send out an MR AM inited event for all previous AMs.
for (AMInfo info : amInfos) {
dispatcher.getEventHandler().handle(new JobHistoryEvent(job.getID(), new AMStartedEvent(info.getAppAttemptId(), info.getStartTime(), info.getContainerId(), info.getNodeManagerHost(), info.getNodeManagerPort(), info.getNodeManagerHttpPort(), appSubmitTime)));
}
// Send out an MR AM inited event for this AM.
dispatcher.getEventHandler().handle(new JobHistoryEvent(job.getID(), new AMStartedEvent(amInfo.getAppAttemptId(), amInfo.getStartTime(), amInfo.getContainerId(), amInfo.getNodeManagerHost(), amInfo.getNodeManagerPort(), amInfo.getNodeManagerHttpPort(), this.forcedState == null ? null : this.forcedState.toString(), appSubmitTime)));
amInfos.add(amInfo);
// metrics system init is really init & start.
// It's more test friendly to put it here.
DefaultMetricsSystem.initialize("MRAppMaster");
boolean initFailed = false;
if (!errorHappenedShutDown) {
// create a job event for job intialization
JobEvent initJobEvent = new JobEvent(job.getID(), JobEventType.JOB_INIT);
// Send init to the job (this does NOT trigger job execution)
// This is a synchronous call, not an event through dispatcher. We want
// job-init to be done completely here.
jobEventDispatcher.handle(initJobEvent);
// If job is still not initialized, an error happened during
// initialization. Must complete starting all of the services so failure
// events can be processed.
initFailed = (((JobImpl) job).getInternalState() != JobStateInternal.INITED);
if (job.isUber()) {
speculatorEventDispatcher.disableSpeculation();
LOG.info("MRAppMaster uberizing job " + job.getID() + " in local container (\"uber-AM\") on node " + nmHost + ":" + nmPort + ".");
} else {
// send init to speculator only for non-uber jobs.
// This won't yet start as dispatcher isn't started yet.
dispatcher.getEventHandler().handle(new SpeculatorEvent(job.getID(), clock.getTime()));
LOG.info("MRAppMaster launching normal, non-uberized, multi-container " + "job " + job.getID() + ".");
}
// Start ClientService here, since it's not initialized if
// errorHappenedShutDown is true
clientService.start();
}
//start all the components
super.serviceStart();
// finally set the job classloader
MRApps.setClassLoader(jobClassLoader, getConfig());
// set job classloader if configured
Limits.init(getConfig());
if (initFailed) {
JobEvent initFailedEvent = new JobEvent(job.getID(), JobEventType.JOB_INIT_FAILED);
jobEventDispatcher.handle(initFailedEvent);
} else {
// All components have started, start the job.
startJobs();
}
}
use of org.apache.hadoop.mapreduce.v2.api.records.AMInfo in project hadoop by apache.
the class TestRecovery method testCrashed.
/**
* AM with 2 maps and 1 reduce. For 1st map, one attempt fails, one attempt
* completely disappears because of failed launch, one attempt gets killed and
* one attempt succeeds. AM crashes after the first tasks finishes and
* recovers completely and succeeds in the second generation.
*
* @throws Exception
*/
@Test
public void testCrashed() throws Exception {
int runCount = 0;
long am1StartTimeEst = System.currentTimeMillis();
MRApp app = new MRAppWithHistory(2, 1, false, this.getClass().getName(), true, ++runCount);
Configuration conf = new Configuration();
conf.setBoolean("mapred.mapper.new-api", true);
conf.setBoolean("mapred.reducer.new-api", true);
conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
Job job = app.submit(conf);
app.waitForState(job, JobState.RUNNING);
long jobStartTime = job.getReport().getStartTime();
//all maps would be running
Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
Iterator<Task> it = job.getTasks().values().iterator();
Task mapTask1 = it.next();
Task mapTask2 = it.next();
Task reduceTask = it.next();
// all maps must be running
app.waitForState(mapTask1, TaskState.RUNNING);
app.waitForState(mapTask2, TaskState.RUNNING);
TaskAttempt task1Attempt1 = mapTask1.getAttempts().values().iterator().next();
TaskAttempt task2Attempt = mapTask2.getAttempts().values().iterator().next();
//before sending the TA_DONE, event make sure attempt has come to
//RUNNING state
app.waitForState(task1Attempt1, TaskAttemptState.RUNNING);
app.waitForState(task2Attempt, TaskAttemptState.RUNNING);
app.waitForState(reduceTask, TaskState.RUNNING);
/////////// Play some games with the TaskAttempts of the first task //////
//send the fail signal to the 1st map task attempt
app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt1.getID(), TaskAttemptEventType.TA_FAILMSG));
app.waitForState(task1Attempt1, TaskAttemptState.FAILED);
int timeOut = 0;
while (mapTask1.getAttempts().size() != 2 && timeOut++ < 10) {
Thread.sleep(2000);
LOG.info("Waiting for next attempt to start");
}
Assert.assertEquals(2, mapTask1.getAttempts().size());
Iterator<TaskAttempt> itr = mapTask1.getAttempts().values().iterator();
itr.next();
TaskAttempt task1Attempt2 = itr.next();
// wait for the second task attempt to be assigned.
waitForContainerAssignment(task1Attempt2);
// This attempt will automatically fail because of the way ContainerLauncher
// is setup
// This attempt 'disappears' from JobHistory and so causes MAPREDUCE-3846
app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt2.getID(), TaskAttemptEventType.TA_CONTAINER_LAUNCH_FAILED));
app.waitForState(task1Attempt2, TaskAttemptState.FAILED);
timeOut = 0;
while (mapTask1.getAttempts().size() != 3 && timeOut++ < 10) {
Thread.sleep(2000);
LOG.info("Waiting for next attempt to start");
}
Assert.assertEquals(3, mapTask1.getAttempts().size());
itr = mapTask1.getAttempts().values().iterator();
itr.next();
itr.next();
TaskAttempt task1Attempt3 = itr.next();
app.waitForState(task1Attempt3, TaskAttemptState.RUNNING);
//send the kill signal to the 1st map 3rd attempt
app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt3.getID(), TaskAttemptEventType.TA_KILL));
app.waitForState(task1Attempt3, TaskAttemptState.KILLED);
timeOut = 0;
while (mapTask1.getAttempts().size() != 4 && timeOut++ < 10) {
Thread.sleep(2000);
LOG.info("Waiting for next attempt to start");
}
Assert.assertEquals(4, mapTask1.getAttempts().size());
itr = mapTask1.getAttempts().values().iterator();
itr.next();
itr.next();
itr.next();
TaskAttempt task1Attempt4 = itr.next();
app.waitForState(task1Attempt4, TaskAttemptState.RUNNING);
//send the done signal to the 1st map 4th attempt
app.getContext().getEventHandler().handle(new TaskAttemptEvent(task1Attempt4.getID(), TaskAttemptEventType.TA_DONE));
/////////// End of games with the TaskAttempts of the first task //////
//wait for first map task to complete
app.waitForState(mapTask1, TaskState.SUCCEEDED);
long task1StartTime = mapTask1.getReport().getStartTime();
long task1FinishTime = mapTask1.getReport().getFinishTime();
//stop the app
app.stop();
//rerun
//in rerun the 1st map will be recovered from previous run
long am2StartTimeEst = System.currentTimeMillis();
app = new MRAppWithHistory(2, 1, false, this.getClass().getName(), false, ++runCount);
conf = new Configuration();
conf.setBoolean(MRJobConfig.MR_AM_JOB_RECOVERY_ENABLE, true);
conf.setBoolean("mapred.mapper.new-api", true);
conf.setBoolean("mapred.reducer.new-api", true);
conf.set(FileOutputFormat.OUTDIR, outputDir.toString());
conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
job = app.submit(conf);
app.waitForState(job, JobState.RUNNING);
//all maps would be running
Assert.assertEquals("No of tasks not correct", 3, job.getTasks().size());
it = job.getTasks().values().iterator();
mapTask1 = it.next();
mapTask2 = it.next();
reduceTask = it.next();
// first map will be recovered, no need to send done
app.waitForState(mapTask1, TaskState.SUCCEEDED);
app.waitForState(mapTask2, TaskState.RUNNING);
task2Attempt = mapTask2.getAttempts().values().iterator().next();
//before sending the TA_DONE, event make sure attempt has come to
//RUNNING state
app.waitForState(task2Attempt, TaskAttemptState.RUNNING);
//send the done signal to the 2nd map task
app.getContext().getEventHandler().handle(new TaskAttemptEvent(mapTask2.getAttempts().values().iterator().next().getID(), TaskAttemptEventType.TA_DONE));
//wait to get it completed
app.waitForState(mapTask2, TaskState.SUCCEEDED);
//wait for reduce to be running before sending done
app.waitForState(reduceTask, TaskState.RUNNING);
//send the done signal to the reduce
app.getContext().getEventHandler().handle(new TaskAttemptEvent(reduceTask.getAttempts().values().iterator().next().getID(), TaskAttemptEventType.TA_DONE));
app.waitForState(job, JobState.SUCCEEDED);
app.verifyCompleted();
Assert.assertEquals("Job Start time not correct", jobStartTime, job.getReport().getStartTime());
Assert.assertEquals("Task Start time not correct", task1StartTime, mapTask1.getReport().getStartTime());
Assert.assertEquals("Task Finish time not correct", task1FinishTime, mapTask1.getReport().getFinishTime());
Assert.assertEquals(2, job.getAMInfos().size());
int attemptNum = 1;
// Verify AMInfo
for (AMInfo amInfo : job.getAMInfos()) {
Assert.assertEquals(attemptNum++, amInfo.getAppAttemptId().getAttemptId());
Assert.assertEquals(amInfo.getAppAttemptId(), amInfo.getContainerId().getApplicationAttemptId());
Assert.assertEquals(MRApp.NM_HOST, amInfo.getNodeManagerHost());
Assert.assertEquals(MRApp.NM_PORT, amInfo.getNodeManagerPort());
Assert.assertEquals(MRApp.NM_HTTP_PORT, amInfo.getNodeManagerHttpPort());
}
long am1StartTimeReal = job.getAMInfos().get(0).getStartTime();
long am2StartTimeReal = job.getAMInfos().get(1).getStartTime();
Assert.assertTrue(am1StartTimeReal >= am1StartTimeEst && am1StartTimeReal <= am2StartTimeEst);
Assert.assertTrue(am2StartTimeReal >= am2StartTimeEst && am2StartTimeReal <= System.currentTimeMillis());
// TODO Add verification of additional data from jobHistory - whatever was
// available in the failed attempt should be available here
}
use of org.apache.hadoop.mapreduce.v2.api.records.AMInfo in project hadoop by apache.
the class TestMRClientService method verifyJobReport.
private void verifyJobReport(JobReport jr) {
Assert.assertNotNull("JobReport is null", jr);
List<AMInfo> amInfos = jr.getAMInfos();
Assert.assertEquals(1, amInfos.size());
Assert.assertEquals(JobState.RUNNING, jr.getJobState());
AMInfo amInfo = amInfos.get(0);
Assert.assertEquals(MRApp.NM_HOST, amInfo.getNodeManagerHost());
Assert.assertEquals(MRApp.NM_PORT, amInfo.getNodeManagerPort());
Assert.assertEquals(MRApp.NM_HTTP_PORT, amInfo.getNodeManagerHttpPort());
Assert.assertEquals(1, amInfo.getAppAttemptId().getAttemptId());
Assert.assertEquals(1, amInfo.getContainerId().getApplicationAttemptId().getAttemptId());
Assert.assertTrue(amInfo.getStartTime() > 0);
Assert.assertEquals(false, jr.isUber());
}
use of org.apache.hadoop.mapreduce.v2.api.records.AMInfo in project hadoop by apache.
the class JobReportPBImpl method addAMInfosToProto.
private synchronized void addAMInfosToProto() {
maybeInitBuilder();
builder.clearAmInfos();
if (this.amInfos == null)
return;
for (AMInfo amInfo : this.amInfos) {
builder.addAmInfos(convertToProtoFormat(amInfo));
}
}
Aggregations