private void addLeafTask(TaskAttemptToSchedulerEvent event) {
TaskAttempt taskAttempt = event.getTaskAttempt();
List<DataLocation> locations = taskAttempt.getTask().getDataLocations();
for (DataLocation location : locations) {
String host = location.getHost();
leafTaskHosts.add(host);
HostVolumeMapping hostVolumeMapping = leafTaskHostMapping.get(host);
if (hostVolumeMapping == null) {
String rack = RackResolver.resolve(host).getNetworkLocation();
hostVolumeMapping = new HostVolumeMapping(host, rack);
leafTaskHostMapping.put(host, hostVolumeMapping);
}
hostVolumeMapping.addTaskAttempt(location.getVolumeId(), taskAttempt);
if (LOG.isDebugEnabled()) {
LOG.debug("Added attempt req to host " + host);
}
HashSet<TaskAttemptId> list = leafTasksRackMapping.get(hostVolumeMapping.getRack());
if (list == null) {
list = new HashSet<>();
leafTasksRackMapping.put(hostVolumeMapping.getRack(), list);
}
list.add(taskAttempt.getId());
if (LOG.isDebugEnabled()) {
LOG.debug("Added attempt req to rack " + hostVolumeMapping.getRack());
}
}
leafTasks.add(taskAttempt.getId());
}
private Set<String> resolveRacks(List<String> nodes) {
Set<String> racks = new HashSet<String>();
if (nodes != null) {
for (String node : nodes) {
// Ensure node requests are accompanied by requests for
// corresponding rack
String rack = RackResolver.resolve(node).getNetworkLocation();
if (rack == null) {
LOG.warn("Failed to resolve rack for node " + node + ".");
} else {
racks.add(rack);
}
}
}
return racks;
}
@Test
public void testHistoryParsingForFailedAttempts() throws Exception {
LOG.info("STARTING testHistoryParsingForFailedAttempts");
try {
Configuration conf = new Configuration();
conf.setClass(
CommonConfigurationKeysPublic.NET_TOPOLOGY_NODE_SWITCH_MAPPING_IMPL_KEY,
MyResolver.class,
DNSToSwitchMapping.class);
RackResolver.init(conf);
MRApp app =
new MRAppWithHistoryWithFailedAttempt(2, 1, true, this.getClass().getName(), true);
app.submit(conf);
Job job = app.getContext().getAllJobs().values().iterator().next();
JobId jobId = job.getID();
app.waitForState(job, JobState.SUCCEEDED);
// make sure all events are flushed
app.waitForState(Service.STATE.STOPPED);
String jobhistoryDir = JobHistoryUtils.getHistoryIntermediateDoneDirForUser(conf);
JobHistory jobHistory = new JobHistory();
jobHistory.init(conf);
JobIndexInfo jobIndexInfo = jobHistory.getJobFileInfo(jobId).getJobIndexInfo();
String jobhistoryFileName = FileNameIndexUtils.getDoneFileName(jobIndexInfo);
Path historyFilePath = new Path(jobhistoryDir, jobhistoryFileName);
FSDataInputStream in = null;
FileContext fc = null;
try {
fc = FileContext.getFileContext(conf);
in = fc.open(fc.makeQualified(historyFilePath));
} catch (IOException ioe) {
LOG.info("Can not open history file: " + historyFilePath, ioe);
throw (new Exception("Can not open History File"));
}
JobHistoryParser parser = new JobHistoryParser(in);
JobInfo jobInfo = parser.parse();
Exception parseException = parser.getParseException();
Assert.assertNull("Caught an expected exception " + parseException, parseException);
int noOffailedAttempts = 0;
Map<TaskID, TaskInfo> allTasks = jobInfo.getAllTasks();
for (Task task : job.getTasks().values()) {
TaskInfo taskInfo = allTasks.get(TypeConverter.fromYarn(task.getID()));
for (TaskAttempt taskAttempt : task.getAttempts().values()) {
TaskAttemptInfo taskAttemptInfo =
taskInfo.getAllTaskAttempts().get(TypeConverter.fromYarn((taskAttempt.getID())));
// Verify rack-name for all task attempts
Assert.assertEquals("rack-name is incorrect", taskAttemptInfo.getRackname(), RACK_NAME);
if (taskAttemptInfo.getTaskStatus().equals("FAILED")) {
noOffailedAttempts++;
}
}
}
Assert.assertEquals("No of Failed tasks doesn't match.", 2, noOffailedAttempts);
} finally {
LOG.info("FINISHED testHistoryParsingForFailedAttempts");
}
}
private void checkHistoryParsing(
final int numMaps, final int numReduces, final int numSuccessfulMaps) throws Exception {
Configuration conf = new Configuration();
conf.set(MRJobConfig.USER_NAME, System.getProperty("user.name"));
long amStartTimeEst = System.currentTimeMillis();
conf.setClass(
CommonConfigurationKeysPublic.NET_TOPOLOGY_NODE_SWITCH_MAPPING_IMPL_KEY,
MyResolver.class,
DNSToSwitchMapping.class);
RackResolver.init(conf);
MRApp app = new MRAppWithHistory(numMaps, numReduces, true, this.getClass().getName(), true);
app.submit(conf);
Job job = app.getContext().getAllJobs().values().iterator().next();
JobId jobId = job.getID();
LOG.info("JOBID is " + TypeConverter.fromYarn(jobId).toString());
app.waitForState(job, JobState.SUCCEEDED);
// make sure all events are flushed
app.waitForState(Service.STATE.STOPPED);
String jobhistoryDir = JobHistoryUtils.getHistoryIntermediateDoneDirForUser(conf);
FileContext fc = null;
try {
fc = FileContext.getFileContext(conf);
} catch (IOException ioe) {
LOG.info("Can not get FileContext", ioe);
throw (new Exception("Can not get File Context"));
}
if (numMaps == numSuccessfulMaps) {
String summaryFileName = JobHistoryUtils.getIntermediateSummaryFileName(jobId);
Path summaryFile = new Path(jobhistoryDir, summaryFileName);
String jobSummaryString = getJobSummary(fc, summaryFile);
Assert.assertNotNull(jobSummaryString);
Assert.assertTrue(jobSummaryString.contains("resourcesPerMap=100"));
Assert.assertTrue(jobSummaryString.contains("resourcesPerReduce=100"));
Map<String, String> jobSummaryElements = new HashMap<String, String>();
StringTokenizer strToken = new StringTokenizer(jobSummaryString, ",");
while (strToken.hasMoreTokens()) {
String keypair = strToken.nextToken();
jobSummaryElements.put(keypair.split("=")[0], keypair.split("=")[1]);
}
Assert.assertEquals(
"JobId does not match", jobId.toString(), jobSummaryElements.get("jobId"));
Assert.assertEquals("JobName does not match", "test", jobSummaryElements.get("jobName"));
Assert.assertTrue(
"submitTime should not be 0", Long.parseLong(jobSummaryElements.get("submitTime")) != 0);
Assert.assertTrue(
"launchTime should not be 0", Long.parseLong(jobSummaryElements.get("launchTime")) != 0);
Assert.assertTrue(
"firstMapTaskLaunchTime should not be 0",
Long.parseLong(jobSummaryElements.get("firstMapTaskLaunchTime")) != 0);
Assert.assertTrue(
"firstReduceTaskLaunchTime should not be 0",
Long.parseLong(jobSummaryElements.get("firstReduceTaskLaunchTime")) != 0);
Assert.assertTrue(
"finishTime should not be 0", Long.parseLong(jobSummaryElements.get("finishTime")) != 0);
Assert.assertEquals(
"Mismatch in num map slots",
numSuccessfulMaps,
Integer.parseInt(jobSummaryElements.get("numMaps")));
Assert.assertEquals(
"Mismatch in num reduce slots",
numReduces,
Integer.parseInt(jobSummaryElements.get("numReduces")));
Assert.assertEquals(
"User does not match", System.getProperty("user.name"), jobSummaryElements.get("user"));
Assert.assertEquals("Queue does not match", "default", jobSummaryElements.get("queue"));
Assert.assertEquals("Status does not match", "SUCCEEDED", jobSummaryElements.get("status"));
}
JobHistory jobHistory = new JobHistory();
jobHistory.init(conf);
HistoryFileInfo fileInfo = jobHistory.getJobFileInfo(jobId);
JobInfo jobInfo;
long numFinishedMaps;
synchronized (fileInfo) {
Path historyFilePath = fileInfo.getHistoryFile();
FSDataInputStream in = null;
LOG.info("JobHistoryFile is: " + historyFilePath);
try {
in = fc.open(fc.makeQualified(historyFilePath));
} catch (IOException ioe) {
LOG.info("Can not open history file: " + historyFilePath, ioe);
throw (new Exception("Can not open History File"));
}
JobHistoryParser parser = new JobHistoryParser(in);
final EventReader realReader = new EventReader(in);
EventReader reader = Mockito.mock(EventReader.class);
if (numMaps == numSuccessfulMaps) {
reader = realReader;
} else {
final AtomicInteger numFinishedEvents = new AtomicInteger(0); // Hack!
Mockito.when(reader.getNextEvent())
.thenAnswer(
new Answer<HistoryEvent>() {
public HistoryEvent answer(InvocationOnMock invocation) throws IOException {
HistoryEvent event = realReader.getNextEvent();
if (event instanceof TaskFinishedEvent) {
numFinishedEvents.incrementAndGet();
}
if (numFinishedEvents.get() <= numSuccessfulMaps) {
return event;
} else {
throw new IOException("test");
}
}
});
}
jobInfo = parser.parse(reader);
numFinishedMaps = computeFinishedMaps(jobInfo, numMaps, numSuccessfulMaps);
if (numFinishedMaps != numMaps) {
Exception parseException = parser.getParseException();
Assert.assertNotNull("Didn't get expected parse exception", parseException);
}
}
Assert.assertEquals(
"Incorrect username ", System.getProperty("user.name"), jobInfo.getUsername());
Assert.assertEquals("Incorrect jobName ", "test", jobInfo.getJobname());
Assert.assertEquals("Incorrect queuename ", "default", jobInfo.getJobQueueName());
Assert.assertEquals("incorrect conf path", "test", jobInfo.getJobConfPath());
Assert.assertEquals("incorrect finishedMap ", numSuccessfulMaps, numFinishedMaps);
Assert.assertEquals("incorrect finishedReduces ", numReduces, jobInfo.getFinishedReduces());
Assert.assertEquals("incorrect uberized ", job.isUber(), jobInfo.getUberized());
Map<TaskID, TaskInfo> allTasks = jobInfo.getAllTasks();
int totalTasks = allTasks.size();
Assert.assertEquals("total number of tasks is incorrect ", (numMaps + numReduces), totalTasks);
// Verify aminfo
Assert.assertEquals(1, jobInfo.getAMInfos().size());
Assert.assertEquals(MRApp.NM_HOST, jobInfo.getAMInfos().get(0).getNodeManagerHost());
AMInfo amInfo = jobInfo.getAMInfos().get(0);
Assert.assertEquals(MRApp.NM_PORT, amInfo.getNodeManagerPort());
Assert.assertEquals(MRApp.NM_HTTP_PORT, amInfo.getNodeManagerHttpPort());
Assert.assertEquals(1, amInfo.getAppAttemptId().getAttemptId());
Assert.assertEquals(
amInfo.getAppAttemptId(), amInfo.getContainerId().getApplicationAttemptId());
Assert.assertTrue(
amInfo.getStartTime() <= System.currentTimeMillis()
&& amInfo.getStartTime() >= amStartTimeEst);
ContainerId fakeCid = BuilderUtils.newContainerId(-1, -1, -1, -1);
// Assert at taskAttempt level
for (TaskInfo taskInfo : allTasks.values()) {
int taskAttemptCount = taskInfo.getAllTaskAttempts().size();
Assert.assertEquals("total number of task attempts ", 1, taskAttemptCount);
TaskAttemptInfo taInfo = taskInfo.getAllTaskAttempts().values().iterator().next();
Assert.assertNotNull(taInfo.getContainerId());
// Verify the wrong ctor is not being used. Remove after mrv1 is removed.
Assert.assertFalse(taInfo.getContainerId().equals(fakeCid));
}
// Deep compare Job and JobInfo
for (Task task : job.getTasks().values()) {
TaskInfo taskInfo = allTasks.get(TypeConverter.fromYarn(task.getID()));
Assert.assertNotNull("TaskInfo not found", taskInfo);
for (TaskAttempt taskAttempt : task.getAttempts().values()) {
TaskAttemptInfo taskAttemptInfo =
taskInfo.getAllTaskAttempts().get(TypeConverter.fromYarn((taskAttempt.getID())));
Assert.assertNotNull("TaskAttemptInfo not found", taskAttemptInfo);
Assert.assertEquals(
"Incorrect shuffle port for task attempt",
taskAttempt.getShufflePort(),
taskAttemptInfo.getShufflePort());
if (numMaps == numSuccessfulMaps) {
Assert.assertEquals(MRApp.NM_HOST, taskAttemptInfo.getHostname());
Assert.assertEquals(MRApp.NM_PORT, taskAttemptInfo.getPort());
// Verify rack-name
Assert.assertEquals("rack-name is incorrect", taskAttemptInfo.getRackname(), RACK_NAME);
}
}
}
}
@Override
protected void serviceInit(Configuration conf) throws Exception {
RackResolver.init(conf);
super.serviceInit(conf);
}
private TaskAttemptId allocateRackTask(String host) {
List<HostVolumeMapping> remainingTasks = Lists.newArrayList(leafTaskHostMapping.values());
String rack = RackResolver.resolve(host).getNetworkLocation();
TaskAttemptId attemptId = null;
if (remainingTasks.size() > 0) {
synchronized (scheduledRequests) {
// find largest remaining task of other host in rack
Collections.sort(
remainingTasks,
new Comparator<HostVolumeMapping>() {
@Override
public int compare(HostVolumeMapping v1, HostVolumeMapping v2) {
// descending remaining tasks
if (v2.remainTasksNum.get() > v1.remainTasksNum.get()) {
return 1;
} else if (v2.remainTasksNum.get() == v1.remainTasksNum.get()) {
return 0;
} else {
return -1;
}
}
});
}
for (HostVolumeMapping tasks : remainingTasks) {
for (int i = 0; i < tasks.getRemainingLocalTaskSize(); i++) {
TaskAttemptId tId = tasks.getTaskAttemptIdByRack(rack);
if (tId == null) break;
if (leafTasks.contains(tId)) {
leafTasks.remove(tId);
attemptId = tId;
break;
}
}
if (attemptId != null) break;
}
}
// find task in rack
if (attemptId == null) {
HashSet<TaskAttemptId> list = leafTasksRackMapping.get(rack);
if (list != null) {
synchronized (list) {
Iterator<TaskAttemptId> iterator = list.iterator();
while (iterator.hasNext()) {
TaskAttemptId tId = iterator.next();
iterator.remove();
if (leafTasks.contains(tId)) {
leafTasks.remove(tId);
attemptId = tId;
break;
}
}
}
}
}
return attemptId;
}