private String getJobSummary(FileContext fc, Path path) throws IOException {
Path qPath = fc.makeQualified(path);
FSDataInputStream in = fc.open(qPath);
String jobSummaryString = in.readUTF();
in.close();
return jobSummaryString;
}
public static Path getPreviousJobHistoryPath(
Configuration conf, ApplicationAttemptId applicationAttemptId) throws IOException {
String jobId = TypeConverter.fromYarn(applicationAttemptId.getApplicationId()).toString();
String jobhistoryDir = JobHistoryUtils.getConfiguredHistoryStagingDirPrefix(conf, jobId);
Path histDirPath = FileContext.getFileContext(conf).makeQualified(new Path(jobhistoryDir));
FileContext fc = FileContext.getFileContext(histDirPath.toUri(), conf);
return fc.makeQualified(
JobHistoryUtils.getStagingJobHistoryFile(
histDirPath, jobId, (applicationAttemptId.getAttemptId() - 1)));
}
private static List<FileStatus> scanDirectory(Path path, FileContext fc, PathFilter pathFilter)
throws IOException {
path = fc.makeQualified(path);
List<FileStatus> jhStatusList = new ArrayList<FileStatus>();
RemoteIterator<FileStatus> fileStatusIter = fc.listStatus(path);
while (fileStatusIter.hasNext()) {
FileStatus fileStatus = fileStatusIter.next();
Path filePath = fileStatus.getPath();
if (fileStatus.isFile() && pathFilter.accept(filePath)) {
jhStatusList.add(fileStatus);
}
}
return jhStatusList;
}
/**
* Clean up older history files.
*
* @throws IOException on any error trying to remove the entries.
*/
@SuppressWarnings("unchecked")
void clean() throws IOException {
// TODO this should be replaced by something that knows about the directory
// structure and will put less of a load on HDFS.
long cutoff = System.currentTimeMillis() - maxHistoryAge;
boolean halted = false;
// TODO Delete YYYY/MM/DD directories.
List<FileStatus> serialDirList = findTimestampedDirectories();
// Sort in ascending order. Relies on YYYY/MM/DD/Serial
Collections.sort(serialDirList);
for (FileStatus serialDir : serialDirList) {
List<FileStatus> historyFileList =
scanDirectoryForHistoryFiles(serialDir.getPath(), doneDirFc);
for (FileStatus historyFile : historyFileList) {
JobIndexInfo jobIndexInfo =
FileNameIndexUtils.getIndexInfo(historyFile.getPath().getName());
long effectiveTimestamp = getEffectiveTimestamp(jobIndexInfo.getFinishTime(), historyFile);
if (effectiveTimestamp <= cutoff) {
HistoryFileInfo fileInfo = this.jobListCache.get(jobIndexInfo.getJobId());
if (fileInfo == null) {
String confFileName =
JobHistoryUtils.getIntermediateConfFileName(jobIndexInfo.getJobId());
fileInfo =
new HistoryFileInfo(
historyFile.getPath(),
new Path(historyFile.getPath().getParent(), confFileName),
null,
jobIndexInfo,
true);
}
deleteJobFromDone(fileInfo);
} else {
halted = true;
break;
}
}
if (!halted) {
doneDirFc.delete(doneDirFc.makeQualified(serialDir.getPath()), true);
removeDirectoryFromSerialNumberIndex(serialDir.getPath());
existingDoneSubdirs.remove(serialDir.getPath());
} else {
break; // Don't scan any more directories.
}
}
}
@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);
}
}
}
}
protected boolean deleteDir(FileStatus serialDir)
throws AccessControlException, FileNotFoundException, UnsupportedFileSystemException,
IOException {
return doneDirFc.delete(doneDirFc.makeQualified(serialDir.getPath()), true);
}
@Override
public void serviceInit(Configuration conf) throws Exception {
conf.set(MRConfig.FRAMEWORK_NAME, MRConfig.YARN_TEZ_FRAMEWORK_NAME);
// blacklisting disabled to prevent scheduling issues
conf.setBoolean(TezConfiguration.TEZ_AM_NODE_BLACKLISTING_ENABLED, false);
if (conf.get(MRJobConfig.MR_AM_STAGING_DIR) == null) {
conf.set(
MRJobConfig.MR_AM_STAGING_DIR,
new File(getTestWorkDir(), "apps_staging_dir" + Path.SEPARATOR).getAbsolutePath());
}
if (conf.get(YarnConfiguration.DEBUG_NM_DELETE_DELAY_SEC) == null) {
// nothing defined. set quick delete value
conf.setLong(YarnConfiguration.DEBUG_NM_DELETE_DELAY_SEC, 0l);
}
File appJarLocalFile = new File(MiniTezCluster.APPJAR);
if (!appJarLocalFile.exists()) {
String message = "TezAppJar " + MiniTezCluster.APPJAR + " not found. Exiting.";
LOG.info(message);
throw new TezUncheckedException(message);
}
FileSystem fs = FileSystem.get(conf);
Path testRootDir = fs.makeQualified(new Path("target", getName() + "-tmpDir"));
Path appRemoteJar = new Path(testRootDir, "TezAppJar.jar");
// Copy AppJar and make it public.
Path appMasterJar = new Path(MiniTezCluster.APPJAR);
fs.copyFromLocalFile(appMasterJar, appRemoteJar);
fs.setPermission(appRemoteJar, new FsPermission("777"));
conf.set(TezConfiguration.TEZ_LIB_URIS, appRemoteJar.toUri().toString());
LOG.info("Set TEZ-LIB-URI to: " + conf.get(TezConfiguration.TEZ_LIB_URIS));
// VMEM monitoring disabled, PMEM monitoring enabled.
conf.setBoolean(YarnConfiguration.NM_PMEM_CHECK_ENABLED, false);
conf.setBoolean(YarnConfiguration.NM_VMEM_CHECK_ENABLED, false);
conf.set(CommonConfigurationKeys.FS_PERMISSIONS_UMASK_KEY, "000");
try {
Path stagingPath =
FileContext.getFileContext(conf)
.makeQualified(new Path(conf.get(MRJobConfig.MR_AM_STAGING_DIR)));
/*
* Re-configure the staging path on Windows if the file system is localFs.
* We need to use a absolute path that contains the drive letter. The unit
* test could run on a different drive than the AM. We can run into the
* issue that job files are localized to the drive where the test runs on,
* while the AM starts on a different drive and fails to find the job
* metafiles. Using absolute path can avoid this ambiguity.
*/
if (Path.WINDOWS) {
if (LocalFileSystem.class.isInstance(stagingPath.getFileSystem(conf))) {
conf.set(
MRJobConfig.MR_AM_STAGING_DIR,
new File(conf.get(MRJobConfig.MR_AM_STAGING_DIR)).getAbsolutePath());
}
}
FileContext fc = FileContext.getFileContext(stagingPath.toUri(), conf);
if (fc.util().exists(stagingPath)) {
LOG.info(stagingPath + " exists! deleting...");
fc.delete(stagingPath, true);
}
LOG.info("mkdir: " + stagingPath);
fc.mkdir(stagingPath, null, true);
// mkdir done directory as well
String doneDir = JobHistoryUtils.getConfiguredHistoryServerDoneDirPrefix(conf);
Path doneDirPath = fc.makeQualified(new Path(doneDir));
fc.mkdir(doneDirPath, null, true);
} catch (IOException e) {
throw new TezUncheckedException("Could not create staging directory. ", e);
}
conf.set(MRConfig.MASTER_ADDRESS, "test");
// configure the shuffle service in NM
conf.setStrings(
YarnConfiguration.NM_AUX_SERVICES,
new String[] {ShuffleHandler.MAPREDUCE_SHUFFLE_SERVICEID});
conf.setClass(
String.format(
YarnConfiguration.NM_AUX_SERVICE_FMT, ShuffleHandler.MAPREDUCE_SHUFFLE_SERVICEID),
ShuffleHandler.class,
Service.class);
// Non-standard shuffle port
conf.setInt(ShuffleHandler.SHUFFLE_PORT_CONFIG_KEY, 0);
conf.setClass(
YarnConfiguration.NM_CONTAINER_EXECUTOR,
DefaultContainerExecutor.class,
ContainerExecutor.class);
// TestMRJobs is for testing non-uberized operation only; see TestUberAM
// for corresponding uberized tests.
conf.setBoolean(MRJobConfig.JOB_UBERTASK_ENABLE, false);
super.serviceInit(conf);
}
private Path getFullyQualifiedPath(String pathString) {
return useFCOption
? mfc.makeQualified(new Path(ROOT_DIR, pathString))
: mfs.makeQualified(new Path(ROOT_DIR, pathString));
}