@Override
public void run() {
if (LOG.isDebugEnabled()) {
LOG.debug(this);
}
boolean error = false;
if (null == user) {
if (baseDirs == null || baseDirs.size() == 0) {
LOG.debug("NM deleting absolute path : " + subDir);
try {
lfs.delete(subDir, true);
} catch (IOException e) {
error = true;
LOG.warn("Failed to delete " + subDir);
}
} else {
for (Path baseDir : baseDirs) {
Path del = subDir == null ? baseDir : new Path(baseDir, subDir);
LOG.debug("NM deleting path : " + del);
try {
lfs.delete(del, true);
} catch (IOException e) {
error = true;
LOG.warn("Failed to delete " + subDir);
}
}
}
} else {
try {
LOG.debug("Deleting path: [" + subDir + "] as user: [" + user + "]");
if (baseDirs == null || baseDirs.size() == 0) {
delService.exec.deleteAsUser(user, subDir, (Path[]) null);
} else {
delService.exec.deleteAsUser(user, subDir, baseDirs.toArray(new Path[0]));
}
} catch (IOException e) {
error = true;
LOG.warn("Failed to delete as user " + user, e);
} catch (InterruptedException e) {
error = true;
LOG.warn("Failed to delete as user " + user, e);
}
}
if (error) {
setSuccess(!error);
}
fileDeletionTaskFinished();
}
@BeforeClass
public static void setup()
throws AccessControlException, FileNotFoundException, UnsupportedFileSystemException,
IOException {
localFS = FileContext.getLocalFSFileContext();
localFS.delete(new Path(localFSDirBase.getAbsolutePath()), true);
localFSDirBase.mkdirs();
// Do not start cluster here
}
/**
* 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 testSuccessfulContainerLaunch()
throws InterruptedException, IOException, YarnException {
FileContext localFS = FileContext.getLocalFSFileContext();
localFS.delete(new Path(localDir.getAbsolutePath()), true);
localFS.delete(new Path(localLogDir.getAbsolutePath()), true);
localFS.delete(new Path(remoteLogDir.getAbsolutePath()), true);
localDir.mkdir();
localLogDir.mkdir();
remoteLogDir.mkdir();
YarnConfiguration conf = new YarnConfiguration();
Context context =
new NMContext(
new NMContainerTokenSecretManager(conf),
new NMTokenSecretManagerInNM(),
null,
null,
new NMNullStateStoreService()) {
@Override
public int getHttpPort() {
return 1234;
}
};
conf.set(YarnConfiguration.NM_LOCAL_DIRS, localDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_LOG_DIRS, localLogDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_REMOTE_APP_LOG_DIR, remoteLogDir.getAbsolutePath());
ContainerExecutor exec = new DefaultContainerExecutor();
exec.setConf(conf);
DeletionService del = new DeletionService(exec);
Dispatcher dispatcher = new AsyncDispatcher();
LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
NodeHealthCheckerService healthChecker =
new NodeHealthCheckerService(NodeManager.getNodeHealthScriptRunner(conf), dirsHandler);
healthChecker.init(conf);
NodeManagerMetrics metrics = NodeManagerMetrics.create();
NodeStatusUpdater nodeStatusUpdater =
new NodeStatusUpdaterImpl(context, dispatcher, healthChecker, metrics) {
@Override
protected ResourceTracker getRMClient() {
return new LocalRMInterface();
};
@Override
protected void stopRMProxy() {
return;
}
@Override
protected void startStatusUpdater() {
return; // Don't start any updating thread.
}
@Override
public long getRMIdentifier() {
return SIMULATED_RM_IDENTIFIER;
}
};
DummyContainerManager containerManager =
new DummyContainerManager(
context,
exec,
del,
nodeStatusUpdater,
metrics,
new ApplicationACLsManager(conf),
dirsHandler);
nodeStatusUpdater.init(conf);
((NMContext) context).setContainerManager(containerManager);
nodeStatusUpdater.start();
containerManager.init(conf);
containerManager.start();
ContainerLaunchContext launchContext =
recordFactory.newRecordInstance(ContainerLaunchContext.class);
ApplicationId applicationId = ApplicationId.newInstance(0, 0);
ApplicationAttemptId applicationAttemptId = ApplicationAttemptId.newInstance(applicationId, 0);
ContainerId cID = ContainerId.newContainerId(applicationAttemptId, 0);
String user = "******";
StartContainerRequest scRequest =
StartContainerRequest.newInstance(
launchContext,
TestContainerManager.createContainerToken(
cID,
SIMULATED_RM_IDENTIFIER,
context.getNodeId(),
user,
context.getContainerTokenSecretManager()));
List<StartContainerRequest> list = new ArrayList<StartContainerRequest>();
list.add(scRequest);
StartContainersRequest allRequests = StartContainersRequest.newInstance(list);
containerManager.startContainers(allRequests);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.RUNNING);
List<ContainerId> containerIds = new ArrayList<ContainerId>();
containerIds.add(cID);
StopContainersRequest stopRequest = StopContainersRequest.newInstance(containerIds);
containerManager.stopContainers(stopRequest);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.COMPLETE);
containerManager.stop();
}
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);
}
@Test
public void testSuccessfulContainerLaunch() throws InterruptedException, IOException {
FileContext localFS = FileContext.getLocalFSFileContext();
localFS.delete(new Path(localDir.getAbsolutePath()), true);
localFS.delete(new Path(localLogDir.getAbsolutePath()), true);
localFS.delete(new Path(remoteLogDir.getAbsolutePath()), true);
localDir.mkdir();
localLogDir.mkdir();
remoteLogDir.mkdir();
Context context = new NMContext();
YarnConfiguration conf = new YarnConfiguration();
conf.set(YarnConfiguration.NM_LOCAL_DIRS, localDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_LOG_DIRS, localLogDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_REMOTE_APP_LOG_DIR, remoteLogDir.getAbsolutePath());
ContainerExecutor exec = new DefaultContainerExecutor();
exec.setConf(conf);
DeletionService del = new DeletionService(exec);
Dispatcher dispatcher = new AsyncDispatcher();
NodeHealthCheckerService healthChecker = new NodeHealthCheckerService();
healthChecker.init(conf);
LocalDirsHandlerService dirsHandler = healthChecker.getDiskHandler();
NodeManagerMetrics metrics = NodeManagerMetrics.create();
ContainerTokenSecretManager containerTokenSecretManager = new ContainerTokenSecretManager();
NodeStatusUpdater nodeStatusUpdater =
new NodeStatusUpdaterImpl(
context, dispatcher, healthChecker, metrics, containerTokenSecretManager) {
@Override
protected ResourceTracker getRMClient() {
return new LocalRMInterface();
};
@Override
protected void startStatusUpdater() {
return; // Don't start any updating thread.
}
};
DummyContainerManager containerManager =
new DummyContainerManager(
context,
exec,
del,
nodeStatusUpdater,
metrics,
containerTokenSecretManager,
new ApplicationACLsManager(conf),
dirsHandler);
containerManager.init(conf);
containerManager.start();
ContainerLaunchContext launchContext =
recordFactory.newRecordInstance(ContainerLaunchContext.class);
ContainerId cID = recordFactory.newRecordInstance(ContainerId.class);
ApplicationId applicationId = recordFactory.newRecordInstance(ApplicationId.class);
applicationId.setClusterTimestamp(0);
applicationId.setId(0);
ApplicationAttemptId applicationAttemptId =
recordFactory.newRecordInstance(ApplicationAttemptId.class);
applicationAttemptId.setApplicationId(applicationId);
applicationAttemptId.setAttemptId(0);
cID.setApplicationAttemptId(applicationAttemptId);
launchContext.setContainerId(cID);
launchContext.setUser("testing");
launchContext.setResource(recordFactory.newRecordInstance(Resource.class));
StartContainerRequest request = recordFactory.newRecordInstance(StartContainerRequest.class);
request.setContainerLaunchContext(launchContext);
containerManager.startContainer(request);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.RUNNING);
StopContainerRequest stopRequest = recordFactory.newRecordInstance(StopContainerRequest.class);
stopRequest.setContainerId(cID);
containerManager.stopContainer(stopRequest);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.COMPLETE);
containerManager.stop();
}
