use of org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerHeartbeatResponse in project hadoop by apache.
the class ContainerLocalizer method localizeFiles.
protected void localizeFiles(LocalizationProtocol nodemanager, CompletionService<Path> cs, UserGroupInformation ugi) throws IOException, YarnException {
while (true) {
try {
LocalizerStatus status = createStatus();
LocalizerHeartbeatResponse response = nodemanager.heartbeat(status);
switch(response.getLocalizerAction()) {
case LIVE:
List<ResourceLocalizationSpec> newRsrcs = response.getResourceSpecs();
for (ResourceLocalizationSpec newRsrc : newRsrcs) {
if (!pendingResources.containsKey(newRsrc.getResource())) {
pendingResources.put(newRsrc.getResource(), cs.submit(download(new Path(newRsrc.getDestinationDirectory().getFile()), newRsrc.getResource(), ugi)));
}
}
break;
case DIE:
// killall running localizations
for (Future<Path> pending : pendingResources.values()) {
pending.cancel(true);
}
status = createStatus();
// ignore response while dying.
try {
nodemanager.heartbeat(status);
} catch (YarnException e) {
// Cannot do anything about this during death stage, let's just log
// it.
e.printStackTrace(System.out);
LOG.error("Heartbeat failed while dying: ", e);
}
return;
}
cs.poll(1000, TimeUnit.MILLISECONDS);
} catch (InterruptedException e) {
return;
} catch (YarnException e) {
// TODO cleanup
throw e;
}
}
}
use of org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerHeartbeatResponse in project hadoop by apache.
the class TestResourceLocalizationService method testDownloadingResourcesOnContainerKill.
@Test(timeout = 20000)
@SuppressWarnings("unchecked")
public void testDownloadingResourcesOnContainerKill() throws Exception {
List<Path> localDirs = new ArrayList<Path>();
String[] sDirs = new String[1];
localDirs.add(lfs.makeQualified(new Path(basedir, 0 + "")));
sDirs[0] = localDirs.get(0).toString();
conf.setStrings(YarnConfiguration.NM_LOCAL_DIRS, sDirs);
DrainDispatcher dispatcher = new DrainDispatcher();
dispatcher.init(conf);
dispatcher.start();
EventHandler<ApplicationEvent> applicationBus = mock(EventHandler.class);
dispatcher.register(ApplicationEventType.class, applicationBus);
EventHandler<ContainerEvent> containerBus = mock(EventHandler.class);
dispatcher.register(ContainerEventType.class, containerBus);
DummyExecutor exec = new DummyExecutor();
LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
dirsHandler.init(conf);
DeletionService delServiceReal = new DeletionService(exec);
DeletionService delService = spy(delServiceReal);
delService.init(new Configuration());
delService.start();
ResourceLocalizationService rawService = new ResourceLocalizationService(dispatcher, exec, delService, dirsHandler, nmContext);
ResourceLocalizationService spyService = spy(rawService);
doReturn(mockServer).when(spyService).createServer();
doReturn(lfs).when(spyService).getLocalFileContext(isA(Configuration.class));
FsPermission defaultPermission = FsPermission.getDirDefault().applyUMask(lfs.getUMask());
FsPermission nmPermission = ResourceLocalizationService.NM_PRIVATE_PERM.applyUMask(lfs.getUMask());
final Path userDir = new Path(sDirs[0].substring("file:".length()), ContainerLocalizer.USERCACHE);
final Path fileDir = new Path(sDirs[0].substring("file:".length()), ContainerLocalizer.FILECACHE);
final Path sysDir = new Path(sDirs[0].substring("file:".length()), ResourceLocalizationService.NM_PRIVATE_DIR);
final FileStatus fs = new FileStatus(0, true, 1, 0, System.currentTimeMillis(), 0, defaultPermission, "", "", new Path(sDirs[0]));
final FileStatus nmFs = new FileStatus(0, true, 1, 0, System.currentTimeMillis(), 0, nmPermission, "", "", sysDir);
doAnswer(new Answer<FileStatus>() {
@Override
public FileStatus answer(InvocationOnMock invocation) throws Throwable {
Object[] args = invocation.getArguments();
if (args.length > 0) {
if (args[0].equals(userDir) || args[0].equals(fileDir)) {
return fs;
}
}
return nmFs;
}
}).when(spylfs).getFileStatus(isA(Path.class));
try {
spyService.init(conf);
spyService.start();
final Application app = mock(Application.class);
final ApplicationId appId = BuilderUtils.newApplicationId(314159265358979L, 3);
String user = "user0";
when(app.getUser()).thenReturn(user);
when(app.getAppId()).thenReturn(appId);
spyService.handle(new ApplicationLocalizationEvent(LocalizationEventType.INIT_APPLICATION_RESOURCES, app));
ArgumentMatcher<ApplicationEvent> matchesAppInit = new ArgumentMatcher<ApplicationEvent>() {
@Override
public boolean matches(Object o) {
ApplicationEvent evt = (ApplicationEvent) o;
return evt.getType() == ApplicationEventType.APPLICATION_INITED && appId == evt.getApplicationID();
}
};
dispatcher.await();
verify(applicationBus).handle(argThat(matchesAppInit));
// Initialize localizer.
Random r = new Random();
long seed = r.nextLong();
System.out.println("SEED: " + seed);
r.setSeed(seed);
final Container c1 = getMockContainer(appId, 42, "user0");
final Container c2 = getMockContainer(appId, 43, "user0");
FSDataOutputStream out = new FSDataOutputStream(new DataOutputBuffer(), null);
doReturn(out).when(spylfs).createInternal(isA(Path.class), isA(EnumSet.class), isA(FsPermission.class), anyInt(), anyShort(), anyLong(), isA(Progressable.class), isA(ChecksumOpt.class), anyBoolean());
final LocalResource resource1 = getPrivateMockedResource(r);
LocalResource resource2 = null;
do {
resource2 = getPrivateMockedResource(r);
} while (resource2 == null || resource2.equals(resource1));
LocalResource resource3 = null;
do {
resource3 = getPrivateMockedResource(r);
} while (resource3 == null || resource3.equals(resource1) || resource3.equals(resource2));
// Send localization requests for container c1 and c2.
final LocalResourceRequest req1 = new LocalResourceRequest(resource1);
final LocalResourceRequest req2 = new LocalResourceRequest(resource2);
final LocalResourceRequest req3 = new LocalResourceRequest(resource3);
Map<LocalResourceVisibility, Collection<LocalResourceRequest>> rsrcs = new HashMap<LocalResourceVisibility, Collection<LocalResourceRequest>>();
List<LocalResourceRequest> privateResourceList = new ArrayList<LocalResourceRequest>();
privateResourceList.add(req1);
privateResourceList.add(req2);
privateResourceList.add(req3);
rsrcs.put(LocalResourceVisibility.PRIVATE, privateResourceList);
spyService.handle(new ContainerLocalizationRequestEvent(c1, rsrcs));
final LocalResourceRequest req1_1 = new LocalResourceRequest(resource2);
Map<LocalResourceVisibility, Collection<LocalResourceRequest>> rsrcs1 = new HashMap<LocalResourceVisibility, Collection<LocalResourceRequest>>();
List<LocalResourceRequest> privateResourceList1 = new ArrayList<LocalResourceRequest>();
privateResourceList1.add(req1_1);
rsrcs1.put(LocalResourceVisibility.PRIVATE, privateResourceList1);
spyService.handle(new ContainerLocalizationRequestEvent(c2, rsrcs1));
dispatcher.await();
// Wait for localizers of both container c1 and c2 to begin.
exec.waitForLocalizers(2);
LocalizerRunner locC1 = spyService.getLocalizerRunner(c1.getContainerId().toString());
final String containerIdStr = c1.getContainerId().toString();
// Heartbeats from container localizer
LocalResourceStatus rsrc1success = mock(LocalResourceStatus.class);
LocalResourceStatus rsrc2pending = mock(LocalResourceStatus.class);
LocalizerStatus stat = mock(LocalizerStatus.class);
when(stat.getLocalizerId()).thenReturn(containerIdStr);
when(rsrc1success.getResource()).thenReturn(resource1);
when(rsrc2pending.getResource()).thenReturn(resource2);
when(rsrc1success.getLocalSize()).thenReturn(4344L);
URL locPath = getPath("/some/path");
when(rsrc1success.getLocalPath()).thenReturn(locPath);
when(rsrc1success.getStatus()).thenReturn(ResourceStatusType.FETCH_SUCCESS);
when(rsrc2pending.getStatus()).thenReturn(ResourceStatusType.FETCH_PENDING);
when(stat.getResources()).thenReturn(Collections.<LocalResourceStatus>emptyList()).thenReturn(Collections.singletonList(rsrc1success)).thenReturn(Collections.singletonList(rsrc2pending)).thenReturn(Collections.singletonList(rsrc2pending)).thenReturn(Collections.<LocalResourceStatus>emptyList());
// First heartbeat which schedules first resource.
LocalizerHeartbeatResponse response = spyService.heartbeat(stat);
assertEquals(LocalizerAction.LIVE, response.getLocalizerAction());
// Second heartbeat which reports first resource as success.
// Second resource is scheduled.
response = spyService.heartbeat(stat);
assertEquals(LocalizerAction.LIVE, response.getLocalizerAction());
final String locPath1 = response.getResourceSpecs().get(0).getDestinationDirectory().getFile();
// Third heartbeat which reports second resource as pending.
// Third resource is scheduled.
response = spyService.heartbeat(stat);
assertEquals(LocalizerAction.LIVE, response.getLocalizerAction());
final String locPath2 = response.getResourceSpecs().get(0).getDestinationDirectory().getFile();
// Container c1 is killed which leads to cleanup
spyService.handle(new ContainerLocalizationCleanupEvent(c1, rsrcs));
// This heartbeat will indicate to container localizer to die as localizer
// runner has stopped.
response = spyService.heartbeat(stat);
assertEquals(LocalizerAction.DIE, response.getLocalizerAction());
exec.setStopLocalization();
dispatcher.await();
// verify container notification
ArgumentMatcher<ContainerEvent> successContainerLoc = new ArgumentMatcher<ContainerEvent>() {
@Override
public boolean matches(Object o) {
ContainerEvent evt = (ContainerEvent) o;
return evt.getType() == ContainerEventType.RESOURCE_LOCALIZED && c1.getContainerId() == evt.getContainerID();
}
};
// Only one resource gets localized for container c1.
verify(containerBus).handle(argThat(successContainerLoc));
Set<Path> paths = Sets.newHashSet(new Path(locPath1), new Path(locPath1 + "_tmp"), new Path(locPath2), new Path(locPath2 + "_tmp"));
// Wait for localizer runner thread for container c1 to finish.
while (locC1.getState() != Thread.State.TERMINATED) {
Thread.sleep(50);
}
// Verify if downloading resources were submitted for deletion.
verify(delService).delete(eq(user), (Path) eq(null), argThat(new DownloadingPathsMatcher(paths)));
LocalResourcesTracker tracker = spyService.getLocalResourcesTracker(LocalResourceVisibility.PRIVATE, "user0", appId);
// Container c1 was killed but this resource was localized before kill
// hence its not removed despite ref cnt being 0.
LocalizedResource rsrc1 = tracker.getLocalizedResource(req1);
assertNotNull(rsrc1);
assertEquals(rsrc1.getState(), ResourceState.LOCALIZED);
assertEquals(rsrc1.getRefCount(), 0);
// Container c1 was killed but this resource is referenced by container c2
// as well hence its ref cnt is 1.
LocalizedResource rsrc2 = tracker.getLocalizedResource(req2);
assertNotNull(rsrc2);
assertEquals(rsrc2.getState(), ResourceState.DOWNLOADING);
assertEquals(rsrc2.getRefCount(), 1);
// As container c1 was killed and this resource was not referenced by any
// other container, hence its removed.
LocalizedResource rsrc3 = tracker.getLocalizedResource(req3);
assertNull(rsrc3);
} finally {
spyService.stop();
dispatcher.stop();
delService.stop();
}
}
use of org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerHeartbeatResponse in project hadoop by apache.
the class TestResourceLocalizationService method testLocalResourcePath.
@Test(timeout = 10000)
@SuppressWarnings("unchecked")
public void testLocalResourcePath() throws Exception {
// test the local path where application and user cache files will be
// localized.
DrainDispatcher dispatcher1 = null;
try {
dispatcher1 = new DrainDispatcher();
String user = "testuser";
ApplicationId appId = BuilderUtils.newApplicationId(1, 1);
// creating one local directory
List<Path> localDirs = new ArrayList<Path>();
String[] sDirs = new String[1];
for (int i = 0; i < 1; ++i) {
localDirs.add(lfs.makeQualified(new Path(basedir, i + "")));
sDirs[i] = localDirs.get(i).toString();
}
conf.setStrings(YarnConfiguration.NM_LOCAL_DIRS, sDirs);
LocalDirsHandlerService localDirHandler = new LocalDirsHandlerService();
localDirHandler.init(conf);
// Registering event handlers
EventHandler<ApplicationEvent> applicationBus = mock(EventHandler.class);
dispatcher1.register(ApplicationEventType.class, applicationBus);
EventHandler<ContainerEvent> containerBus = mock(EventHandler.class);
dispatcher1.register(ContainerEventType.class, containerBus);
ContainerExecutor exec = mock(ContainerExecutor.class);
DeletionService delService = mock(DeletionService.class);
LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
// initializing directory handler.
dirsHandler.init(conf);
dispatcher1.init(conf);
dispatcher1.start();
ResourceLocalizationService rls = new ResourceLocalizationService(dispatcher1, exec, delService, localDirHandler, nmContext);
dispatcher1.register(LocalizationEventType.class, rls);
rls.init(conf);
rls.handle(createApplicationLocalizationEvent(user, appId));
// We need to pre-populate the LocalizerRunner as the
// Resource Localization Service code internally starts them which
// definitely we don't want.
// creating new container and populating corresponding localizer runner
// Container - 1
Container container1 = createMockContainer(user, 1);
String localizerId1 = container1.getContainerId().toString();
rls.getPrivateLocalizers().put(localizerId1, rls.new LocalizerRunner(new LocalizerContext(user, container1.getContainerId(), null), localizerId1));
// Creating two requests for container
// 1) Private resource
// 2) Application resource
LocalResourceRequest reqPriv = new LocalResourceRequest(new Path("file:///tmp1"), 123L, LocalResourceType.FILE, LocalResourceVisibility.PRIVATE, "");
List<LocalResourceRequest> privList = new ArrayList<LocalResourceRequest>();
privList.add(reqPriv);
LocalResourceRequest reqApp = new LocalResourceRequest(new Path("file:///tmp2"), 123L, LocalResourceType.FILE, LocalResourceVisibility.APPLICATION, "");
List<LocalResourceRequest> appList = new ArrayList<LocalResourceRequest>();
appList.add(reqApp);
Map<LocalResourceVisibility, Collection<LocalResourceRequest>> rsrcs = new HashMap<LocalResourceVisibility, Collection<LocalResourceRequest>>();
rsrcs.put(LocalResourceVisibility.APPLICATION, appList);
rsrcs.put(LocalResourceVisibility.PRIVATE, privList);
dispatcher1.getEventHandler().handle(new ContainerLocalizationRequestEvent(container1, rsrcs));
// Now waiting for resource download to start. Here actual will not start
// Only the resources will be populated into pending list.
Assert.assertTrue(waitForPrivateDownloadToStart(rls, localizerId1, 2, 5000));
// Validating user and application cache paths
String userCachePath = StringUtils.join(Path.SEPARATOR, Arrays.asList(localDirs.get(0).toUri().getRawPath(), ContainerLocalizer.USERCACHE, user, ContainerLocalizer.FILECACHE));
String userAppCachePath = StringUtils.join(Path.SEPARATOR, Arrays.asList(localDirs.get(0).toUri().getRawPath(), ContainerLocalizer.USERCACHE, user, ContainerLocalizer.APPCACHE, appId.toString(), ContainerLocalizer.FILECACHE));
// Now the Application and private resources may come in any order
// for download.
// For User cahce :
// returned destinationPath = user cache path + random number
// For App cache :
// returned destinationPath = user app cache path + random number
int returnedResources = 0;
boolean appRsrc = false, privRsrc = false;
while (returnedResources < 2) {
LocalizerHeartbeatResponse response = rls.heartbeat(createLocalizerStatus(localizerId1));
for (ResourceLocalizationSpec resourceSpec : response.getResourceSpecs()) {
returnedResources++;
Path destinationDirectory = new Path(resourceSpec.getDestinationDirectory().getFile());
if (resourceSpec.getResource().getVisibility() == LocalResourceVisibility.APPLICATION) {
appRsrc = true;
Assert.assertEquals(userAppCachePath, destinationDirectory.getParent().toUri().toString());
} else if (resourceSpec.getResource().getVisibility() == LocalResourceVisibility.PRIVATE) {
privRsrc = true;
Assert.assertEquals(userCachePath, destinationDirectory.getParent().toUri().toString());
} else {
throw new Exception("Unexpected resource received.");
}
}
}
// We should receive both the resources (Application and Private)
Assert.assertTrue(appRsrc && privRsrc);
} finally {
if (dispatcher1 != null) {
dispatcher1.stop();
}
}
}
use of org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerHeartbeatResponse in project hadoop by apache.
the class TestPBLocalizerRPC method dieHBResponse.
static LocalizerHeartbeatResponse dieHBResponse() {
LocalizerHeartbeatResponse response = recordFactory.newRecordInstance(LocalizerHeartbeatResponse.class);
response.setLocalizerAction(LocalizerAction.DIE);
return response;
}
use of org.apache.hadoop.yarn.server.nodemanager.api.protocolrecords.LocalizerHeartbeatResponse in project hadoop by apache.
the class TestPBRecordImpl method testLocalizerHeartbeatResponseSerDe.
@Test(timeout = 10000)
public void testLocalizerHeartbeatResponseSerDe() throws Exception {
LocalizerHeartbeatResponse rsrcS = createLocalizerHeartbeatResponse();
assertTrue(rsrcS instanceof LocalizerHeartbeatResponsePBImpl);
LocalizerHeartbeatResponsePBImpl rsrcPb = (LocalizerHeartbeatResponsePBImpl) rsrcS;
DataOutputBuffer out = new DataOutputBuffer();
rsrcPb.getProto().writeDelimitedTo(out);
DataInputBuffer in = new DataInputBuffer();
in.reset(out.getData(), 0, out.getLength());
LocalizerHeartbeatResponseProto rsrcPbD = LocalizerHeartbeatResponseProto.parseDelimitedFrom(in);
assertNotNull(rsrcPbD);
LocalizerHeartbeatResponse rsrcD = new LocalizerHeartbeatResponsePBImpl(rsrcPbD);
assertEquals(rsrcS, rsrcD);
assertEquals(createResource(), rsrcS.getResourceSpecs().get(0).getResource());
assertEquals(createResource(), rsrcD.getResourceSpecs().get(0).getResource());
}
Aggregations