use of org.apache.hadoop.yarn.api.records.NMToken in project hadoop by apache.
the class TestDistributedScheduling method testAMRMClient.
/**
* Validates if AMRMClient can be used with Distributed Scheduling turned on.
*
* @throws Exception
*/
@Test(timeout = 120000)
@SuppressWarnings("unchecked")
public void testAMRMClient() throws Exception {
AMRMClientImpl<AMRMClient.ContainerRequest> amClient = null;
try {
Priority priority = Priority.newInstance(1);
Priority priority2 = Priority.newInstance(2);
Resource capability = Resource.newInstance(1024, 1);
List<NodeReport> nodeReports = rmClient.getNodeReports(NodeState.RUNNING);
String node = nodeReports.get(0).getNodeId().getHost();
String rack = nodeReports.get(0).getRackName();
String[] nodes = new String[] { node };
String[] racks = new String[] { rack };
// start am rm client
amClient = new AMRMClientImpl(client);
amClient.init(yarnConf);
amClient.start();
amClient.registerApplicationMaster(NetUtils.getHostname(), 1024, "");
assertEquals(0, amClient.ask.size());
assertEquals(0, amClient.release.size());
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, null, null, priority2, 0, true, null, ExecutionTypeRequest.newInstance(ExecutionType.OPPORTUNISTIC, true)));
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, null, null, priority2, 0, true, null, ExecutionTypeRequest.newInstance(ExecutionType.OPPORTUNISTIC, true)));
amClient.removeContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.removeContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.removeContainerRequest(new AMRMClient.ContainerRequest(capability, null, null, priority2, 0, true, null, ExecutionTypeRequest.newInstance(ExecutionType.OPPORTUNISTIC, true)));
RemoteRequestsTable<ContainerRequest> remoteRequestsTable = amClient.getTable(0);
int containersRequestedNode = remoteRequestsTable.get(priority, node, ExecutionType.GUARANTEED, capability).remoteRequest.getNumContainers();
int containersRequestedRack = remoteRequestsTable.get(priority, rack, ExecutionType.GUARANTEED, capability).remoteRequest.getNumContainers();
int containersRequestedAny = remoteRequestsTable.get(priority, ResourceRequest.ANY, ExecutionType.GUARANTEED, capability).remoteRequest.getNumContainers();
int oppContainersRequestedAny = remoteRequestsTable.get(priority2, ResourceRequest.ANY, ExecutionType.OPPORTUNISTIC, capability).remoteRequest.getNumContainers();
assertEquals(2, containersRequestedNode);
assertEquals(2, containersRequestedRack);
assertEquals(2, containersRequestedAny);
assertEquals(1, oppContainersRequestedAny);
assertEquals(4, amClient.ask.size());
assertEquals(0, amClient.release.size());
// RM should allocate container within 2 calls to allocate()
int allocatedContainerCount = 0;
int iterationsLeft = 10;
Set<ContainerId> releases = new TreeSet<>();
amClient.getNMTokenCache().clearCache();
Assert.assertEquals(0, amClient.getNMTokenCache().numberOfTokensInCache());
HashMap<String, Token> receivedNMTokens = new HashMap<>();
while (allocatedContainerCount < (containersRequestedAny + oppContainersRequestedAny) && iterationsLeft-- > 0) {
AllocateResponse allocResponse = amClient.allocate(0.1f);
assertEquals(0, amClient.ask.size());
assertEquals(0, amClient.release.size());
allocatedContainerCount += allocResponse.getAllocatedContainers().size();
for (Container container : allocResponse.getAllocatedContainers()) {
ContainerId rejectContainerId = container.getId();
releases.add(rejectContainerId);
}
for (NMToken token : allocResponse.getNMTokens()) {
String nodeID = token.getNodeId().toString();
receivedNMTokens.put(nodeID, token.getToken());
}
if (allocatedContainerCount < containersRequestedAny) {
// sleep to let NM's heartbeat to RM and trigger allocations
sleep(100);
}
}
assertEquals(allocatedContainerCount, containersRequestedAny + oppContainersRequestedAny);
for (ContainerId rejectContainerId : releases) {
amClient.releaseAssignedContainer(rejectContainerId);
}
assertEquals(3, amClient.release.size());
assertEquals(0, amClient.ask.size());
// need to tell the AMRMClient that we dont need these resources anymore
amClient.removeContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.removeContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.removeContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority2, 0, true, null, ExecutionTypeRequest.newInstance(ExecutionType.OPPORTUNISTIC, true)));
assertEquals(4, amClient.ask.size());
// test RPC exception handling
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amClient.addContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority2, 0, true, null, ExecutionTypeRequest.newInstance(ExecutionType.OPPORTUNISTIC, true)));
final AMRMClient amc = amClient;
ApplicationMasterProtocol realRM = amClient.rmClient;
try {
ApplicationMasterProtocol mockRM = mock(ApplicationMasterProtocol.class);
when(mockRM.allocate(any(AllocateRequest.class))).thenAnswer(new Answer<AllocateResponse>() {
public AllocateResponse answer(InvocationOnMock invocation) throws Exception {
amc.removeContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amc.removeContainerRequest(new AMRMClient.ContainerRequest(capability, nodes, racks, priority));
amc.removeContainerRequest(new AMRMClient.ContainerRequest(capability, null, null, priority2, 0, true, null, ExecutionTypeRequest.newInstance(ExecutionType.OPPORTUNISTIC, true)));
throw new Exception();
}
});
amClient.rmClient = mockRM;
amClient.allocate(0.1f);
} catch (Exception ioe) {
} finally {
amClient.rmClient = realRM;
}
assertEquals(3, amClient.release.size());
assertEquals(6, amClient.ask.size());
iterationsLeft = 3;
// do a few iterations to ensure RM is not going send new containers
while (iterationsLeft-- > 0) {
// inform RM of rejection
AllocateResponse allocResponse = amClient.allocate(0.1f);
// RM did not send new containers because AM does not need any
assertEquals(0, allocResponse.getAllocatedContainers().size());
if (allocResponse.getCompletedContainersStatuses().size() > 0) {
for (ContainerStatus cStatus : allocResponse.getCompletedContainersStatuses()) {
if (releases.contains(cStatus.getContainerId())) {
assertEquals(cStatus.getState(), ContainerState.COMPLETE);
assertEquals(-100, cStatus.getExitStatus());
releases.remove(cStatus.getContainerId());
}
}
}
if (iterationsLeft > 0) {
// sleep to make sure NM's heartbeat
sleep(100);
}
}
assertEquals(0, amClient.ask.size());
assertEquals(0, amClient.release.size());
amClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, null, null);
} finally {
if (amClient != null && amClient.getServiceState() == Service.STATE.STARTED) {
amClient.stop();
}
}
}
use of org.apache.hadoop.yarn.api.records.NMToken in project hadoop by apache.
the class ApplicationMasterService method registerApplicationMaster.
@Override
public RegisterApplicationMasterResponse registerApplicationMaster(RegisterApplicationMasterRequest request) throws YarnException, IOException {
AMRMTokenIdentifier amrmTokenIdentifier = YarnServerSecurityUtils.authorizeRequest();
ApplicationAttemptId applicationAttemptId = amrmTokenIdentifier.getApplicationAttemptId();
ApplicationId appID = applicationAttemptId.getApplicationId();
AllocateResponseLock lock = responseMap.get(applicationAttemptId);
if (lock == null) {
RMAuditLogger.logFailure(this.rmContext.getRMApps().get(appID).getUser(), AuditConstants.REGISTER_AM, "Application doesn't exist in cache " + applicationAttemptId, "ApplicationMasterService", "Error in registering application master", appID, applicationAttemptId);
throwApplicationDoesNotExistInCacheException(applicationAttemptId);
}
// Allow only one thread in AM to do registerApp at a time.
synchronized (lock) {
AllocateResponse lastResponse = lock.getAllocateResponse();
if (hasApplicationMasterRegistered(applicationAttemptId)) {
String message = "Application Master is already registered : " + appID;
LOG.warn(message);
RMAuditLogger.logFailure(this.rmContext.getRMApps().get(appID).getUser(), AuditConstants.REGISTER_AM, "", "ApplicationMasterService", message, appID, applicationAttemptId);
throw new InvalidApplicationMasterRequestException(message);
}
this.amLivelinessMonitor.receivedPing(applicationAttemptId);
RMApp app = this.rmContext.getRMApps().get(appID);
// Setting the response id to 0 to identify if the
// application master is register for the respective attemptid
lastResponse.setResponseId(0);
lock.setAllocateResponse(lastResponse);
LOG.info("AM registration " + applicationAttemptId);
this.rmContext.getDispatcher().getEventHandler().handle(new RMAppAttemptRegistrationEvent(applicationAttemptId, request.getHost(), request.getRpcPort(), request.getTrackingUrl()));
RMAuditLogger.logSuccess(app.getUser(), AuditConstants.REGISTER_AM, "ApplicationMasterService", appID, applicationAttemptId);
// Pick up min/max resource from scheduler...
RegisterApplicationMasterResponse response = recordFactory.newRecordInstance(RegisterApplicationMasterResponse.class);
response.setMaximumResourceCapability(rScheduler.getMaximumResourceCapability(app.getQueue()));
response.setApplicationACLs(app.getRMAppAttempt(applicationAttemptId).getSubmissionContext().getAMContainerSpec().getApplicationACLs());
response.setQueue(app.getQueue());
if (UserGroupInformation.isSecurityEnabled()) {
LOG.info("Setting client token master key");
response.setClientToAMTokenMasterKey(java.nio.ByteBuffer.wrap(rmContext.getClientToAMTokenSecretManager().getMasterKey(applicationAttemptId).getEncoded()));
}
// and corresponding NM tokens.
if (app.getApplicationSubmissionContext().getKeepContainersAcrossApplicationAttempts()) {
List<Container> transferredContainers = rScheduler.getTransferredContainers(applicationAttemptId);
if (!transferredContainers.isEmpty()) {
response.setContainersFromPreviousAttempts(transferredContainers);
List<NMToken> nmTokens = new ArrayList<NMToken>();
for (Container container : transferredContainers) {
try {
NMToken token = rmContext.getNMTokenSecretManager().createAndGetNMToken(app.getUser(), applicationAttemptId, container);
if (null != token) {
nmTokens.add(token);
}
} catch (IllegalArgumentException e) {
// will be automatically retried by RMProxy in RPC layer.
if (e.getCause() instanceof UnknownHostException) {
throw (UnknownHostException) e.getCause();
}
}
}
response.setNMTokensFromPreviousAttempts(nmTokens);
LOG.info("Application " + appID + " retrieved " + transferredContainers.size() + " containers from previous" + " attempts and " + nmTokens.size() + " NM tokens.");
}
}
response.setSchedulerResourceTypes(rScheduler.getSchedulingResourceTypes());
return response;
}
}
use of org.apache.hadoop.yarn.api.records.NMToken in project hadoop by apache.
the class RegisterApplicationMasterResponsePBImpl method initLocalNewNMTokenList.
private synchronized void initLocalNewNMTokenList() {
RegisterApplicationMasterResponseProtoOrBuilder p = viaProto ? proto : builder;
List<NMTokenProto> list = p.getNmTokensFromPreviousAttemptsList();
nmTokens = new ArrayList<NMToken>();
for (NMTokenProto t : list) {
nmTokens.add(convertFromProtoFormat(t));
}
}
use of org.apache.hadoop.yarn.api.records.NMToken in project hadoop by apache.
the class NMTokenSecretManagerInRM method createAndGetNMToken.
public NMToken createAndGetNMToken(String applicationSubmitter, ApplicationAttemptId appAttemptId, Container container) {
try {
this.writeLock.lock();
HashSet<NodeId> nodeSet = this.appAttemptToNodeKeyMap.get(appAttemptId);
NMToken nmToken = null;
if (nodeSet != null) {
if (!nodeSet.contains(container.getNodeId())) {
LOG.info("Sending NMToken for nodeId : " + container.getNodeId() + " for container : " + container.getId());
Token token = createNMToken(container.getId().getApplicationAttemptId(), container.getNodeId(), applicationSubmitter);
nmToken = NMToken.newInstance(container.getNodeId(), token);
nodeSet.add(container.getNodeId());
}
}
return nmToken;
} finally {
this.writeLock.unlock();
}
}
use of org.apache.hadoop.yarn.api.records.NMToken in project hadoop by apache.
the class MockResourceManagerFacade method allocate.
@SuppressWarnings("deprecation")
@Override
public AllocateResponse allocate(AllocateRequest request) throws YarnException, IOException {
if (request.getAskList() != null && request.getAskList().size() > 0 && request.getReleaseList() != null && request.getReleaseList().size() > 0) {
Assert.fail("The mock RM implementation does not support receiving " + "askList and releaseList in the same heartbeat");
}
String amrmToken = getAppIdentifier();
ArrayList<Container> containerList = new ArrayList<Container>();
if (request.getAskList() != null) {
for (ResourceRequest rr : request.getAskList()) {
for (int i = 0; i < rr.getNumContainers(); i++) {
ContainerId containerId = ContainerId.newInstance(getApplicationAttemptId(1), containerIndex.incrementAndGet());
Container container = Records.newRecord(Container.class);
container.setId(containerId);
container.setPriority(rr.getPriority());
// We don't use the node for running containers in the test cases. So
// it is OK to hard code it to some dummy value
NodeId nodeId = NodeId.newInstance(!Strings.isNullOrEmpty(rr.getResourceName()) ? rr.getResourceName() : "dummy", 1000);
container.setNodeId(nodeId);
container.setResource(rr.getCapability());
containerList.add(container);
synchronized (applicationContainerIdMap) {
// Keep track of the containers returned to this application. We
// will need it in future
Assert.assertTrue("The application id is Not registered before allocate(): " + amrmToken, applicationContainerIdMap.containsKey(amrmToken));
List<ContainerId> ids = applicationContainerIdMap.get(amrmToken);
ids.add(containerId);
this.allocatedContainerMap.put(containerId, container);
}
}
}
}
if (request.getReleaseList() != null && request.getReleaseList().size() > 0) {
Log.getLog().info("Releasing containers: " + request.getReleaseList().size());
synchronized (applicationContainerIdMap) {
Assert.assertTrue("The application id is not registered before allocate(): " + amrmToken, applicationContainerIdMap.containsKey(amrmToken));
List<ContainerId> ids = applicationContainerIdMap.get(amrmToken);
for (ContainerId id : request.getReleaseList()) {
boolean found = false;
for (ContainerId c : ids) {
if (c.equals(id)) {
found = true;
break;
}
}
Assert.assertTrue("ContainerId " + id + " being released is not valid for application: " + conf.get("AMRMTOKEN"), found);
ids.remove(id);
// Return the released container back to the AM with new fake Ids. The
// test case does not care about the IDs. The IDs are faked because
// otherwise the LRM will throw duplication identifier exception. This
// returning of fake containers is ONLY done for testing purpose - for
// the test code to get confirmation that the sub-cluster resource
// managers received the release request
ContainerId fakeContainerId = ContainerId.newInstance(getApplicationAttemptId(1), containerIndex.incrementAndGet());
Container fakeContainer = allocatedContainerMap.get(id);
fakeContainer.setId(fakeContainerId);
containerList.add(fakeContainer);
}
}
}
Log.getLog().info("Allocating containers: " + containerList.size() + " for application attempt: " + conf.get("AMRMTOKEN"));
// Always issue a new AMRMToken as if RM rolled master key
Token newAMRMToken = Token.newInstance(new byte[0], "", new byte[0], "");
return AllocateResponse.newInstance(0, new ArrayList<ContainerStatus>(), containerList, new ArrayList<NodeReport>(), null, AMCommand.AM_RESYNC, 1, null, new ArrayList<NMToken>(), newAMRMToken, new ArrayList<UpdatedContainer>());
}
Aggregations