use of org.apache.hadoop.yarn.api.records.Priority in project hadoop by apache.
the class TestDistributedScheduling method testAMOpportunistic.
/**
* Check if an AM can ask for opportunistic containers and get them.
* @throws Exception
*/
@Test
public void testAMOpportunistic() throws Exception {
// Basic container to request
Resource capability = Resource.newInstance(1024, 1);
Priority priority = Priority.newInstance(1);
// Get the cluster topology
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 };
// Create an AM to request resources
AMRMClient<AMRMClient.ContainerRequest> amClient = null;
try {
amClient = new AMRMClientImpl<AMRMClient.ContainerRequest>(client);
amClient.init(yarnConf);
amClient.start();
amClient.registerApplicationMaster(NetUtils.getHostname(), 1024, "");
// AM requests an opportunistic container
ExecutionTypeRequest execTypeRequest = ExecutionTypeRequest.newInstance(ExecutionType.OPPORTUNISTIC, true);
ContainerRequest containerRequest = new AMRMClient.ContainerRequest(capability, nodes, racks, priority, 0, true, null, execTypeRequest);
amClient.addContainerRequest(containerRequest);
// Wait until the container is allocated
ContainerId opportunisticContainerId = null;
for (int i = 0; i < 10 && opportunisticContainerId == null; i++) {
AllocateResponse allocResponse = amClient.allocate(0.1f);
List<Container> allocatedContainers = allocResponse.getAllocatedContainers();
for (Container allocatedContainer : allocatedContainers) {
// Check that this is the container we required
assertEquals(ExecutionType.OPPORTUNISTIC, allocatedContainer.getExecutionType());
opportunisticContainerId = allocatedContainer.getId();
}
sleep(100);
}
assertNotNull(opportunisticContainerId);
// The RM sees the container as OPPORTUNISTIC
ResourceScheduler scheduler = cluster.getResourceManager().getResourceScheduler();
RMContainer rmContainer = scheduler.getRMContainer(opportunisticContainerId);
assertEquals(ExecutionType.OPPORTUNISTIC, rmContainer.getExecutionType());
// Release the opportunistic container
amClient.releaseAssignedContainer(opportunisticContainerId);
// Wait for the release container to appear
boolean released = false;
for (int i = 0; i < 10 && !released; i++) {
AllocateResponse allocResponse = amClient.allocate(0.1f);
List<ContainerStatus> completedContainers = allocResponse.getCompletedContainersStatuses();
for (ContainerStatus completedContainer : completedContainers) {
ContainerId completedContainerId = completedContainer.getContainerId();
assertEquals(completedContainerId, opportunisticContainerId);
released = true;
}
if (!released) {
sleep(100);
}
}
assertTrue(released);
// The RM shouldn't see the container anymore
rmContainer = scheduler.getRMContainer(opportunisticContainerId);
assertNull(rmContainer);
// Clean the AM
amClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, null, null);
} finally {
if (amClient != null && amClient.getServiceState() == Service.STATE.STARTED) {
amClient.close();
}
}
}
use of org.apache.hadoop.yarn.api.records.Priority 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.Priority in project hadoop by apache.
the class ClientRMService method updateApplicationPriority.
@Override
public UpdateApplicationPriorityResponse updateApplicationPriority(UpdateApplicationPriorityRequest request) throws YarnException, IOException {
ApplicationId applicationId = request.getApplicationId();
Priority newAppPriority = request.getApplicationPriority();
UserGroupInformation callerUGI = getCallerUgi(applicationId, AuditConstants.UPDATE_APP_PRIORITY);
RMApp application = verifyUserAccessForRMApp(applicationId, callerUGI, AuditConstants.UPDATE_APP_PRIORITY);
UpdateApplicationPriorityResponse response = recordFactory.newRecordInstance(UpdateApplicationPriorityResponse.class);
// Update priority only when app is tracked by the scheduler
if (!ACTIVE_APP_STATES.contains(application.getState())) {
if (application.isAppInCompletedStates()) {
// If Application is in any of the final states, change priority
// can be skipped rather throwing exception.
RMAuditLogger.logSuccess(callerUGI.getShortUserName(), AuditConstants.UPDATE_APP_PRIORITY, "ClientRMService", applicationId);
response.setApplicationPriority(application.getApplicationPriority());
return response;
}
String msg = "Application in " + application.getState() + " state cannot update priority.";
RMAuditLogger.logFailure(callerUGI.getShortUserName(), AuditConstants.UPDATE_APP_PRIORITY, "UNKNOWN", "ClientRMService", msg);
throw new YarnException(msg);
}
try {
rmAppManager.updateApplicationPriority(callerUGI, application.getApplicationId(), newAppPriority);
} catch (YarnException ex) {
RMAuditLogger.logFailure(callerUGI.getShortUserName(), AuditConstants.UPDATE_APP_PRIORITY, "UNKNOWN", "ClientRMService", ex.getMessage());
throw ex;
}
RMAuditLogger.logSuccess(callerUGI.getShortUserName(), AuditConstants.UPDATE_APP_PRIORITY, "ClientRMService", applicationId);
response.setApplicationPriority(application.getApplicationPriority());
return response;
}
use of org.apache.hadoop.yarn.api.records.Priority in project hadoop by apache.
the class RMAppManager method createAndPopulateNewRMApp.
private RMAppImpl createAndPopulateNewRMApp(ApplicationSubmissionContext submissionContext, long submitTime, String user, boolean isRecovery, long startTime) throws YarnException {
if (!isRecovery) {
// Do queue mapping
if (rmContext.getQueuePlacementManager() != null) {
// We only do queue mapping when it's a new application
rmContext.getQueuePlacementManager().placeApplication(submissionContext, user);
}
// fail the submission if configured application timeout value is invalid
RMServerUtils.validateApplicationTimeouts(submissionContext.getApplicationTimeouts());
}
ApplicationId applicationId = submissionContext.getApplicationId();
ResourceRequest amReq = null;
try {
amReq = validateAndCreateResourceRequest(submissionContext, isRecovery);
} catch (InvalidLabelResourceRequestException e) {
// after recovery and user can see what's going on and react accordingly.
if (isRecovery && !YarnConfiguration.areNodeLabelsEnabled(this.conf)) {
if (LOG.isDebugEnabled()) {
LOG.debug("AMResourceRequest is not created for " + applicationId + ". NodeLabel is not enabled in cluster, but AM resource " + "request contains a label expression.");
}
} else {
throw e;
}
}
// Verify and get the update application priority and set back to
// submissionContext
UserGroupInformation userUgi = UserGroupInformation.createRemoteUser(user);
Priority appPriority = scheduler.checkAndGetApplicationPriority(submissionContext.getPriority(), userUgi, submissionContext.getQueue(), applicationId);
submissionContext.setPriority(appPriority);
// For now, exclude FS for the acl check.
if (!isRecovery && YarnConfiguration.isAclEnabled(conf) && scheduler instanceof CapacityScheduler) {
String queueName = submissionContext.getQueue();
String appName = submissionContext.getApplicationName();
CSQueue csqueue = ((CapacityScheduler) scheduler).getQueue(queueName);
if (null != csqueue && !authorizer.checkPermission(new AccessRequest(csqueue.getPrivilegedEntity(), userUgi, SchedulerUtils.toAccessType(QueueACL.SUBMIT_APPLICATIONS), applicationId.toString(), appName, Server.getRemoteAddress(), null)) && !authorizer.checkPermission(new AccessRequest(csqueue.getPrivilegedEntity(), userUgi, SchedulerUtils.toAccessType(QueueACL.ADMINISTER_QUEUE), applicationId.toString(), appName, Server.getRemoteAddress(), null))) {
throw RPCUtil.getRemoteException(new AccessControlException("User " + user + " does not have permission to submit " + applicationId + " to queue " + submissionContext.getQueue()));
}
}
// Create RMApp
RMAppImpl application = new RMAppImpl(applicationId, rmContext, this.conf, submissionContext.getApplicationName(), user, submissionContext.getQueue(), submissionContext, this.scheduler, this.masterService, submitTime, submissionContext.getApplicationType(), submissionContext.getApplicationTags(), amReq, startTime);
// influence each other
if (rmContext.getRMApps().putIfAbsent(applicationId, application) != null) {
String message = "Application with id " + applicationId + " is already present! Cannot add a duplicate!";
LOG.warn(message);
throw new YarnException(message);
}
if (YarnConfiguration.timelineServiceV2Enabled(conf)) {
// Start timeline collector for the submitted app
application.startTimelineCollector();
}
// Inform the ACLs Manager
this.applicationACLsManager.addApplication(applicationId, submissionContext.getAMContainerSpec().getApplicationACLs());
String appViewACLs = submissionContext.getAMContainerSpec().getApplicationACLs().get(ApplicationAccessType.VIEW_APP);
rmContext.getSystemMetricsPublisher().appACLsUpdated(application, appViewACLs, System.currentTimeMillis());
return application;
}
use of org.apache.hadoop.yarn.api.records.Priority in project hadoop by apache.
the class RMAppManager method updateApplicationPriority.
/**
* updateApplicationPriority will invoke scheduler api to update the
* new priority to RM and StateStore.
* @param callerUGI user
* @param applicationId Application Id
* @param newAppPriority proposed new application priority
* @throws YarnException Handle exceptions
*/
public void updateApplicationPriority(UserGroupInformation callerUGI, ApplicationId applicationId, Priority newAppPriority) throws YarnException {
RMApp app = this.rmContext.getRMApps().get(applicationId);
synchronized (applicationId) {
if (app == null || app.isAppInCompletedStates()) {
return;
}
// Create a future object to capture exceptions from StateStore.
SettableFuture<Object> future = SettableFuture.create();
// Invoke scheduler api to update priority in scheduler and to
// State Store.
Priority appPriority = rmContext.getScheduler().updateApplicationPriority(newAppPriority, applicationId, future, callerUGI);
if (app.getApplicationPriority().equals(appPriority)) {
return;
}
Futures.get(future, YarnException.class);
// update in-memory
((RMAppImpl) app).setApplicationPriority(appPriority);
}
// Update the changed application state to timeline server
rmContext.getSystemMetricsPublisher().appUpdated(app, System.currentTimeMillis());
}
Aggregations