Examples with RMAppAttempt org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt used on opensource projects

Example 1 with RMAppAttempt

use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.

the class TestWorkPreservingRMRestart method testDynamicQueueRecovery.

// Test work preserving recovery of apps running under reservation.
// This involves:
// 1. Setting up a dynamic reservable queue,
// 2. Submitting an app to it,
// 3. Failing over RM,
// 4. Validating that the app is recovered post failover,
// 5. Check if all running containers are recovered,
// 6. Verify the scheduler state like attempt info,
// 7. Verify the queue/user metrics for the dynamic reservable queue.
@Test(timeout = 30000)
public void testDynamicQueueRecovery() throws Exception {
    conf.setBoolean(CapacitySchedulerConfiguration.ENABLE_USER_METRICS, true);
    conf.set(CapacitySchedulerConfiguration.RESOURCE_CALCULATOR_CLASS, DominantResourceCalculator.class.getName());
    // 1. Set up dynamic reservable queue.
    Configuration schedulerConf = getSchedulerDynamicConfiguration();
    int containerMemory = 1024;
    Resource containerResource = Resource.newInstance(containerMemory, 1);
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(schedulerConf);
    rm1 = new MockRM(schedulerConf, memStore);
    rm1.start();
    MockNM nm1 = new MockNM("127.0.0.1:1234", 8192, rm1.getResourceTrackerService());
    nm1.registerNode();
    // 2. Run plan follower to update the added node & then submit app to
    // dynamic queue.
    rm1.getRMContext().getReservationSystem().synchronizePlan(ReservationSystemTestUtil.reservationQ, true);
    RMApp app1 = rm1.submitApp(200, "dynamicQApp", UserGroupInformation.getCurrentUser().getShortUserName(), null, ReservationSystemTestUtil.getReservationQueueName());
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    // clear queue metrics
    rm1.clearQueueMetrics(app1);
    // 3. Fail over (restart) RM.
    rm2 = new MockRM(schedulerConf, memStore);
    rm2.start();
    nm1.setResourceTrackerService(rm2.getResourceTrackerService());
    // 4. Validate app is recovered post failover.
    RMApp recoveredApp1 = rm2.getRMContext().getRMApps().get(app1.getApplicationId());
    RMAppAttempt loadedAttempt1 = recoveredApp1.getCurrentAppAttempt();
    NMContainerStatus amContainer = TestRMRestart.createNMContainerStatus(am1.getApplicationAttemptId(), 1, ContainerState.RUNNING);
    NMContainerStatus runningContainer = TestRMRestart.createNMContainerStatus(am1.getApplicationAttemptId(), 2, ContainerState.RUNNING);
    NMContainerStatus completedContainer = TestRMRestart.createNMContainerStatus(am1.getApplicationAttemptId(), 3, ContainerState.COMPLETE);
    nm1.registerNode(Arrays.asList(amContainer, runningContainer, completedContainer), null);
    // Wait for RM to settle down on recovering containers.
    waitForNumContainersToRecover(2, rm2, am1.getApplicationAttemptId());
    Set<ContainerId> launchedContainers = ((RMNodeImpl) rm2.getRMContext().getRMNodes().get(nm1.getNodeId())).getLaunchedContainers();
    assertTrue(launchedContainers.contains(amContainer.getContainerId()));
    assertTrue(launchedContainers.contains(runningContainer.getContainerId()));
    // 5. Check RMContainers are re-recreated and the container state is
    // correct.
    rm2.waitForState(nm1, amContainer.getContainerId(), RMContainerState.RUNNING);
    rm2.waitForState(nm1, runningContainer.getContainerId(), RMContainerState.RUNNING);
    rm2.waitForContainerToComplete(loadedAttempt1, completedContainer);
    AbstractYarnScheduler scheduler = (AbstractYarnScheduler) rm2.getResourceScheduler();
    SchedulerNode schedulerNode1 = scheduler.getSchedulerNode(nm1.getNodeId());
    // ********* check scheduler node state.*******
    // 2 running containers.
    Resource usedResources = Resources.multiply(containerResource, 2);
    Resource nmResource = Resource.newInstance(nm1.getMemory(), nm1.getvCores());
    assertTrue(schedulerNode1.isValidContainer(amContainer.getContainerId()));
    assertTrue(schedulerNode1.isValidContainer(runningContainer.getContainerId()));
    assertFalse(schedulerNode1.isValidContainer(completedContainer.getContainerId()));
    // 2 launched containers, 1 completed container
    assertEquals(2, schedulerNode1.getNumContainers());
    assertEquals(Resources.subtract(nmResource, usedResources), schedulerNode1.getUnallocatedResource());
    assertEquals(usedResources, schedulerNode1.getAllocatedResource());
    Resource availableResources = Resources.subtract(nmResource, usedResources);
    // 6. Verify the scheduler state like attempt info.
    Map<ApplicationId, SchedulerApplication<SchedulerApplicationAttempt>> sa = ((AbstractYarnScheduler) rm2.getResourceScheduler()).getSchedulerApplications();
    SchedulerApplication<SchedulerApplicationAttempt> schedulerApp = sa.get(recoveredApp1.getApplicationId());
    // 7. Verify the queue/user metrics for the dynamic reservable queue.
    if (getSchedulerType() == SchedulerType.CAPACITY) {
        checkCSQueue(rm2, schedulerApp, nmResource, nmResource, usedResources, 2);
    } else {
        checkFSQueue(rm2, schedulerApp, usedResources, availableResources);
    }
    // *********** check scheduler attempt state.********
    SchedulerApplicationAttempt schedulerAttempt = schedulerApp.getCurrentAppAttempt();
    assertTrue(schedulerAttempt.getLiveContainers().contains(scheduler.getRMContainer(amContainer.getContainerId())));
    assertTrue(schedulerAttempt.getLiveContainers().contains(scheduler.getRMContainer(runningContainer.getContainerId())));
    assertEquals(schedulerAttempt.getCurrentConsumption(), usedResources);
    // *********** check appSchedulingInfo state ***********
    assertEquals((1L << 40) + 1L, schedulerAttempt.getNewContainerId());
}

Example 2 with RMAppAttempt

use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.

the class TestZKRMStateStore method testFencedState.

@Test
public void testFencedState() throws Exception {
    TestZKRMStateStoreTester zkTester = new TestZKRMStateStoreTester();
    RMStateStore store = zkTester.getRMStateStore();
    // Move state to FENCED from ACTIVE
    store.updateFencedState();
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    long submitTime = System.currentTimeMillis();
    long startTime = submitTime + 1000;
    // Add a new app
    RMApp mockApp = mock(RMApp.class);
    ApplicationSubmissionContext context = new ApplicationSubmissionContextPBImpl();
    when(mockApp.getSubmitTime()).thenReturn(submitTime);
    when(mockApp.getStartTime()).thenReturn(startTime);
    when(mockApp.getApplicationSubmissionContext()).thenReturn(context);
    when(mockApp.getUser()).thenReturn("test");
    store.storeNewApplication(mockApp);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // Add a new attempt
    ClientToAMTokenSecretManagerInRM clientToAMTokenMgr = new ClientToAMTokenSecretManagerInRM();
    ApplicationAttemptId attemptId = ApplicationAttemptId.fromString("appattempt_1234567894321_0001_000001");
    SecretKey clientTokenMasterKey = clientToAMTokenMgr.createMasterKey(attemptId);
    RMAppAttemptMetrics mockRmAppAttemptMetrics = mock(RMAppAttemptMetrics.class);
    Container container = new ContainerPBImpl();
    container.setId(ContainerId.fromString("container_1234567891234_0001_01_000001"));
    RMAppAttempt mockAttempt = mock(RMAppAttempt.class);
    when(mockAttempt.getAppAttemptId()).thenReturn(attemptId);
    when(mockAttempt.getMasterContainer()).thenReturn(container);
    when(mockAttempt.getClientTokenMasterKey()).thenReturn(clientTokenMasterKey);
    when(mockAttempt.getRMAppAttemptMetrics()).thenReturn(mockRmAppAttemptMetrics);
    when(mockRmAppAttemptMetrics.getAggregateAppResourceUsage()).thenReturn(new AggregateAppResourceUsage(0, 0));
    store.storeNewApplicationAttempt(mockAttempt);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    long finishTime = submitTime + 1000;
    // Update attempt
    ApplicationAttemptStateData newAttemptState = ApplicationAttemptStateData.newInstance(attemptId, container, store.getCredentialsFromAppAttempt(mockAttempt), startTime, RMAppAttemptState.FINISHED, "testUrl", "test", FinalApplicationStatus.SUCCEEDED, 100, finishTime, 0, 0, 0, 0);
    store.updateApplicationAttemptState(newAttemptState);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // Update app
    ApplicationStateData appState = ApplicationStateData.newInstance(submitTime, startTime, context, "test");
    store.updateApplicationState(appState);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // Remove app
    store.removeApplication(mockApp);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // store RM delegation token;
    RMDelegationTokenIdentifier dtId1 = new RMDelegationTokenIdentifier(new Text("owner1"), new Text("renewer1"), new Text("realuser1"));
    Long renewDate1 = new Long(System.currentTimeMillis());
    dtId1.setSequenceNumber(1111);
    store.storeRMDelegationToken(dtId1, renewDate1);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    store.updateRMDelegationToken(dtId1, renewDate1);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // remove delegation key;
    store.removeRMDelegationToken(dtId1);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // store delegation master key;
    DelegationKey key = new DelegationKey(1234, 4321, "keyBytes".getBytes());
    store.storeRMDTMasterKey(key);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // remove delegation master key;
    store.removeRMDTMasterKey(key);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    // store or update AMRMToken;
    store.storeOrUpdateAMRMTokenSecretManager(null, false);
    assertEquals("RMStateStore should have been in fenced state", true, store.isFencedState());
    store.close();
}

Example 3 with RMAppAttempt

use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.

the class RMStateStoreTestBase method storeAttempt.

protected RMAppAttempt storeAttempt(RMStateStore store, ApplicationAttemptId attemptId, String containerIdStr, Token<AMRMTokenIdentifier> appToken, SecretKey clientTokenMasterKey, TestDispatcher dispatcher) throws Exception {
    RMAppAttemptMetrics mockRmAppAttemptMetrics = mock(RMAppAttemptMetrics.class);
    Container container = new ContainerPBImpl();
    container.setId(ContainerId.fromString(containerIdStr));
    RMAppAttempt mockAttempt = mock(RMAppAttempt.class);
    when(mockAttempt.getAppAttemptId()).thenReturn(attemptId);
    when(mockAttempt.getMasterContainer()).thenReturn(container);
    when(mockAttempt.getAMRMToken()).thenReturn(appToken);
    when(mockAttempt.getClientTokenMasterKey()).thenReturn(clientTokenMasterKey);
    when(mockAttempt.getRMAppAttemptMetrics()).thenReturn(mockRmAppAttemptMetrics);
    when(mockRmAppAttemptMetrics.getAggregateAppResourceUsage()).thenReturn(new AggregateAppResourceUsage(0, 0));
    dispatcher.attemptId = attemptId;
    store.storeNewApplicationAttempt(mockAttempt);
    waitNotify(dispatcher);
    return mockAttempt;
}

Example 4 with RMAppAttempt

use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.

the class TestContainerAllocation method testAMContainerAllocationWhenDNSUnavailable.

// This is to test fetching AM container will be retried, if AM container is
// not fetchable since DNS is unavailable causing container token/NMtoken
// creation failure.
@Test(timeout = 30000)
public void testAMContainerAllocationWhenDNSUnavailable() throws Exception {
    MockRM rm1 = new MockRM(conf) {

        @Override
        protected RMSecretManagerService createRMSecretManagerService() {
            return new TestRMSecretManagerService(conf, rmContext);
        }
    };
    rm1.start();
    MockNM nm1 = rm1.registerNode("unknownhost:1234", 8000);
    RMApp app1;
    try {
        SecurityUtilTestHelper.setTokenServiceUseIp(true);
        app1 = rm1.submitApp(200);
        RMAppAttempt attempt = app1.getCurrentAppAttempt();
        nm1.nodeHeartbeat(true);
        // fetching am container will fail, keep retrying 5 times.
        while (numRetries <= 5) {
            nm1.nodeHeartbeat(true);
            Thread.sleep(1000);
            Assert.assertEquals(RMAppAttemptState.SCHEDULED, attempt.getAppAttemptState());
            System.out.println("Waiting for am container to be allocated.");
        }
    } finally {
        SecurityUtilTestHelper.setTokenServiceUseIp(false);
    }
    MockRM.launchAndRegisterAM(app1, rm1, nm1);
}

Example 5 with RMAppAttempt

use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.

the class TestCapacitySchedulerNodeLabelUpdate method testBlacklistAMDisableLabel.

@Test(timeout = 30000)
public void testBlacklistAMDisableLabel() throws Exception {
    conf.setBoolean(YarnConfiguration.AM_SCHEDULING_NODE_BLACKLISTING_ENABLED, true);
    conf.setFloat(YarnConfiguration.AM_SCHEDULING_NODE_BLACKLISTING_DISABLE_THRESHOLD, 0.5f);
    mgr.addToCluserNodeLabelsWithDefaultExclusivity(ImmutableSet.of("x", "y"));
    mgr.addLabelsToNode(ImmutableMap.of(NodeId.newInstance("h2", 0), toSet("x"), NodeId.newInstance("h3", 0), toSet("x"), NodeId.newInstance("h6", 0), toSet("x")));
    mgr.addLabelsToNode(ImmutableMap.of(NodeId.newInstance("h4", 0), toSet("y"), NodeId.newInstance("h5", 0), toSet("y"), NodeId.newInstance("h7", 0), toSet("y")));
    MockRM rm = new MockRM(getConfigurationWithQueueLabels(conf)) {

        @Override
        public RMNodeLabelsManager createNodeLabelManager() {
            return mgr;
        }
    };
    rm.getRMContext().setNodeLabelManager(mgr);
    rm.start();
    // Nodes in label default h1,h8,h9
    // Nodes in label x h2,h3,h6
    // Nodes in label y h4,h5,h7
    MockNM nm1 = rm.registerNode("h1:1234", 2048);
    MockNM nm2 = rm.registerNode("h2:1234", 2048);
    rm.registerNode("h3:1234", 2048);
    rm.registerNode("h4:1234", 2048);
    rm.registerNode("h5:1234", 2048);
    rm.registerNode("h6:1234", 2048);
    rm.registerNode("h7:1234", 2048);
    rm.registerNode("h8:1234", 2048);
    rm.registerNode("h9:1234", 2048);
    // Submit app with AM container launched on default partition i.e. h1.
    RMApp app = rm.submitApp(GB, "app", "user", null, "a");
    MockRM.launchAndRegisterAM(app, rm, nm1);
    RMAppAttempt appAttempt = app.getCurrentAppAttempt();
    // Add default node blacklist from default
    appAttempt.getAMBlacklistManager().addNode("h1");
    ResourceBlacklistRequest blacklistUpdates = appAttempt.getAMBlacklistManager().getBlacklistUpdates();
    Assert.assertEquals(1, blacklistUpdates.getBlacklistAdditions().size());
    Assert.assertEquals(0, blacklistUpdates.getBlacklistRemovals().size());
    // Adding second node from default parition
    appAttempt.getAMBlacklistManager().addNode("h8");
    blacklistUpdates = appAttempt.getAMBlacklistManager().getBlacklistUpdates();
    Assert.assertEquals(0, blacklistUpdates.getBlacklistAdditions().size());
    Assert.assertEquals(2, blacklistUpdates.getBlacklistRemovals().size());
    // Submission in label x
    RMApp applabel = rm.submitApp(GB, "app", "user", null, "a", "x");
    MockRM.launchAndRegisterAM(applabel, rm, nm2);
    RMAppAttempt appAttemptlabelx = applabel.getCurrentAppAttempt();
    appAttemptlabelx.getAMBlacklistManager().addNode("h2");
    ResourceBlacklistRequest blacklistUpdatesOnx = appAttemptlabelx.getAMBlacklistManager().getBlacklistUpdates();
    Assert.assertEquals(1, blacklistUpdatesOnx.getBlacklistAdditions().size());
    Assert.assertEquals(0, blacklistUpdatesOnx.getBlacklistRemovals().size());
    // Adding second node from default parition
    appAttemptlabelx.getAMBlacklistManager().addNode("h3");
    blacklistUpdatesOnx = appAttempt.getAMBlacklistManager().getBlacklistUpdates();
    Assert.assertEquals(0, blacklistUpdatesOnx.getBlacklistAdditions().size());
    Assert.assertEquals(2, blacklistUpdatesOnx.getBlacklistRemovals().size());
    rm.close();
}