Examples with ApplicationAttemptStateData org.apache.hadoop.yarn.server.resourcemanager.recovery.records.ApplicationAttemptStateData used on opensource projects

Example 6 with ApplicationAttemptStateData

use of org.apache.hadoop.yarn.server.resourcemanager.recovery.records.ApplicationAttemptStateData in project hadoop by apache.

the class TestRMRestart method testRMRestartOnMaxAppAttempts.

@Test(timeout = 60000)
public void testRMRestartOnMaxAppAttempts() throws Exception {
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, YarnConfiguration.DEFAULT_RM_AM_MAX_ATTEMPTS);
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    RMState rmState = memStore.getState();
    Map<ApplicationId, ApplicationStateData> rmAppState = rmState.getApplicationState();
    MockRM rm1 = createMockRM(conf, memStore);
    rm1.start();
    MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm1.getResourceTrackerService());
    nm1.registerNode();
    // submit an app with maxAppAttempts equals to 1
    RMApp app1 = rm1.submitApp(200, "name", "user", new HashMap<ApplicationAccessType, String>(), false, "default", 1, null);
    // submit an app with maxAppAttempts equals to -1
    RMApp app2 = rm1.submitApp(200, "name", "user", new HashMap<ApplicationAccessType, String>(), false, "default", -1, null);
    // assert app1 info is saved
    ApplicationStateData appState = rmAppState.get(app1.getApplicationId());
    Assert.assertNotNull(appState);
    Assert.assertEquals(0, appState.getAttemptCount());
    Assert.assertEquals(appState.getApplicationSubmissionContext().getApplicationId(), app1.getApplicationSubmissionContext().getApplicationId());
    // Allocate the AM
    nm1.nodeHeartbeat(true);
    RMAppAttempt attempt = app1.getCurrentAppAttempt();
    ApplicationAttemptId attemptId1 = attempt.getAppAttemptId();
    rm1.waitForState(attemptId1, RMAppAttemptState.ALLOCATED);
    Assert.assertEquals(1, appState.getAttemptCount());
    ApplicationAttemptStateData attemptState = appState.getAttempt(attemptId1);
    Assert.assertNotNull(attemptState);
    Assert.assertEquals(BuilderUtils.newContainerId(attemptId1, 1), attemptState.getMasterContainer().getId());
    // Setting AMLivelinessMonitor interval to be 3 Secs.
    conf.setInt(YarnConfiguration.RM_AM_EXPIRY_INTERVAL_MS, 3000);
    // start new RM   
    MockRM rm2 = createMockRM(conf, memStore);
    rm2.start();
    // verify that maxAppAttempts is set to global value
    Assert.assertEquals(2, rm2.getRMContext().getRMApps().get(app2.getApplicationId()).getMaxAppAttempts());
    // app1 and app2 are loaded back, but app1 failed because it's
    // hitting max-retry.
    Assert.assertEquals(2, rm2.getRMContext().getRMApps().size());
    rm2.waitForState(app1.getApplicationId(), RMAppState.FAILED);
    rm2.waitForState(app2.getApplicationId(), RMAppState.ACCEPTED);
    // app1 failed state is saved in state store. app2 final saved state is not
    // determined yet.
    Assert.assertEquals(RMAppState.FAILED, rmAppState.get(app1.getApplicationId()).getState());
    Assert.assertNull(rmAppState.get(app2.getApplicationId()).getState());
}

Example 7 with ApplicationAttemptStateData

use of org.apache.hadoop.yarn.server.resourcemanager.recovery.records.ApplicationAttemptStateData in project hadoop by apache.

the class TestRMRestart method testAppAttemptTokensRestoredOnRMRestart.

@Test(timeout = 60000)
public void testAppAttemptTokensRestoredOnRMRestart() throws Exception {
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 2);
    conf.set(CommonConfigurationKeysPublic.HADOOP_SECURITY_AUTHENTICATION, "kerberos");
    UserGroupInformation.setConfiguration(conf);
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    RMState rmState = memStore.getState();
    Map<ApplicationId, ApplicationStateData> rmAppState = rmState.getApplicationState();
    MockRM rm1 = new TestSecurityMockRM(conf, memStore);
    rm1.start();
    MockNM nm1 = new MockNM("0.0.0.0:4321", 15120, rm1.getResourceTrackerService());
    nm1.registerNode();
    // submit an app
    RMApp app1 = rm1.submitApp(200, "name", "user", new HashMap<ApplicationAccessType, String>(), "default");
    // assert app info is saved
    ApplicationStateData appState = rmAppState.get(app1.getApplicationId());
    Assert.assertNotNull(appState);
    // Allocate the AM
    nm1.nodeHeartbeat(true);
    RMAppAttempt attempt1 = app1.getCurrentAppAttempt();
    ApplicationAttemptId attemptId1 = attempt1.getAppAttemptId();
    rm1.waitForState(attemptId1, RMAppAttemptState.ALLOCATED);
    // assert attempt info is saved
    ApplicationAttemptStateData attemptState = appState.getAttempt(attemptId1);
    Assert.assertNotNull(attemptState);
    Assert.assertEquals(BuilderUtils.newContainerId(attemptId1, 1), attemptState.getMasterContainer().getId());
    // the clientTokenMasterKey that are generated when
    // RMAppAttempt is created,
    byte[] clientTokenMasterKey = attempt1.getClientTokenMasterKey().getEncoded();
    // assert application credentials are saved
    Credentials savedCredentials = attemptState.getAppAttemptTokens();
    Assert.assertArrayEquals("client token master key not saved", clientTokenMasterKey, savedCredentials.getSecretKey(RMStateStore.AM_CLIENT_TOKEN_MASTER_KEY_NAME));
    // start new RM
    MockRM rm2 = new TestSecurityMockRM(conf, memStore);
    rm2.start();
    RMApp loadedApp1 = rm2.getRMContext().getRMApps().get(app1.getApplicationId());
    RMAppAttempt loadedAttempt1 = loadedApp1.getRMAppAttempt(attemptId1);
    // assert loaded attempt recovered
    Assert.assertNotNull(loadedAttempt1);
    // assert client token master key is recovered back to api-versioned
    // client token master key
    Assert.assertEquals("client token master key not restored", attempt1.getClientTokenMasterKey(), loadedAttempt1.getClientTokenMasterKey());
    // assert ClientTokenSecretManager also knows about the key
    Assert.assertArrayEquals(clientTokenMasterKey, rm2.getClientToAMTokenSecretManager().getMasterKey(attemptId1).getEncoded());
    // assert AMRMTokenSecretManager also knows about the AMRMToken password
    Token<AMRMTokenIdentifier> amrmToken = loadedAttempt1.getAMRMToken();
    Assert.assertArrayEquals(amrmToken.getPassword(), rm2.getRMContext().getAMRMTokenSecretManager().retrievePassword(amrmToken.decodeIdentifier()));
}

Example 8 with ApplicationAttemptStateData

use of org.apache.hadoop.yarn.server.resourcemanager.recovery.records.ApplicationAttemptStateData in project hadoop by apache.

the class TestRMRestart method testRMRestartSucceededApp.

@Test(timeout = 60000)
public void testRMRestartSucceededApp() throws Exception {
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, YarnConfiguration.DEFAULT_RM_AM_MAX_ATTEMPTS);
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    RMState rmState = memStore.getState();
    Map<ApplicationId, ApplicationStateData> rmAppState = rmState.getApplicationState();
    // start RM
    MockRM rm1 = createMockRM(conf, memStore);
    rm1.start();
    MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm1.getResourceTrackerService());
    nm1.registerNode();
    // create an app and finish the app.
    RMApp app0 = rm1.submitApp(200);
    MockAM am0 = launchAM(app0, rm1, nm1);
    // unregister am
    FinishApplicationMasterRequest req = FinishApplicationMasterRequest.newInstance(FinalApplicationStatus.SUCCEEDED, "diagnostics", "trackingUrl");
    finishApplicationMaster(app0, rm1, nm1, am0, req);
    // check the state store about the unregistered info.
    ApplicationStateData appState = rmAppState.get(app0.getApplicationId());
    ApplicationAttemptStateData attemptState0 = appState.getAttempt(am0.getApplicationAttemptId());
    Assert.assertEquals("diagnostics", attemptState0.getDiagnostics());
    Assert.assertEquals(FinalApplicationStatus.SUCCEEDED, attemptState0.getFinalApplicationStatus());
    Assert.assertEquals("trackingUrl", attemptState0.getFinalTrackingUrl());
    Assert.assertEquals(app0.getFinishTime(), appState.getFinishTime());
    // restart rm
    MockRM rm2 = createMockRM(conf, memStore);
    rm2.start();
    // verify application report returns the same app info as the app info
    // before RM restarts.
    ApplicationReport appReport = verifyAppReportAfterRMRestart(app0, rm2);
    Assert.assertEquals(FinalApplicationStatus.SUCCEEDED, appReport.getFinalApplicationStatus());
    Assert.assertEquals("trackingUrl", appReport.getOriginalTrackingUrl());
}

Example 9 with ApplicationAttemptStateData

use of org.apache.hadoop.yarn.server.resourcemanager.recovery.records.ApplicationAttemptStateData 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 10 with ApplicationAttemptStateData

use of org.apache.hadoop.yarn.server.resourcemanager.recovery.records.ApplicationAttemptStateData in project hadoop by apache.

the class TestWorkPreservingRMRestart method testAppStateSavedButAttemptStateNotSaved.

// Test that if application state was saved, but attempt state was not saved.
// RM should start correctly.
@Test(timeout = 20000)
public void testAppStateSavedButAttemptStateNotSaved() throws Exception {
    MemoryRMStateStore memStore = new MemoryRMStateStore() {

        @Override
        public synchronized void updateApplicationAttemptStateInternal(ApplicationAttemptId appAttemptId, ApplicationAttemptStateData attemptState) {
        // do nothing;
        // simulate the failure that attempt final state is not saved.
        }
    };
    memStore.init(conf);
    rm1 = new MockRM(conf, memStore);
    rm1.start();
    MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm1.getResourceTrackerService());
    nm1.registerNode();
    RMApp app1 = rm1.submitApp(200);
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    MockRM.finishAMAndVerifyAppState(app1, rm1, nm1, am1);
    ApplicationStateData appSavedState = memStore.getState().getApplicationState().get(app1.getApplicationId());
    // check that app state is  saved.
    assertEquals(RMAppState.FINISHED, appSavedState.getState());
    // check that attempt state is not saved.
    assertNull(appSavedState.getAttempt(am1.getApplicationAttemptId()).getState());
    rm2 = new MockRM(conf, memStore);
    rm2.start();
    RMApp recoveredApp1 = rm2.getRMContext().getRMApps().get(app1.getApplicationId());
    assertEquals(RMAppState.FINISHED, recoveredApp1.getState());
    // check that attempt state is recovered correctly.
    assertEquals(RMAppAttemptState.FINISHED, recoveredApp1.getCurrentAppAttempt().getState());
}