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

Example 11 with ApplicationStateData

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

the class TestRMRestart method finishApplicationMaster.

private void finishApplicationMaster(RMApp rmApp, MockRM rm, MockNM nm, MockAM am, FinishApplicationMasterRequest req) throws Exception {
    RMState rmState = ((MemoryRMStateStore) rm.getRMContext().getStateStore()).getState();
    Map<ApplicationId, ApplicationStateData> rmAppState = rmState.getApplicationState();
    am.unregisterAppAttempt(req, true);
    rm.waitForState(am.getApplicationAttemptId(), RMAppAttemptState.FINISHING);
    nm.nodeHeartbeat(am.getApplicationAttemptId(), 1, ContainerState.COMPLETE);
    rm.waitForState(am.getApplicationAttemptId(), RMAppAttemptState.FINISHED);
    rm.waitForState(rmApp.getApplicationId(), RMAppState.FINISHED);
    // check that app/attempt is saved with the final state
    ApplicationStateData appState = rmAppState.get(rmApp.getApplicationId());
    Assert.assertEquals(RMAppState.FINISHED, appState.getState());
    Assert.assertEquals(RMAppAttemptState.FINISHED, appState.getAttempt(am.getApplicationAttemptId()).getState());
}

Example 12 with ApplicationStateData

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

the class TestRMRestart method testRMRestartOnMissingAttempts.

@Test(timeout = 60000)
public void testRMRestartOnMissingAttempts() throws Exception {
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 5);
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    // 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);
    ApplicationStateData app0State = memStore.getState().getApplicationState().get(app0.getApplicationId());
    MockAM am0 = launchAndFailAM(app0, rm1, nm1);
    MockAM am1 = launchAndFailAM(app0, rm1, nm1);
    MockAM am2 = launchAndFailAM(app0, rm1, nm1);
    MockAM am3 = launchAM(app0, rm1, nm1);
    // am1 is missed from MemoryRMStateStore
    memStore.removeApplicationAttemptInternal(am1.getApplicationAttemptId());
    ApplicationAttemptStateData am2State = app0State.getAttempt(am2.getApplicationAttemptId());
    // am2's state is not consistent: MemoryRMStateStore just saved its initial
    // state and failed to store its final state
    am2State.setState(null);
    // restart rm
    MockRM rm2 = createMockRM(conf, memStore);
    rm2.start();
    Assert.assertEquals(1, rm2.getRMContext().getRMApps().size());
    RMApp recoveredApp0 = rm2.getRMContext().getRMApps().values().iterator().next();
    Map<ApplicationAttemptId, RMAppAttempt> recoveredAppAttempts = recoveredApp0.getAppAttempts();
    Assert.assertEquals(3, recoveredAppAttempts.size());
    Assert.assertEquals(RMAppAttemptState.FAILED, recoveredAppAttempts.get(am0.getApplicationAttemptId()).getAppAttemptState());
    Assert.assertEquals(RMAppAttemptState.FAILED, recoveredAppAttempts.get(am2.getApplicationAttemptId()).getAppAttemptState());
    Assert.assertEquals(RMAppAttemptState.LAUNCHED, recoveredAppAttempts.get(am3.getApplicationAttemptId()).getAppAttemptState());
    Assert.assertEquals(5, ((RMAppImpl) app0).getNextAttemptId());
}

Example 13 with ApplicationStateData

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

the class TestRMRestart method testRMRestartWaitForPreviousSucceededAttempt.

// Test RM restarts after previous attempt succeeded and was saved into state
// store but before the RMAppAttempt notifies RMApp that it has succeeded. On
// recovery, RMAppAttempt should send the AttemptFinished event to RMApp so
// that RMApp can recover its state.
@Test(timeout = 60000)
public void testRMRestartWaitForPreviousSucceededAttempt() throws Exception {
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 2);
    MemoryRMStateStore memStore = new MemoryRMStateStore() {

        int count = 0;

        @Override
        public void updateApplicationStateInternal(ApplicationId appId, ApplicationStateData appStateData) throws Exception {
            if (count == 0) {
                // do nothing; simulate app final state is not saved.
                LOG.info(appId + " final state is not saved.");
                count++;
            } else {
                super.updateApplicationStateInternal(appId, appStateData);
            }
        }
    };
    memStore.init(conf);
    RMState rmState = memStore.getState();
    Map<ApplicationId, ApplicationStateData> rmAppState = rmState.getApplicationState();
    // start RM
    MockRM rm1 = createMockRM(conf, memStore);
    rm1.start();
    MockNM nm1 = rm1.registerNode("127.0.0.1:1234", 15120);
    RMApp app0 = rm1.submitApp(200);
    MockAM am0 = MockRM.launchAndRegisterAM(app0, rm1, nm1);
    FinishApplicationMasterRequest req = FinishApplicationMasterRequest.newInstance(FinalApplicationStatus.SUCCEEDED, "", "");
    am0.unregisterAppAttempt(req, true);
    rm1.waitForState(am0.getApplicationAttemptId(), RMAppAttemptState.FINISHING);
    // app final state is not saved. This guarantees that RMApp cannot be
    // recovered via its own saved state, but only via the event notification
    // from the RMAppAttempt on recovery.
    Assert.assertNull(rmAppState.get(app0.getApplicationId()).getState());
    // start RM
    MockRM rm2 = createMockRM(conf, memStore);
    nm1.setResourceTrackerService(rm2.getResourceTrackerService());
    rm2.start();
    rm2.waitForState(app0.getCurrentAppAttempt().getAppAttemptId(), RMAppAttemptState.FINISHED);
    rm2.waitForState(app0.getApplicationId(), RMAppState.FINISHED);
    // app final state is saved via the finish event from attempt.
    Assert.assertEquals(RMAppState.FINISHED, rmAppState.get(app0.getApplicationId()).getState());
}

Example 14 with ApplicationStateData

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

the class TestRMRestart method testRMRestartTimelineCollectorContext.

@Test(timeout = 60000)
public void testRMRestartTimelineCollectorContext() throws Exception {
    conf.setBoolean(YarnConfiguration.TIMELINE_SERVICE_ENABLED, true);
    conf.setFloat(YarnConfiguration.TIMELINE_SERVICE_VERSION, 2.0f);
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    RMState rmState = memStore.getState();
    Map<ApplicationId, ApplicationStateData> rmAppState = rmState.getApplicationState();
    MockRM rm1 = null;
    MockRM rm2 = null;
    try {
        rm1 = createMockRM(conf, memStore);
        rm1.start();
        MockNM nm1 = new MockNM("127.0.0.1:1234", 15120, rm1.getResourceTrackerService());
        nm1.registerNode();
        // submit an app.
        RMApp app = rm1.submitApp(200, "name", "user", new HashMap<ApplicationAccessType, String>(), false, "default", -1, null);
        // Check if app info has been saved.
        ApplicationStateData appState = rmAppState.get(app.getApplicationId());
        Assert.assertNotNull(appState);
        Assert.assertEquals(0, appState.getAttemptCount());
        Assert.assertEquals(appState.getApplicationSubmissionContext().getApplicationId(), app.getApplicationSubmissionContext().getApplicationId());
        // Allocate the AM
        nm1.nodeHeartbeat(true);
        RMAppAttempt attempt = app.getCurrentAppAttempt();
        ApplicationAttemptId attemptId1 = attempt.getAppAttemptId();
        rm1.waitForState(attemptId1, RMAppAttemptState.ALLOCATED);
        ApplicationId appId = app.getApplicationId();
        TimelineCollectorContext contextBeforeRestart = rm1.getRMContext().getRMTimelineCollectorManager().get(appId).getTimelineEntityContext();
        // Restart RM.
        rm2 = createMockRM(conf, memStore);
        rm2.start();
        Assert.assertEquals(1, rm2.getRMContext().getRMApps().size());
        rm2.waitForState(app.getApplicationId(), RMAppState.ACCEPTED);
        TimelineCollectorContext contextAfterRestart = rm2.getRMContext().getRMTimelineCollectorManager().get(appId).getTimelineEntityContext();
        Assert.assertEquals("Collector contexts for an app should be same " + "across restarts", contextBeforeRestart, contextAfterRestart);
    } finally {
        conf.setBoolean(YarnConfiguration.TIMELINE_SERVICE_ENABLED, false);
        if (rm1 != null) {
            rm1.close();
        }
        if (rm2 != null) {
            rm2.close();
        }
    }
}

Example 15 with ApplicationStateData

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

the class TestRMRestart method testRMDelegationTokenRestoredOnRMRestart.

@Test(timeout = 60000)
public void testRMDelegationTokenRestoredOnRMRestart() throws Exception {
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 2);
    conf.set(CommonConfigurationKeysPublic.HADOOP_SECURITY_AUTHENTICATION, "kerberos");
    conf.set(YarnConfiguration.RM_ADDRESS, "localhost:8032");
    UserGroupInformation.setConfiguration(conf);
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    RMState rmState = memStore.getState();
    Map<ApplicationId, ApplicationStateData> rmAppState = rmState.getApplicationState();
    Map<RMDelegationTokenIdentifier, Long> rmDTState = rmState.getRMDTSecretManagerState().getTokenState();
    Set<DelegationKey> rmDTMasterKeyState = rmState.getRMDTSecretManagerState().getMasterKeyState();
    MockRM rm1 = new TestSecurityMockRM(conf, memStore);
    rm1.start();
    // create an empty credential
    Credentials ts = new Credentials();
    // request a token and add into credential
    GetDelegationTokenRequest request1 = GetDelegationTokenRequest.newInstance("renewer1");
    UserGroupInformation.getCurrentUser().setAuthenticationMethod(AuthMethod.KERBEROS);
    GetDelegationTokenResponse response1 = rm1.getClientRMService().getDelegationToken(request1);
    org.apache.hadoop.yarn.api.records.Token delegationToken1 = response1.getRMDelegationToken();
    Token<RMDelegationTokenIdentifier> token1 = ConverterUtils.convertFromYarn(delegationToken1, rmAddr);
    RMDelegationTokenIdentifier dtId1 = token1.decodeIdentifier();
    HashSet<RMDelegationTokenIdentifier> tokenIdentSet = new HashSet<RMDelegationTokenIdentifier>();
    ts.addToken(token1.getService(), token1);
    tokenIdentSet.add(dtId1);
    // submit an app with customized credential
    RMApp app = rm1.submitApp(200, "name", "user", new HashMap<ApplicationAccessType, String>(), false, "default", 1, ts);
    // assert app info is saved
    ApplicationStateData appState = rmAppState.get(app.getApplicationId());
    Assert.assertNotNull(appState);
    // assert all master keys are saved
    Set<DelegationKey> allKeysRM1 = rm1.getRMContext().getRMDelegationTokenSecretManager().getAllMasterKeys();
    Assert.assertEquals(allKeysRM1, rmDTMasterKeyState);
    // assert all tokens are saved
    Map<RMDelegationTokenIdentifier, Long> allTokensRM1 = rm1.getRMContext().getRMDelegationTokenSecretManager().getAllTokens();
    Assert.assertEquals(tokenIdentSet, allTokensRM1.keySet());
    Assert.assertEquals(allTokensRM1, rmDTState);
    // assert sequence number is saved
    Assert.assertEquals(rm1.getRMContext().getRMDelegationTokenSecretManager().getLatestDTSequenceNumber(), rmState.getRMDTSecretManagerState().getDTSequenceNumber());
    // request one more token
    GetDelegationTokenRequest request2 = GetDelegationTokenRequest.newInstance("renewer2");
    GetDelegationTokenResponse response2 = rm1.getClientRMService().getDelegationToken(request2);
    org.apache.hadoop.yarn.api.records.Token delegationToken2 = response2.getRMDelegationToken();
    Token<RMDelegationTokenIdentifier> token2 = ConverterUtils.convertFromYarn(delegationToken2, rmAddr);
    RMDelegationTokenIdentifier dtId2 = token2.decodeIdentifier();
    // cancel token2
    try {
        rm1.getRMContext().getRMDelegationTokenSecretManager().cancelToken(token2, UserGroupInformation.getCurrentUser().getUserName());
    } catch (Exception e) {
        Assert.fail();
    }
    // Assert the token which has the latest delegationTokenSequenceNumber is removed
    Assert.assertEquals(rm1.getRMContext().getRMDelegationTokenSecretManager().getLatestDTSequenceNumber(), dtId2.getSequenceNumber());
    Assert.assertFalse(rmDTState.containsKey(dtId2));
    // start new RM
    MockRM rm2 = new TestSecurityMockRM(conf, memStore);
    rm2.start();
    // assert master keys and tokens are populated back to DTSecretManager
    Map<RMDelegationTokenIdentifier, Long> allTokensRM2 = rm2.getRMContext().getRMDelegationTokenSecretManager().getAllTokens();
    Assert.assertEquals(allTokensRM2.keySet(), allTokensRM1.keySet());
    // rm2 has its own master keys when it starts, we use containsAll here
    Assert.assertTrue(rm2.getRMContext().getRMDelegationTokenSecretManager().getAllMasterKeys().containsAll(allKeysRM1));
    // assert sequenceNumber is properly recovered,
    // even though the token which has max sequenceNumber is not stored
    Assert.assertEquals(rm1.getRMContext().getRMDelegationTokenSecretManager().getLatestDTSequenceNumber(), rm2.getRMContext().getRMDelegationTokenSecretManager().getLatestDTSequenceNumber());
    // renewDate before renewing
    Long renewDateBeforeRenew = allTokensRM2.get(dtId1);
    try {
        // Sleep for one millisecond to make sure renewDataAfterRenew is greater
        Thread.sleep(1);
        // renew recovered token
        rm2.getRMContext().getRMDelegationTokenSecretManager().renewToken(token1, "renewer1");
    } catch (Exception e) {
        Assert.fail();
    }
    allTokensRM2 = rm2.getRMContext().getRMDelegationTokenSecretManager().getAllTokens();
    Long renewDateAfterRenew = allTokensRM2.get(dtId1);
    // assert token is renewed
    Assert.assertTrue(renewDateAfterRenew > renewDateBeforeRenew);
    // assert new token is added into state store
    Assert.assertTrue(rmDTState.containsValue(renewDateAfterRenew));
    // assert old token is removed from state store
    Assert.assertFalse(rmDTState.containsValue(renewDateBeforeRenew));
    try {
        rm2.getRMContext().getRMDelegationTokenSecretManager().cancelToken(token1, UserGroupInformation.getCurrentUser().getUserName());
    } catch (Exception e) {
        Assert.fail();
    }
    // assert token is removed from state after its cancelled
    allTokensRM2 = rm2.getRMContext().getRMDelegationTokenSecretManager().getAllTokens();
    Assert.assertFalse(allTokensRM2.containsKey(dtId1));
    Assert.assertFalse(rmDTState.containsKey(dtId1));
}