Examples with RMState org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore.RMState used on opensource projects

Example 31 with RMState

use of org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore.RMState in project hadoop by apache.

the class TestRMRestart method testRMRestartFailedApp.

@Test(timeout = 60000)
public void testRMRestartFailedApp() throws Exception {
    conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 1);
    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 app and launch the AM
    RMApp app0 = rm1.submitApp(200);
    MockAM am0 = launchAM(app0, rm1, nm1);
    // fail the AM by sending CONTAINER_FINISHED event without registering.
    nm1.nodeHeartbeat(am0.getApplicationAttemptId(), 1, ContainerState.COMPLETE);
    rm1.waitForState(am0.getApplicationAttemptId(), RMAppAttemptState.FAILED);
    rm1.waitForState(app0.getApplicationId(), RMAppState.FAILED);
    // assert the app/attempt failed state is saved.
    ApplicationStateData appState = rmAppState.get(app0.getApplicationId());
    Assert.assertEquals(RMAppState.FAILED, appState.getState());
    Assert.assertEquals(RMAppAttemptState.FAILED, appState.getAttempt(am0.getApplicationAttemptId()).getState());
    // start new RM
    MockRM rm2 = createMockRM(conf, memStore);
    rm2.start();
    RMApp loadedApp0 = rm2.getRMContext().getRMApps().get(app0.getApplicationId());
    rm2.waitForState(app0.getApplicationId(), RMAppState.FAILED);
    rm2.waitForState(am0.getApplicationAttemptId(), RMAppAttemptState.FAILED);
    // no new attempt is created.
    Assert.assertEquals(1, loadedApp0.getAppAttempts().size());
    verifyAppReportAfterRMRestart(app0, rm2);
    Assert.assertTrue(app0.getDiagnostics().toString().contains("Failing the application."));
// failed diagnostics from attempt is lost because the diagnostics from
// attempt is not yet available by the time app is saving the app state.
}