use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt 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()));
}
use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.
the class TestRMApplicationHistoryWriter method createRMAppAttempt.
private static RMAppAttempt createRMAppAttempt(ApplicationAttemptId appAttemptId) {
RMAppAttempt appAttempt = mock(RMAppAttempt.class);
when(appAttempt.getAppAttemptId()).thenReturn(appAttemptId);
when(appAttempt.getHost()).thenReturn("test host");
when(appAttempt.getRpcPort()).thenReturn(-100);
Container container = mock(Container.class);
when(container.getId()).thenReturn(ContainerId.newContainerId(appAttemptId, 1));
when(appAttempt.getMasterContainer()).thenReturn(container);
when(appAttempt.getDiagnostics()).thenReturn("test diagnostics info");
when(appAttempt.getTrackingUrl()).thenReturn("test url");
when(appAttempt.getFinalApplicationStatus()).thenReturn(FinalApplicationStatus.UNDEFINED);
return appAttempt;
}
use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.
the class TestAMRMRPCNodeUpdates method testAMRMUnusableNodes.
@Test
public void testAMRMUnusableNodes() throws Exception {
MockNM nm1 = rm.registerNode("127.0.0.1:1234", 10000);
MockNM nm2 = rm.registerNode("127.0.0.2:1234", 10000);
MockNM nm3 = rm.registerNode("127.0.0.3:1234", 10000);
MockNM nm4 = rm.registerNode("127.0.0.4:1234", 10000);
dispatcher.await();
RMApp app1 = rm.submitApp(2000);
// Trigger the scheduling so the AM gets 'launched' on nm1
nm1.nodeHeartbeat(true);
RMAppAttempt attempt1 = app1.getCurrentAppAttempt();
MockAM am1 = rm.sendAMLaunched(attempt1.getAppAttemptId());
// register AM returns no unusable node
am1.registerAppAttempt();
// allocate request returns no updated node
AllocateRequest allocateRequest1 = AllocateRequest.newInstance(0, 0F, null, null, null);
AllocateResponse response1 = allocate(attempt1.getAppAttemptId(), allocateRequest1);
List<NodeReport> updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(0, updatedNodes.size());
syncNodeHeartbeat(nm4, false);
// allocate request returns updated node
allocateRequest1 = AllocateRequest.newInstance(response1.getResponseId(), 0F, null, null, null);
response1 = allocate(attempt1.getAppAttemptId(), allocateRequest1);
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
NodeReport nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.UNHEALTHY, nr.getNodeState());
// resending the allocate request returns the same result
response1 = allocate(attempt1.getAppAttemptId(), allocateRequest1);
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.UNHEALTHY, nr.getNodeState());
syncNodeLost(nm3);
// subsequent allocate request returns delta
allocateRequest1 = AllocateRequest.newInstance(response1.getResponseId(), 0F, null, null, null);
response1 = allocate(attempt1.getAppAttemptId(), allocateRequest1);
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm3.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.LOST, nr.getNodeState());
// registering another AM gives it the complete failed list
RMApp app2 = rm.submitApp(2000);
// Trigger nm2 heartbeat so that AM gets launched on it
nm2.nodeHeartbeat(true);
RMAppAttempt attempt2 = app2.getCurrentAppAttempt();
MockAM am2 = rm.sendAMLaunched(attempt2.getAppAttemptId());
// register AM returns all unusable nodes
am2.registerAppAttempt();
// allocate request returns no updated node
AllocateRequest allocateRequest2 = AllocateRequest.newInstance(0, 0F, null, null, null);
AllocateResponse response2 = allocate(attempt2.getAppAttemptId(), allocateRequest2);
updatedNodes = response2.getUpdatedNodes();
Assert.assertEquals(0, updatedNodes.size());
syncNodeHeartbeat(nm4, true);
// both AM's should get delta updated nodes
allocateRequest1 = AllocateRequest.newInstance(response1.getResponseId(), 0F, null, null, null);
response1 = allocate(attempt1.getAppAttemptId(), allocateRequest1);
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.RUNNING, nr.getNodeState());
allocateRequest2 = AllocateRequest.newInstance(response2.getResponseId(), 0F, null, null, null);
response2 = allocate(attempt2.getAppAttemptId(), allocateRequest2);
updatedNodes = response2.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.RUNNING, nr.getNodeState());
// subsequent allocate calls should return no updated nodes
allocateRequest2 = AllocateRequest.newInstance(response2.getResponseId(), 0F, null, null, null);
response2 = allocate(attempt2.getAppAttemptId(), allocateRequest2);
updatedNodes = response2.getUpdatedNodes();
Assert.assertEquals(0, updatedNodes.size());
// how to do the above for LOST node
}
use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.
the class TestAMRMRPCResponseId method testARRMResponseId.
@Test
public void testARRMResponseId() throws Exception {
MockNM nm1 = rm.registerNode("h1:1234", 5000);
RMApp app = rm.submitApp(2000);
// Trigger the scheduling so the AM gets 'launched'
nm1.nodeHeartbeat(true);
RMAppAttempt attempt = app.getCurrentAppAttempt();
MockAM am = rm.sendAMLaunched(attempt.getAppAttemptId());
am.registerAppAttempt();
AllocateRequest allocateRequest = AllocateRequest.newInstance(0, 0F, null, null, null);
AllocateResponse response = allocate(attempt.getAppAttemptId(), allocateRequest);
Assert.assertEquals(1, response.getResponseId());
Assert.assertTrue(response.getAMCommand() == null);
allocateRequest = AllocateRequest.newInstance(response.getResponseId(), 0F, null, null, null);
response = allocate(attempt.getAppAttemptId(), allocateRequest);
Assert.assertEquals(2, response.getResponseId());
/* try resending */
response = allocate(attempt.getAppAttemptId(), allocateRequest);
Assert.assertEquals(2, response.getResponseId());
/** try sending old request again **/
allocateRequest = AllocateRequest.newInstance(0, 0F, null, null, null);
try {
allocate(attempt.getAppAttemptId(), allocateRequest);
Assert.fail();
} catch (Exception e) {
Assert.assertTrue(e.getCause() instanceof InvalidApplicationMasterRequestException);
}
}
use of org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttempt in project hadoop by apache.
the class TestSystemMetricsPublisher method testPublishAppAttemptMetricsForUnmanagedAM.
@Test(timeout = 10000)
public void testPublishAppAttemptMetricsForUnmanagedAM() throws Exception {
ApplicationAttemptId appAttemptId = ApplicationAttemptId.newInstance(ApplicationId.newInstance(0, 1), 1);
RMAppAttempt appAttempt = createRMAppAttempt(appAttemptId, true);
metricsPublisher.appAttemptRegistered(appAttempt, Integer.MAX_VALUE + 1L);
RMApp app = mock(RMApp.class);
when(app.getFinalApplicationStatus()).thenReturn(FinalApplicationStatus.UNDEFINED);
metricsPublisher.appAttemptFinished(appAttempt, RMAppAttemptState.FINISHED, app, Integer.MAX_VALUE + 2L);
TimelineEntity entity = null;
do {
entity = store.getEntity(appAttemptId.toString(), AppAttemptMetricsConstants.ENTITY_TYPE, EnumSet.allOf(Field.class));
// ensure two events are both published before leaving the loop
} while (entity == null || entity.getEvents().size() < 2);
}
Aggregations