Examples with MockNM org.apache.hadoop.yarn.server.resourcemanager.MockNM used on opensource projects

Example 21 with MockNM

use of org.apache.hadoop.yarn.server.resourcemanager.MockNM in project hadoop by apache.

the class TestFairScheduler method testRefreshQueuesWhenRMHA.

@Test(timeout = 120000)
public void testRefreshQueuesWhenRMHA() throws Exception {
    conf.setBoolean(YarnConfiguration.AUTO_FAILOVER_ENABLED, false);
    conf.setBoolean(YarnConfiguration.RECOVERY_ENABLED, true);
    conf.setBoolean(FairSchedulerConfiguration.ALLOW_UNDECLARED_POOLS, false);
    conf.setBoolean(FairSchedulerConfiguration.USER_AS_DEFAULT_QUEUE, false);
    conf.set(FairSchedulerConfiguration.ALLOCATION_FILE, ALLOC_FILE);
    HAServiceProtocol.StateChangeRequestInfo requestInfo = new HAServiceProtocol.StateChangeRequestInfo(HAServiceProtocol.RequestSource.REQUEST_BY_USER);
    // 1. start a standby RM, file 'ALLOC_FILE' is empty, so there is no queues
    MockRM rm1 = new MockRM(conf, null);
    rm1.init(conf);
    rm1.start();
    rm1.getAdminService().transitionToStandby(requestInfo);
    // 2. add a new queue "test_queue"
    PrintWriter out = new PrintWriter(new FileWriter(ALLOC_FILE));
    out.println("<?xml version=\"1.0\"?>");
    out.println("<allocations>");
    out.println("<queue name=\"test_queue\">");
    out.println("  <maxRunningApps>3</maxRunningApps>");
    out.println("</queue>");
    out.println("</allocations>");
    out.close();
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    // 3. start a active RM
    MockRM rm2 = new MockRM(conf, memStore);
    rm2.init(conf);
    rm2.start();
    MockNM nm = new MockNM("127.0.0.1:1234", 15120, rm2.getResourceTrackerService());
    nm.registerNode();
    rm2.getAdminService().transitionToActive(requestInfo);
    // 4. submit a app to the new added queue "test_queue"
    RMApp app = rm2.submitApp(200, "test_app", "user", null, "test_queue");
    RMAppAttempt attempt0 = app.getCurrentAppAttempt();
    nm.nodeHeartbeat(true);
    MockAM am0 = rm2.sendAMLaunched(attempt0.getAppAttemptId());
    am0.registerAppAttempt();
    assertEquals("root.test_queue", app.getQueue());
    // 5. Transit rm1 to active, recover app
    ((RMContextImpl) rm1.getRMContext()).setStateStore(memStore);
    rm1.getAdminService().transitionToActive(requestInfo);
    rm1.drainEvents();
    assertEquals(1, rm1.getRMContext().getRMApps().size());
    RMApp recoveredApp = rm1.getRMContext().getRMApps().values().iterator().next();
    assertEquals("root.test_queue", recoveredApp.getQueue());
    rm1.stop();
    rm2.stop();
}

Example 22 with MockNM

use of org.apache.hadoop.yarn.server.resourcemanager.MockNM in project hadoop by apache.

the class TestWorkPreservingRMRestartForNodeLabel method testWorkPreservingRestartForNodeLabel.

@Test
public void testWorkPreservingRestartForNodeLabel() throws Exception {
    // This test is pretty much similar to testContainerAllocateWithLabel.
    // Difference is, this test doesn't specify label expression in ResourceRequest,
    // instead, it uses default queue label expression
    // set node -> label
    mgr.addToCluserNodeLabelsWithDefaultExclusivity(ImmutableSet.of("x", "y"));
    mgr.addLabelsToNode(ImmutableMap.of(NodeId.newInstance("h1", 0), toSet("x"), NodeId.newInstance("h2", 0), toSet("y")));
    MemoryRMStateStore memStore = new MemoryRMStateStore();
    memStore.init(conf);
    conf = TestUtils.getConfigurationWithDefaultQueueLabels(conf);
    // inject node label manager
    MockRM rm1 = new MockRM(conf, memStore) {

        @Override
        public RMNodeLabelsManager createNodeLabelManager() {
            return mgr;
        }
    };
    rm1.getRMContext().setNodeLabelManager(mgr);
    rm1.start();
    // label = x
    MockNM nm1 = rm1.registerNode("h1:1234", 8000);
    // label = y
    MockNM nm2 = rm1.registerNode("h2:1234", 8000);
    // label = <empty>
    MockNM nm3 = rm1.registerNode("h3:1234", 8000);
    ContainerId containerId;
    // launch an app to queue a1 (label = x), and check all container will
    // be allocated in h1
    RMApp app1 = rm1.submitApp(200, "app", "user", null, "a1");
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    // request a container.
    am1.allocate("*", 1024, 1, new ArrayList<ContainerId>());
    containerId = ContainerId.newContainerId(am1.getApplicationAttemptId(), 2);
    Assert.assertTrue(rm1.waitForState(nm1, containerId, RMContainerState.ALLOCATED));
    checkRMContainerLabelExpression(ContainerId.newContainerId(am1.getApplicationAttemptId(), 1), rm1, "x");
    checkRMContainerLabelExpression(ContainerId.newContainerId(am1.getApplicationAttemptId(), 2), rm1, "x");
    // launch an app to queue b1 (label = y), and check all container will
    // be allocated in h2
    RMApp app2 = rm1.submitApp(200, "app", "user", null, "b1");
    MockAM am2 = MockRM.launchAndRegisterAM(app2, rm1, nm2);
    // request a container.
    am2.allocate("*", 1024, 1, new ArrayList<ContainerId>());
    containerId = ContainerId.newContainerId(am2.getApplicationAttemptId(), 2);
    Assert.assertTrue(rm1.waitForState(nm2, containerId, RMContainerState.ALLOCATED));
    checkRMContainerLabelExpression(ContainerId.newContainerId(am2.getApplicationAttemptId(), 1), rm1, "y");
    checkRMContainerLabelExpression(ContainerId.newContainerId(am2.getApplicationAttemptId(), 2), rm1, "y");
    // launch an app to queue c1 (label = ""), and check all container will
    // be allocated in h3
    RMApp app3 = rm1.submitApp(200, "app", "user", null, "c1");
    MockAM am3 = MockRM.launchAndRegisterAM(app3, rm1, nm3);
    // request a container.
    am3.allocate("*", 1024, 1, new ArrayList<ContainerId>());
    containerId = ContainerId.newContainerId(am3.getApplicationAttemptId(), 2);
    Assert.assertTrue(rm1.waitForState(nm3, containerId, RMContainerState.ALLOCATED));
    checkRMContainerLabelExpression(ContainerId.newContainerId(am3.getApplicationAttemptId(), 1), rm1, "");
    checkRMContainerLabelExpression(ContainerId.newContainerId(am3.getApplicationAttemptId(), 2), rm1, "");
    // Re-start RM
    mgr = new NullRMNodeLabelsManager();
    mgr.init(conf);
    mgr.addToCluserNodeLabelsWithDefaultExclusivity(ImmutableSet.of("x", "y"));
    mgr.addLabelsToNode(ImmutableMap.of(NodeId.newInstance("h1", 0), toSet("x"), NodeId.newInstance("h2", 0), toSet("y")));
    MockRM rm2 = new MockRM(conf, memStore) {

        @Override
        public RMNodeLabelsManager createNodeLabelManager() {
            return mgr;
        }
    };
    rm2.start();
    nm1.setResourceTrackerService(rm2.getResourceTrackerService());
    nm2.setResourceTrackerService(rm2.getResourceTrackerService());
    nm3.setResourceTrackerService(rm2.getResourceTrackerService());
    // recover app
    NMContainerStatus app1c1 = TestRMRestart.createNMContainerStatus(am1.getApplicationAttemptId(), 1, ContainerState.RUNNING, "x");
    NMContainerStatus app1c2 = TestRMRestart.createNMContainerStatus(am1.getApplicationAttemptId(), 2, ContainerState.RUNNING, "x");
    nm1.registerNode(Arrays.asList(app1c1, app1c2), null);
    waitForNumContainersToRecover(2, rm2, am1.getApplicationAttemptId());
    checkRMContainerLabelExpression(ContainerId.newContainerId(am1.getApplicationAttemptId(), 1), rm1, "x");
    checkRMContainerLabelExpression(ContainerId.newContainerId(am1.getApplicationAttemptId(), 2), rm1, "x");
    NMContainerStatus app2c1 = TestRMRestart.createNMContainerStatus(am2.getApplicationAttemptId(), 1, ContainerState.RUNNING, "y");
    NMContainerStatus app2c2 = TestRMRestart.createNMContainerStatus(am2.getApplicationAttemptId(), 2, ContainerState.RUNNING, "y");
    nm2.registerNode(Arrays.asList(app2c1, app2c2), null);
    waitForNumContainersToRecover(2, rm2, am2.getApplicationAttemptId());
    checkRMContainerLabelExpression(ContainerId.newContainerId(am2.getApplicationAttemptId(), 1), rm1, "y");
    checkRMContainerLabelExpression(ContainerId.newContainerId(am2.getApplicationAttemptId(), 2), rm1, "y");
    NMContainerStatus app3c1 = TestRMRestart.createNMContainerStatus(am3.getApplicationAttemptId(), 1, ContainerState.RUNNING, "");
    NMContainerStatus app3c2 = TestRMRestart.createNMContainerStatus(am3.getApplicationAttemptId(), 2, ContainerState.RUNNING, "");
    nm3.registerNode(Arrays.asList(app3c1, app3c2), null);
    waitForNumContainersToRecover(2, rm2, am3.getApplicationAttemptId());
    checkRMContainerLabelExpression(ContainerId.newContainerId(am3.getApplicationAttemptId(), 1), rm1, "");
    checkRMContainerLabelExpression(ContainerId.newContainerId(am3.getApplicationAttemptId(), 2), rm1, "");
    // Check recovered resource usage
    checkAppResourceUsage("x", app1.getApplicationId(), rm1, 2 * GB);
    checkAppResourceUsage("y", app2.getApplicationId(), rm1, 2 * GB);
    checkAppResourceUsage("", app3.getApplicationId(), rm1, 2 * GB);
    checkQueueResourceUsage("x", "a1", rm1, 2 * GB);
    checkQueueResourceUsage("y", "b1", rm1, 2 * GB);
    checkQueueResourceUsage("", "c1", rm1, 2 * GB);
    checkQueueResourceUsage("x", "a", rm1, 2 * GB);
    checkQueueResourceUsage("y", "b", rm1, 2 * GB);
    checkQueueResourceUsage("", "c", rm1, 2 * GB);
    checkQueueResourceUsage("x", "root", rm1, 2 * GB);
    checkQueueResourceUsage("y", "root", rm1, 2 * GB);
    checkQueueResourceUsage("", "root", rm1, 2 * GB);
    rm1.close();
    rm2.close();
}

Example 23 with MockNM

use of org.apache.hadoop.yarn.server.resourcemanager.MockNM in project hadoop by apache.

the class TestRMWebServicesSchedulerActivities method testAssignWithoutAvailableResource.

@Test
public void testAssignWithoutAvailableResource() throws Exception {
    //Start RM so that it accepts app submissions
    rm.start();
    MockNM nm = new MockNM("127.0.0.1:1234", 1 * 1024, rm.getResourceTrackerService());
    nm.registerNode();
    try {
        RMApp app1 = rm.submitApp(1024, "app1", "user1", null, "b1");
        MockAM am1 = MockRM.launchAndRegisterAM(app1, rm, nm);
        am1.allocate(Arrays.asList(ResourceRequest.newInstance(Priority.UNDEFINED, "127.0.0.1", Resources.createResource(1024), 10), ResourceRequest.newInstance(Priority.UNDEFINED, "/default-rack", Resources.createResource(1024), 10), ResourceRequest.newInstance(Priority.UNDEFINED, "*", Resources.createResource(1024), 10)), null);
        //Get JSON
        WebResource r = resource();
        MultivaluedMapImpl params = new MultivaluedMapImpl();
        params.add("nodeId", "127.0.0.1");
        ClientResponse response = r.path("ws").path("v1").path("cluster").path("scheduler/activities").queryParams(params).accept(MediaType.APPLICATION_JSON).get(ClientResponse.class);
        assertEquals(MediaType.APPLICATION_JSON_TYPE + "; " + JettyUtils.UTF_8, response.getType().toString());
        JSONObject json = response.getEntity(JSONObject.class);
        nm.nodeHeartbeat(true);
        Thread.sleep(1000);
        //Get JSON
        response = r.path("ws").path("v1").path("cluster").path("scheduler/activities").queryParams(params).accept(MediaType.APPLICATION_JSON).get(ClientResponse.class);
        assertEquals(MediaType.APPLICATION_JSON_TYPE + "; " + JettyUtils.UTF_8, response.getType().toString());
        json = response.getEntity(JSONObject.class);
        verifyNumberOfAllocations(json, 0);
    } finally {
        rm.stop();
    }
}

Example 24 with MockNM

use of org.apache.hadoop.yarn.server.resourcemanager.MockNM in project hadoop by apache.

the class TestCapacitySchedulerSurgicalPreemption method testPriorityPreemptionWhenAllQueuesAreBelowGuaranteedCapacities.

@Test(timeout = 60000)
public void testPriorityPreemptionWhenAllQueuesAreBelowGuaranteedCapacities() throws Exception {
    /**
     * Test case: Submit two application (app1/app2) to different queues, queue
     * structure:
     *
     * <pre>
     *             Root
     *            /  |  \
     *           a   b   c
     *          10   20  70
     * </pre>
     *
     * 1) Two nodes (n1/n2) in the cluster, each of them has 20G.
     *
     * 2) app1 submit to queue-b first, it asked 6 * 1G containers
     * We will allocate 4 on n1 (including AM) and 3 on n2.
     *
     * 3) app2 submit to queue-c, ask for one 18G container (for AM)
     *
     * After preemption, we should expect:
     * Preempt 3 containers from app1 and AM of app2 successfully allocated.
     */
    conf.setPUOrderingPolicyUnderUtilizedPreemptionEnabled(true);
    conf.setPUOrderingPolicyUnderUtilizedPreemptionDelay(1000);
    conf.setQueueOrderingPolicy(CapacitySchedulerConfiguration.ROOT, CapacitySchedulerConfiguration.QUEUE_PRIORITY_UTILIZATION_ORDERING_POLICY);
    // Queue c has higher priority than a/b
    conf.setQueuePriority(CapacitySchedulerConfiguration.ROOT + ".c", 1);
    MockRM rm1 = new MockRM(conf);
    rm1.getRMContext().setNodeLabelManager(mgr);
    rm1.start();
    MockNM nm1 = rm1.registerNode("h1:1234", 20 * GB);
    MockNM nm2 = rm1.registerNode("h2:1234", 20 * GB);
    CapacityScheduler cs = (CapacityScheduler) rm1.getResourceScheduler();
    RMNode rmNode1 = rm1.getRMContext().getRMNodes().get(nm1.getNodeId());
    RMNode rmNode2 = rm1.getRMContext().getRMNodes().get(nm2.getNodeId());
    // launch an app to queue, AM container should be launched in nm1
    RMApp app1 = rm1.submitApp(1 * GB, "app", "user", null, "b");
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    am1.allocate("*", 1 * GB, 6, new ArrayList<>());
    // Do allocation for node1/node2
    for (int i = 0; i < 3; i++) {
        cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
        cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
    }
    // App1 should have 7 containers now, so the abs-used-cap of b is
    // 7 / 40 = 17.5% < 20% (guaranteed)
    FiCaSchedulerApp schedulerApp1 = cs.getApplicationAttempt(am1.getApplicationAttemptId());
    Assert.assertEquals(7, schedulerApp1.getLiveContainers().size());
    // 4 from n1 and 3 from n2
    waitNumberOfLiveContainersOnNodeFromApp(cs.getNode(rmNode1.getNodeID()), am1.getApplicationAttemptId(), 4);
    waitNumberOfLiveContainersOnNodeFromApp(cs.getNode(rmNode2.getNodeID()), am1.getApplicationAttemptId(), 3);
    // Submit app2 to queue-c and asks for a 1G container for AM
    RMApp app2 = rm1.submitApp(18 * GB, "app", "user", null, "c");
    FiCaSchedulerApp schedulerApp2 = cs.getApplicationAttempt(ApplicationAttemptId.newInstance(app2.getApplicationId(), 1));
    while (cs.getNode(rmNode1.getNodeID()).getReservedContainer() == null) {
        cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
        Thread.sleep(10);
    }
    // Call editSchedule immediately: containers are not selected
    ProportionalCapacityPreemptionPolicy editPolicy = (ProportionalCapacityPreemptionPolicy) getSchedulingEditPolicy(rm1);
    editPolicy.editSchedule();
    Assert.assertEquals(0, editPolicy.getToPreemptContainers().size());
    // Sleep the timeout interval, we should be able to see containers selected
    Thread.sleep(1000);
    editPolicy.editSchedule();
    Assert.assertEquals(2, editPolicy.getToPreemptContainers().size());
    // Call editSchedule again: selected containers are killed, and new AM
    // container launched
    editPolicy.editSchedule();
    // Do allocation till reserved container allocated
    while (cs.getNode(rmNode1.getNodeID()).getReservedContainer() != null) {
        cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
        Thread.sleep(10);
    }
    waitNumberOfLiveContainersFromApp(schedulerApp2, 1);
    rm1.close();
}

Example 25 with MockNM

use of org.apache.hadoop.yarn.server.resourcemanager.MockNM in project hadoop by apache.

the class TestCapacitySchedulerSurgicalPreemption method testSurgicalPreemptionWithAvailableResource.

@Test(timeout = 60000)
public void testSurgicalPreemptionWithAvailableResource() throws Exception {
    /**
     * Test case: Submit two application (app1/app2) to different queues, queue
     * structure:
     *
     * <pre>
     *             Root
     *            /  |  \
     *           a   b   c
     *          10   20  70
     * </pre>
     *
     * 1) Two nodes (n1/n2) in the cluster, each of them has 20G.
     *
     * 2) app1 submit to queue-b, asks for 1G * 5
     *
     * 3) app2 submit to queue-c, ask for one 4G container (for AM)
     *
     * After preemption, we should expect:
     * Preempt 3 containers from app1 and AM of app2 successfully allocated.
     */
    MockRM rm1 = new MockRM(conf);
    rm1.getRMContext().setNodeLabelManager(mgr);
    rm1.start();
    MockNM nm1 = rm1.registerNode("h1:1234", 20 * GB);
    MockNM nm2 = rm1.registerNode("h2:1234", 20 * GB);
    CapacityScheduler cs = (CapacityScheduler) rm1.getResourceScheduler();
    RMNode rmNode1 = rm1.getRMContext().getRMNodes().get(nm1.getNodeId());
    RMNode rmNode2 = rm1.getRMContext().getRMNodes().get(nm2.getNodeId());
    // launch an app to queue, AM container should be launched in nm1
    RMApp app1 = rm1.submitApp(1 * GB, "app", "user", null, "a");
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    am1.allocate("*", 1 * GB, 38, new ArrayList<ContainerId>());
    // Do allocation for node1/node2
    for (int i = 0; i < 38; i++) {
        cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
        cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
    }
    // App1 should have 31 containers now
    FiCaSchedulerApp schedulerApp1 = cs.getApplicationAttempt(am1.getApplicationAttemptId());
    Assert.assertEquals(39, schedulerApp1.getLiveContainers().size());
    // 17 from n1 and 16 from n2
    waitNumberOfLiveContainersOnNodeFromApp(cs.getNode(rmNode1.getNodeID()), am1.getApplicationAttemptId(), 20);
    waitNumberOfLiveContainersOnNodeFromApp(cs.getNode(rmNode2.getNodeID()), am1.getApplicationAttemptId(), 19);
    // Submit app2 to queue-c and asks for a 4G container for AM
    RMApp app2 = rm1.submitApp(4 * GB, "app", "user", null, "c");
    FiCaSchedulerApp schedulerApp2 = cs.getApplicationAttempt(ApplicationAttemptId.newInstance(app2.getApplicationId(), 1));
    // Call editSchedule: containers are selected to be preemption candidate
    ProportionalCapacityPreemptionPolicy editPolicy = (ProportionalCapacityPreemptionPolicy) getSchedulingEditPolicy(rm1);
    editPolicy.editSchedule();
    Assert.assertEquals(3, editPolicy.getToPreemptContainers().size());
    // Call editSchedule again: selected containers are killed
    editPolicy.editSchedule();
    waitNumberOfLiveContainersFromApp(schedulerApp1, 36);
    // Call allocation, containers are reserved
    cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
    cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
    waitNumberOfReservedContainersFromApp(schedulerApp2, 1);
    // Call editSchedule twice and allocation once, container should get allocated
    editPolicy.editSchedule();
    editPolicy.editSchedule();
    int tick = 0;
    while (schedulerApp2.getLiveContainers().size() != 1 && tick < 10) {
        cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
        cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
        tick++;
        Thread.sleep(100);
    }
    waitNumberOfReservedContainersFromApp(schedulerApp2, 0);
    rm1.close();
}