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

Example 21 with RMNode

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

the class TestFairScheduler method testStrictLocality.

@Test
public void testStrictLocality() throws IOException {
    scheduler.init(conf);
    scheduler.start();
    scheduler.reinitialize(conf, resourceManager.getRMContext());
    RMNode node1 = MockNodes.newNodeInfo(1, Resources.createResource(1024), 1, "127.0.0.1");
    NodeAddedSchedulerEvent nodeEvent1 = new NodeAddedSchedulerEvent(node1);
    scheduler.handle(nodeEvent1);
    RMNode node2 = MockNodes.newNodeInfo(1, Resources.createResource(1024), 2, "127.0.0.2");
    NodeAddedSchedulerEvent nodeEvent2 = new NodeAddedSchedulerEvent(node2);
    scheduler.handle(nodeEvent2);
    ApplicationAttemptId attId1 = createSchedulingRequest(1024, "queue1", "user1", 0);
    ResourceRequest nodeRequest = createResourceRequest(1024, node1.getHostName(), 1, 1, true);
    ResourceRequest rackRequest = createResourceRequest(1024, node1.getRackName(), 1, 1, false);
    ResourceRequest anyRequest = createResourceRequest(1024, ResourceRequest.ANY, 1, 1, false);
    createSchedulingRequestExistingApplication(nodeRequest, attId1);
    createSchedulingRequestExistingApplication(rackRequest, attId1);
    createSchedulingRequestExistingApplication(anyRequest, attId1);
    scheduler.update();
    NodeUpdateSchedulerEvent node1UpdateEvent = new NodeUpdateSchedulerEvent(node1);
    NodeUpdateSchedulerEvent node2UpdateEvent = new NodeUpdateSchedulerEvent(node2);
    // no matter how many heartbeats, node2 should never get a container
    FSAppAttempt app = scheduler.getSchedulerApp(attId1);
    for (int i = 0; i < 10; i++) {
        scheduler.handle(node2UpdateEvent);
        assertEquals(0, app.getLiveContainers().size());
        assertEquals(0, app.getReservedContainers().size());
    }
    // then node1 should get the container
    scheduler.handle(node1UpdateEvent);
    assertEquals(1, app.getLiveContainers().size());
}

Example 22 with RMNode

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

the class TestRMWebServicesNodes method testNodesQueryStateLost.

@Test
public void testNodesQueryStateLost() throws JSONException, Exception {
    WebResource r = resource();
    RMNode rmnode1 = getRunningRMNode("h1", 1234, 5120);
    sendLostEvent(rmnode1);
    RMNode rmnode2 = getRunningRMNode("h2", 1235, 5121);
    sendLostEvent(rmnode2);
    ClientResponse response = r.path("ws").path("v1").path("cluster").path("nodes").queryParam("states", NodeState.LOST.toString()).accept(MediaType.APPLICATION_JSON).get(ClientResponse.class);
    assertEquals(MediaType.APPLICATION_JSON_TYPE + "; " + JettyUtils.UTF_8, response.getType().toString());
    JSONObject json = response.getEntity(JSONObject.class);
    JSONObject nodes = json.getJSONObject("nodes");
    assertEquals("incorrect number of elements", 1, nodes.length());
    JSONArray nodeArray = nodes.getJSONArray("node");
    assertEquals("incorrect number of elements", 2, nodeArray.length());
    for (int i = 0; i < nodeArray.length(); ++i) {
        JSONObject info = nodeArray.getJSONObject(i);
        String[] node = info.get("id").toString().split(":");
        NodeId nodeId = NodeId.newInstance(node[0], Integer.parseInt(node[1]));
        RMNode rmNode = rm.getRMContext().getInactiveRMNodes().get(nodeId);
        WebServicesTestUtils.checkStringMatch("nodeHTTPAddress", "", info.getString("nodeHTTPAddress"));
        if (rmNode != null) {
            WebServicesTestUtils.checkStringMatch("state", rmNode.getState().toString(), info.getString("state"));
        }
    }
}

Example 23 with RMNode

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

the class TestRMWebServicesNodes method testNodesHelper.

public void testNodesHelper(String path, String media) throws JSONException, Exception {
    WebResource r = resource();
    RMNode rmnode1 = getRunningRMNode("h1", 1234, 5120);
    RMNode rmnode2 = getRunningRMNode("h2", 1235, 5121);
    ClientResponse response = r.path("ws").path("v1").path("cluster").path(path).accept(media).get(ClientResponse.class);
    assertEquals(MediaType.APPLICATION_JSON_TYPE + "; " + JettyUtils.UTF_8, response.getType().toString());
    JSONObject json = response.getEntity(JSONObject.class);
    assertEquals("incorrect number of elements", 1, json.length());
    JSONObject nodes = json.getJSONObject("nodes");
    assertEquals("incorrect number of elements", 1, nodes.length());
    JSONArray nodeArray = nodes.getJSONArray("node");
    assertEquals("incorrect number of elements", 2, nodeArray.length());
    JSONObject info = nodeArray.getJSONObject(0);
    String id = info.get("id").toString();
    if (id.matches("h1:1234")) {
        verifyNodeInfo(info, rmnode1);
        verifyNodeInfo(nodeArray.getJSONObject(1), rmnode2);
    } else {
        verifyNodeInfo(info, rmnode2);
        verifyNodeInfo(nodeArray.getJSONObject(1), rmnode1);
    }
}

Example 24 with RMNode

use of org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNode 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 RMNode

use of org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNode 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();
}