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

Example 11 with MockAM

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

the class TestContainerAllocation method testReservedContainerMetricsOnDecommisionedNode.

@Test(timeout = 60000)
public void testReservedContainerMetricsOnDecommisionedNode() throws Exception {
    /**
     * Test case: Submit two application (app1/app2) to a queue. And there's one
     * node with 8G resource in the cluster. App1 allocates a 6G container, Then
     * app2 asks for a 4G container. App2's request will be reserved on the
     * node.
     *
     * Before next node heartbeat, app1 container is completed/killed. So app1
     * container which was reserved will be allocated.
     */
    // inject node label manager
    MockRM rm1 = new MockRM();
    rm1.getRMContext().setNodeLabelManager(mgr);
    rm1.start();
    MockNM nm1 = rm1.registerNode("h1:1234", 8 * GB);
    MockNM nm2 = rm1.registerNode("h2:1234", 8 * GB);
    // launch an app to queue, AM container should be launched in nm1
    RMApp app1 = rm1.submitApp(1 * GB, "app", "user", null, "default");
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    // launch another app to queue, AM container should be launched in nm1
    RMApp app2 = rm1.submitApp(1 * GB, "app", "user", null, "default");
    MockAM am2 = MockRM.launchAndRegisterAM(app2, rm1, nm1);
    am1.allocate("*", 4 * GB, 1, new ArrayList<ContainerId>());
    am2.allocate("*", 4 * GB, 1, new ArrayList<ContainerId>());
    CapacityScheduler cs = (CapacityScheduler) rm1.getResourceScheduler();
    RMNode rmNode1 = rm1.getRMContext().getRMNodes().get(nm1.getNodeId());
    LeafQueue leafQueue = (LeafQueue) cs.getQueue("default");
    // Do node heartbeats 2 times
    // First time will allocate container for app1, second time will reserve
    // container for app2
    cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
    cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
    // App2 will get preference to be allocated on node1, and node1 will be all
    // used by App2.
    FiCaSchedulerApp schedulerApp1 = cs.getApplicationAttempt(am1.getApplicationAttemptId());
    FiCaSchedulerApp schedulerApp2 = cs.getApplicationAttempt(am2.getApplicationAttemptId());
    // Check if a 4G container allocated for app1, and nothing allocated for app2
    Assert.assertEquals(2, schedulerApp1.getLiveContainers().size());
    Assert.assertEquals(1, schedulerApp2.getLiveContainers().size());
    Assert.assertTrue(schedulerApp2.getReservedContainers().size() > 0);
    // NM1 has available resource = 2G (8G - 2 * 1G - 4G)
    Assert.assertEquals(2 * GB, cs.getNode(nm1.getNodeId()).getUnallocatedResource().getMemorySize());
    Assert.assertNotNull(cs.getNode(nm1.getNodeId()).getReservedContainer());
    // Usage of queue = 4G + 2 * 1G + 4G (reserved)
    Assert.assertEquals(10 * GB, cs.getRootQueue().getQueueResourceUsage().getUsed().getMemorySize());
    Assert.assertEquals(4 * GB, cs.getRootQueue().getQueueResourceUsage().getReserved().getMemorySize());
    Assert.assertEquals(4 * GB, leafQueue.getQueueResourceUsage().getReserved().getMemorySize());
    // Remove the node
    cs.handle(new NodeRemovedSchedulerEvent(rmNode1));
    // Check all container cancelled for app1 and app2
    Assert.assertEquals(0, schedulerApp1.getLiveContainers().size());
    Assert.assertEquals(0, schedulerApp2.getLiveContainers().size());
    Assert.assertFalse(schedulerApp2.getReservedContainers().size() > 0);
    // Usage and Reserved capacity of queue is 0
    Assert.assertEquals(0 * GB, cs.getRootQueue().getQueueResourceUsage().getUsed().getMemorySize());
    Assert.assertEquals(0 * GB, cs.getRootQueue().getQueueResourceUsage().getReserved().getMemorySize());
    Assert.assertEquals(0 * GB, leafQueue.getQueueResourceUsage().getReserved().getMemorySize());
    rm1.close();
}

Example 12 with MockAM

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

the class TestContainerAllocation method testAssignMultipleOffswitchContainers.

@Test(timeout = 60000)
public void testAssignMultipleOffswitchContainers() throws Exception {
    MockRM rm1 = new MockRM();
    rm1.getRMContext().setNodeLabelManager(mgr);
    rm1.start();
    MockNM nm1 = rm1.registerNode("h1:1234", 80 * GB);
    // launch an app to queue, AM container should be launched in nm1
    RMApp app1 = rm1.submitApp(1 * GB, "app", "user", null, "default");
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    am1.allocate("*", 1 * GB, 5, new ArrayList<ContainerId>());
    CapacityScheduler cs = (CapacityScheduler) rm1.getResourceScheduler();
    RMNode rmNode1 = rm1.getRMContext().getRMNodes().get(nm1.getNodeId());
    // Do node heartbeats once
    cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
    FiCaSchedulerApp schedulerApp1 = cs.getApplicationAttempt(am1.getApplicationAttemptId());
    // App1 will get one container allocated (plus AM container
    Assert.assertEquals(2, schedulerApp1.getLiveContainers().size());
    // Set assign multiple off-switch containers to 3
    CapacitySchedulerConfiguration newCSConf = new CapacitySchedulerConfiguration();
    newCSConf.setInt(CapacitySchedulerConfiguration.OFFSWITCH_PER_HEARTBEAT_LIMIT, 3);
    cs.reinitialize(newCSConf, rm1.getRMContext());
    // Do node heartbeats once
    cs.handle(new NodeUpdateSchedulerEvent(rmNode1));
    // App1 will get 3 new container allocated (plus 2 previously allocated
    // container)
    Assert.assertEquals(5, schedulerApp1.getLiveContainers().size());
    rm1.close();
}

Example 13 with MockAM

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

the class TestContainerAllocation method getLogAggregationContextFromContainerToken.

private LogAggregationContext getLogAggregationContextFromContainerToken(MockRM rm1, MockNM nm1, LogAggregationContext logAggregationContext) throws Exception {
    RMApp app2 = rm1.submitApp(200, logAggregationContext);
    MockAM am2 = MockRM.launchAndRegisterAM(app2, rm1, nm1);
    nm1.nodeHeartbeat(true);
    // request a container.
    am2.allocate("127.0.0.1", 512, 1, new ArrayList<ContainerId>());
    ContainerId containerId = ContainerId.newContainerId(am2.getApplicationAttemptId(), 2);
    rm1.waitForState(nm1, containerId, RMContainerState.ALLOCATED);
    // acquire the container.
    List<Container> containers = am2.allocate(new ArrayList<ResourceRequest>(), new ArrayList<ContainerId>()).getAllocatedContainers();
    Assert.assertEquals(containerId, containers.get(0).getId());
    // container token is generated.
    Assert.assertNotNull(containers.get(0).getContainerToken());
    ContainerTokenIdentifier token = BuilderUtils.newContainerTokenIdentifier(containers.get(0).getContainerToken());
    return token.getLogAggregationContext();
}

Example 14 with MockAM

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

the class TestIncreaseAllocationExpirer method testContainerIncreaseAllocationExpiration.

@Test
public void testContainerIncreaseAllocationExpiration() throws Exception {
    /**
     * 1. Allocate 1 container: containerId2 (1G)
     * 2. Increase resource of containerId2: 1G -> 3G
     * 3. AM acquires the token
     * 4. AM does not use the token
     * 5. Verify containerId2's resource usage falls back to
     *    1G after the increase token expires
     */
    // Set the allocation expiration to 5 seconds
    conf.setLong(YarnConfiguration.RM_CONTAINER_ALLOC_EXPIRY_INTERVAL_MS, 5000);
    MockRM rm1 = new MockRM(conf);
    rm1.start();
    MockNM nm1 = rm1.registerNode("127.0.0.1:1234", 20 * GB);
    RMApp app1 = rm1.submitApp(1 * GB, "app", "user", null, "default");
    MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1);
    nm1.nodeHeartbeat(app1.getCurrentAppAttempt().getAppAttemptId(), 1, ContainerState.RUNNING);
    am1.allocate("127.0.0.1", 1 * GB, 1, new ArrayList<ContainerId>());
    ContainerId containerId2 = ContainerId.newContainerId(am1.getApplicationAttemptId(), 2);
    rm1.waitForState(nm1, containerId2, RMContainerState.ALLOCATED);
    List<Container> containers = am1.allocate(null, null).getAllocatedContainers();
    Assert.assertEquals(containerId2, containers.get(0).getId());
    Assert.assertNotNull(containers.get(0).getContainerToken());
    checkUsedResource(rm1, "default", 2 * GB, null);
    FiCaSchedulerApp app = TestUtils.getFiCaSchedulerApp(rm1, app1.getApplicationId());
    Assert.assertEquals(2 * GB, app.getAppAttemptResourceUsage().getUsed().getMemorySize());
    verifyAvailableResourceOfSchedulerNode(rm1, nm1.getNodeId(), 18 * GB);
    nm1.nodeHeartbeat(app1.getCurrentAppAttempt().getAppAttemptId(), 2, ContainerState.RUNNING);
    rm1.waitForState(nm1, containerId2, RMContainerState.RUNNING);
    // am1 asks to increase containerId2 from 1GB to 3GB
    am1.sendContainerResizingRequest(Collections.singletonList(UpdateContainerRequest.newInstance(0, containerId2, ContainerUpdateType.INCREASE_RESOURCE, Resources.createResource(3 * GB), null)));
    // Kick off scheduling and wait for 1 second;
    nm1.nodeHeartbeat(true);
    Thread.sleep(1000);
    // Start container increase allocation expirer
    am1.allocate(null, null);
    // Verify resource usage
    checkUsedResource(rm1, "default", 4 * GB, null);
    Assert.assertEquals(4 * GB, app.getAppAttemptResourceUsage().getUsed().getMemorySize());
    verifyAvailableResourceOfSchedulerNode(rm1, nm1.getNodeId(), 16 * GB);
    // Wait long enough for the increase token to expire, and for the roll
    // back action to complete
    Thread.sleep(10000);
    // Verify container size is 1G
    am1.allocate(null, null);
    Assert.assertEquals(1 * GB, rm1.getResourceScheduler().getRMContainer(containerId2).getAllocatedResource().getMemorySize());
    // Verify total resource usage is 2G
    checkUsedResource(rm1, "default", 2 * GB, null);
    Assert.assertEquals(2 * GB, app.getAppAttemptResourceUsage().getUsed().getMemorySize());
    // Verify available resource is rolled back to 18GB
    verifyAvailableResourceOfSchedulerNode(rm1, nm1.getNodeId(), 18 * GB);
    rm1.stop();
}

Example 15 with MockAM

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

the class TestCapacitySchedulerLazyPreemption method testPreemptionConsidersNodeLocalityDelay.

@Test(timeout = 60000)
public void testPreemptionConsidersNodeLocalityDelay() throws Exception {
    /**
     * Test case: same as testSimplePreemption steps 1-3.
     *
     * Step 4: app2 asks for 1G container with locality specified, so it needs
     * to wait for missed-opportunity before get scheduled.
     * Check if system waits missed-opportunity before finish killable container
     */
    MockRM rm1 = new MockRM(conf);
    rm1.getRMContext().setNodeLabelManager(mgr);
    rm1.start();
    MockNM nm1 = rm1.registerNode("h1:1234", 4 * GB);
    MockNM nm2 = rm1.registerNode("h2:1234", 4 * 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, 6, new ArrayList<ContainerId>());
    // Do allocation 3 times 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, and no available resource for cluster
    FiCaSchedulerApp schedulerApp1 = cs.getApplicationAttempt(am1.getApplicationAttemptId());
    Assert.assertEquals(7, schedulerApp1.getLiveContainers().size());
    // Submit app2 to queue-c and asks for a 1G container for AM
    RMApp app2 = rm1.submitApp(1 * GB, "app", "user", null, "c");
    MockAM am2 = MockRM.launchAndRegisterAM(app2, rm1, nm2);
    // NM1/NM2 has available resource = 0G
    Assert.assertEquals(0 * GB, cs.getNode(nm1.getNodeId()).getUnallocatedResource().getMemorySize());
    Assert.assertEquals(0 * GB, cs.getNode(nm2.getNodeId()).getUnallocatedResource().getMemorySize());
    // AM asks for a 1 * GB container with unknown host and unknown rack
    am2.allocate(Arrays.asList(ResourceRequest.newInstance(Priority.newInstance(1), ResourceRequest.ANY, Resources.createResource(1 * GB), 1), ResourceRequest.newInstance(Priority.newInstance(1), "unknownhost", Resources.createResource(1 * GB), 1), ResourceRequest.newInstance(Priority.newInstance(1), "/default-rack", Resources.createResource(1 * GB), 1)), null);
    // Get edit policy and do one update
    SchedulingEditPolicy editPolicy = getSchedulingEditPolicy(rm1);
    // Call edit schedule twice, and check if one container from app1 marked
    // to be "killable"
    editPolicy.editSchedule();
    editPolicy.editSchedule();
    PreemptionManager pm = cs.getPreemptionManager();
    Map<ContainerId, RMContainer> killableContainers = waitKillableContainersSize(pm, "a", RMNodeLabelsManager.NO_LABEL, 1);
    Assert.assertEquals(killableContainers.entrySet().iterator().next().getKey().getApplicationAttemptId(), am1.getApplicationAttemptId());
    // Call CS.handle once to see if container preempted
    cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
    FiCaSchedulerApp schedulerApp2 = cs.getApplicationAttempt(am2.getApplicationAttemptId());
    // App1 has 7 containers, and app2 has 1 containers (no container preempted)
    Assert.assertEquals(7, schedulerApp1.getLiveContainers().size());
    Assert.assertEquals(1, schedulerApp2.getLiveContainers().size());
    // Do allocation again, one container will be preempted
    cs.handle(new NodeUpdateSchedulerEvent(rmNode2));
    // App1 has 6 containers, and app2 has 2 containers (new container allocated)
    Assert.assertEquals(6, schedulerApp1.getLiveContainers().size());
    Assert.assertEquals(2, schedulerApp2.getLiveContainers().size());
    rm1.close();
}