Search in sources :

Example 21 with FiCaSchedulerNode

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode in project hadoop by apache.

the class TestReservations method testAssignToQueue.

@Test
public void testAssignToQueue() throws Exception {
    CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
    setup(csConf);
    // Manipulate queue 'a'
    LeafQueue a = stubLeafQueue((LeafQueue) queues.get(A));
    // Users
    final String user_0 = "user_0";
    // Submit applications
    final ApplicationAttemptId appAttemptId_0 = TestUtils.getMockApplicationAttemptId(0, 0);
    FiCaSchedulerApp app_0 = new FiCaSchedulerApp(appAttemptId_0, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    app_0 = spy(app_0);
    Mockito.doNothing().when(app_0).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    rmContext.getRMApps().put(app_0.getApplicationId(), mock(RMApp.class));
    a.submitApplicationAttempt(app_0, user_0);
    final ApplicationAttemptId appAttemptId_1 = TestUtils.getMockApplicationAttemptId(1, 0);
    FiCaSchedulerApp app_1 = new FiCaSchedulerApp(appAttemptId_1, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    app_1 = spy(app_1);
    Mockito.doNothing().when(app_1).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    a.submitApplicationAttempt(app_1, user_0);
    // Setup some nodes
    String host_0 = "host_0";
    FiCaSchedulerNode node_0 = TestUtils.getMockNode(host_0, DEFAULT_RACK, 0, 8 * GB);
    String host_1 = "host_1";
    FiCaSchedulerNode node_1 = TestUtils.getMockNode(host_1, DEFAULT_RACK, 0, 8 * GB);
    String host_2 = "host_2";
    FiCaSchedulerNode node_2 = TestUtils.getMockNode(host_2, DEFAULT_RACK, 0, 8 * GB);
    when(csContext.getNode(node_0.getNodeID())).thenReturn(node_0);
    when(csContext.getNode(node_1.getNodeID())).thenReturn(node_1);
    when(csContext.getNode(node_2.getNodeID())).thenReturn(node_2);
    Map<ApplicationAttemptId, FiCaSchedulerApp> apps = ImmutableMap.of(app_0.getApplicationAttemptId(), app_0, app_1.getApplicationAttemptId(), app_1);
    Map<NodeId, FiCaSchedulerNode> nodes = ImmutableMap.of(node_0.getNodeID(), node_0, node_1.getNodeID(), node_1, node_2.getNodeID(), node_2);
    final int numNodes = 2;
    Resource clusterResource = Resources.createResource(numNodes * (8 * GB));
    when(csContext.getNumClusterNodes()).thenReturn(numNodes);
    // Setup resource-requests
    Priority priorityAM = TestUtils.createMockPriority(1);
    Priority priorityMap = TestUtils.createMockPriority(5);
    Priority priorityReduce = TestUtils.createMockPriority(10);
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 2 * GB, 1, true, priorityAM, recordFactory)));
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 5 * GB, 2, true, priorityReduce, recordFactory)));
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 3 * GB, 2, true, priorityMap, recordFactory)));
    // Start testing...
    // Only AM
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(2 * GB, a.getUsedResources().getMemorySize());
    assertEquals(2 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(2 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(14 * GB, a.getMetrics().getAvailableMB());
    assertEquals(2 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_1.getAllocatedResource().getMemorySize());
    // Only 1 map - simulating reduce
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(5 * GB, a.getUsedResources().getMemorySize());
    assertEquals(5 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(5 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(11 * GB, a.getMetrics().getAvailableMB());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_1.getAllocatedResource().getMemorySize());
    // Only 1 map to other node - simulating reduce
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_1, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(8 * GB, a.getUsedResources().getMemorySize());
    assertEquals(8 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(8 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(8 * GB, a.getMetrics().getAvailableMB());
    assertEquals(null, node_0.getReservedContainer());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(3 * GB, node_1.getAllocatedResource().getMemorySize());
    // now add in reservations and make sure it continues if config set
    // allocate to queue so that the potential new capacity is greater then
    // absoluteMaxCapacity
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(13 * GB, a.getUsedResources().getMemorySize());
    assertEquals(8 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(5 * GB, a.getMetrics().getReservedMB());
    assertEquals(8 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(3 * GB, a.getMetrics().getAvailableMB());
    assertEquals(3 * GB, app_0.getHeadroom().getMemorySize());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(3 * GB, node_1.getAllocatedResource().getMemorySize());
    ResourceLimits limits = new ResourceLimits(Resources.createResource(13 * GB));
    boolean res = a.canAssignToThisQueue(Resources.createResource(13 * GB), RMNodeLabelsManager.NO_LABEL, limits, Resources.createResource(3 * GB), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY);
    assertTrue(res);
    // 16GB total, 13GB consumed (8 allocated, 5 reserved). asking for 5GB so we would have to
    // unreserve 2GB to get the total 5GB needed.
    // also note vcore checks not enabled
    assertEquals(0, limits.getHeadroom().getMemorySize());
    refreshQueuesTurnOffReservationsContLook(a, csConf);
    // should return false since reservations continue look is off.
    limits = new ResourceLimits(Resources.createResource(13 * GB));
    res = a.canAssignToThisQueue(Resources.createResource(13 * GB), RMNodeLabelsManager.NO_LABEL, limits, Resources.createResource(3 * GB), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY);
    assertFalse(res);
}
Also used : RMApp(org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp) FiCaSchedulerNode(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode) Priority(org.apache.hadoop.yarn.api.records.Priority) AMState(org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt.AMState) Resource(org.apache.hadoop.yarn.api.records.Resource) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) FiCaSchedulerApp(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp) ResourceLimits(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceLimits) NodeId(org.apache.hadoop.yarn.api.records.NodeId) ActiveUsersManager(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager) Test(org.junit.Test)

Example 22 with FiCaSchedulerNode

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode in project hadoop by apache.

the class TestReservations method testReservation.

@Test
@SuppressWarnings("unchecked")
public void testReservation() throws Exception {
    // Test that we now unreserve and use a node that has space
    CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
    setup(csConf);
    // Manipulate queue 'a'
    LeafQueue a = stubLeafQueue((LeafQueue) queues.get(A));
    // Users
    final String user_0 = "user_0";
    // Submit applications
    final ApplicationAttemptId appAttemptId_0 = TestUtils.getMockApplicationAttemptId(0, 0);
    FiCaSchedulerApp app_0 = new FiCaSchedulerApp(appAttemptId_0, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    app_0 = spy(app_0);
    Mockito.doNothing().when(app_0).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    rmContext.getRMApps().put(app_0.getApplicationId(), mock(RMApp.class));
    a.submitApplicationAttempt(app_0, user_0);
    final ApplicationAttemptId appAttemptId_1 = TestUtils.getMockApplicationAttemptId(1, 0);
    FiCaSchedulerApp app_1 = new FiCaSchedulerApp(appAttemptId_1, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    app_1 = spy(app_1);
    Mockito.doNothing().when(app_1).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    a.submitApplicationAttempt(app_1, user_0);
    // Setup some nodes
    String host_0 = "host_0";
    FiCaSchedulerNode node_0 = TestUtils.getMockNode(host_0, DEFAULT_RACK, 0, 8 * GB);
    String host_1 = "host_1";
    FiCaSchedulerNode node_1 = TestUtils.getMockNode(host_1, DEFAULT_RACK, 0, 8 * GB);
    String host_2 = "host_2";
    FiCaSchedulerNode node_2 = TestUtils.getMockNode(host_2, DEFAULT_RACK, 0, 8 * GB);
    Map<ApplicationAttemptId, FiCaSchedulerApp> apps = ImmutableMap.of(app_0.getApplicationAttemptId(), app_0, app_1.getApplicationAttemptId(), app_1);
    Map<NodeId, FiCaSchedulerNode> nodes = ImmutableMap.of(node_0.getNodeID(), node_0, node_1.getNodeID(), node_1, node_2.getNodeID(), node_2);
    when(csContext.getNode(node_0.getNodeID())).thenReturn(node_0);
    when(csContext.getNode(node_1.getNodeID())).thenReturn(node_1);
    when(csContext.getNode(node_2.getNodeID())).thenReturn(node_2);
    cs.getNodeTracker().addNode(node_0);
    cs.getNodeTracker().addNode(node_1);
    cs.getNodeTracker().addNode(node_2);
    final int numNodes = 3;
    Resource clusterResource = Resources.createResource(numNodes * (8 * GB));
    when(csContext.getNumClusterNodes()).thenReturn(numNodes);
    // Setup resource-requests
    Priority priorityAM = TestUtils.createMockPriority(1);
    Priority priorityMap = TestUtils.createMockPriority(5);
    Priority priorityReduce = TestUtils.createMockPriority(10);
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 2 * GB, 1, true, priorityAM, recordFactory)));
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 5 * GB, 2, true, priorityReduce, recordFactory)));
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 3 * GB, 2, true, priorityMap, recordFactory)));
    // Start testing...
    // Only AM
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(2 * GB, a.getUsedResources().getMemorySize());
    assertEquals(2 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(2 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(22 * GB, a.getMetrics().getAvailableMB());
    assertEquals(2 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
    // Only 1 map - simulating reduce
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(5 * GB, a.getUsedResources().getMemorySize());
    assertEquals(5 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(5 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(19 * GB, a.getMetrics().getAvailableMB());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
    // Only 1 map to other node - simulating reduce
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_1, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(8 * GB, a.getUsedResources().getMemorySize());
    assertEquals(8 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(8 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(16 * GB, a.getMetrics().getAvailableMB());
    assertEquals(16 * GB, app_0.getHeadroom().getMemorySize());
    assertEquals(null, node_0.getReservedContainer());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(3 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
    assertEquals(2, app_0.getOutstandingAsksCount(toSchedulerKey(priorityReduce)));
    // try to assign reducer (5G on node 0 and should reserve)
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(13 * GB, a.getUsedResources().getMemorySize());
    assertEquals(8 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(5 * GB, a.getMetrics().getReservedMB());
    assertEquals(8 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(11 * GB, a.getMetrics().getAvailableMB());
    assertEquals(11 * GB, app_0.getHeadroom().getMemorySize());
    assertEquals(5 * GB, node_0.getReservedContainer().getReservedResource().getMemorySize());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(3 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
    assertEquals(2, app_0.getOutstandingAsksCount(toSchedulerKey(priorityReduce)));
    // assign reducer to node 2
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_2, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(18 * GB, a.getUsedResources().getMemorySize());
    assertEquals(13 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(5 * GB, a.getMetrics().getReservedMB());
    assertEquals(13 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(6 * GB, a.getMetrics().getAvailableMB());
    assertEquals(6 * GB, app_0.getHeadroom().getMemorySize());
    assertEquals(5 * GB, node_0.getReservedContainer().getReservedResource().getMemorySize());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(3 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(5 * GB, node_2.getAllocatedResource().getMemorySize());
    assertEquals(1, app_0.getOutstandingAsksCount(toSchedulerKey(priorityReduce)));
    // node_1 heartbeat and unreserves from node_0 in order to allocate
    // on node_1
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_1, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(18 * GB, a.getUsedResources().getMemorySize());
    assertEquals(18 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(18 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(6 * GB, a.getMetrics().getAvailableMB());
    assertEquals(6 * GB, app_0.getHeadroom().getMemorySize());
    assertEquals(null, node_0.getReservedContainer());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(8 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(5 * GB, node_2.getAllocatedResource().getMemorySize());
    assertEquals(0, app_0.getOutstandingAsksCount(toSchedulerKey(priorityReduce)));
}
Also used : RMApp(org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp) FiCaSchedulerNode(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode) Priority(org.apache.hadoop.yarn.api.records.Priority) AMState(org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt.AMState) Resource(org.apache.hadoop.yarn.api.records.Resource) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) FiCaSchedulerApp(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp) ResourceLimits(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceLimits) NodeId(org.apache.hadoop.yarn.api.records.NodeId) ActiveUsersManager(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager) Test(org.junit.Test)

Example 23 with FiCaSchedulerNode

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode in project hadoop by apache.

the class TestReservations method testReservationLimitOtherUsers.

// Test that hitting a reservation limit and needing to unreserve
// does not affect assigning containers for other users
@Test
@SuppressWarnings("unchecked")
public void testReservationLimitOtherUsers() throws Exception {
    CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
    setup(csConf, true);
    // Manipulate queue 'a'
    LeafQueue a = stubLeafQueue((LeafQueue) queues.get(A));
    // Users
    final String user_0 = "user_0";
    final String user_1 = "user_1";
    // Submit applications
    final ApplicationAttemptId appAttemptId_0 = TestUtils.getMockApplicationAttemptId(0, 0);
    FiCaSchedulerApp app_0 = new FiCaSchedulerApp(appAttemptId_0, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    app_0 = spy(app_0);
    Mockito.doNothing().when(app_0).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    rmContext.getRMApps().put(app_0.getApplicationId(), mock(RMApp.class));
    a.submitApplicationAttempt(app_0, user_0);
    final ApplicationAttemptId appAttemptId_1 = TestUtils.getMockApplicationAttemptId(1, 0);
    FiCaSchedulerApp app_1 = new FiCaSchedulerApp(appAttemptId_1, user_1, a, mock(ActiveUsersManager.class), spyRMContext);
    app_1 = spy(app_1);
    Mockito.doNothing().when(app_1).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    rmContext.getRMApps().put(app_1.getApplicationId(), mock(RMApp.class));
    a.submitApplicationAttempt(app_1, user_1);
    // Setup some nodes
    String host_0 = "host_0";
    FiCaSchedulerNode node_0 = TestUtils.getMockNode(host_0, DEFAULT_RACK, 0, 8 * GB);
    String host_1 = "host_1";
    FiCaSchedulerNode node_1 = TestUtils.getMockNode(host_1, DEFAULT_RACK, 0, 8 * GB);
    String host_2 = "host_2";
    FiCaSchedulerNode node_2 = TestUtils.getMockNode(host_2, DEFAULT_RACK, 0, 8 * GB);
    when(csContext.getNode(node_0.getNodeID())).thenReturn(node_0);
    when(csContext.getNode(node_1.getNodeID())).thenReturn(node_1);
    when(csContext.getNode(node_2.getNodeID())).thenReturn(node_2);
    Map<ApplicationAttemptId, FiCaSchedulerApp> apps = ImmutableMap.of(app_0.getApplicationAttemptId(), app_0, app_1.getApplicationAttemptId(), app_1);
    Map<NodeId, FiCaSchedulerNode> nodes = ImmutableMap.of(node_0.getNodeID(), node_0, node_1.getNodeID(), node_1, node_2.getNodeID(), node_2);
    cs.getNodeTracker().addNode(node_0);
    cs.getNodeTracker().addNode(node_1);
    cs.getNodeTracker().addNode(node_2);
    final int numNodes = 3;
    Resource clusterResource = Resources.createResource(numNodes * (8 * GB));
    when(csContext.getNumClusterNodes()).thenReturn(numNodes);
    // Setup resource-requests
    Priority priorityAM = TestUtils.createMockPriority(1);
    Priority priorityMap = TestUtils.createMockPriority(5);
    Priority priorityReduce = TestUtils.createMockPriority(10);
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 2 * GB, 1, true, priorityAM, recordFactory)));
    app_1.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 2 * GB, 1, true, priorityAM, recordFactory)));
    // Start testing...
    // Only AM
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(2 * GB, a.getUsedResources().getMemorySize());
    assertEquals(2 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, app_1.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(2 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(22 * GB, a.getMetrics().getAvailableMB());
    assertEquals(2 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_1, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(4 * GB, a.getUsedResources().getMemorySize());
    assertEquals(2 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(2 * GB, app_1.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(4 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(20 * GB, a.getMetrics().getAvailableMB());
    assertEquals(2 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(2 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
    // Add a few requests to each app
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 8 * GB, 2, true, priorityMap, recordFactory)));
    app_1.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 2 * GB, 2, true, priorityMap, recordFactory)));
    // add a reservation for app_0
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(12 * GB, a.getUsedResources().getMemorySize());
    assertEquals(2 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(2 * GB, app_1.getCurrentConsumption().getMemorySize());
    assertEquals(8 * GB, a.getMetrics().getReservedMB());
    assertEquals(4 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(12 * GB, a.getMetrics().getAvailableMB());
    assertEquals(2 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(2 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
    // next assignment is beyond user limit for user_0 but it should assign to
    // app_1 for user_1
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_1, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(14 * GB, a.getUsedResources().getMemorySize());
    assertEquals(2 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(4 * GB, app_1.getCurrentConsumption().getMemorySize());
    assertEquals(8 * GB, a.getMetrics().getReservedMB());
    assertEquals(6 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(10 * GB, a.getMetrics().getAvailableMB());
    assertEquals(2 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(4 * GB, node_1.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_2.getAllocatedResource().getMemorySize());
}
Also used : RMApp(org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp) FiCaSchedulerNode(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode) Priority(org.apache.hadoop.yarn.api.records.Priority) AMState(org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt.AMState) Resource(org.apache.hadoop.yarn.api.records.Resource) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) FiCaSchedulerApp(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp) ResourceLimits(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceLimits) NodeId(org.apache.hadoop.yarn.api.records.NodeId) ActiveUsersManager(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager) Test(org.junit.Test)

Example 24 with FiCaSchedulerNode

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode in project hadoop by apache.

the class TestReservations method testFindNodeToUnreserve.

@Test
public void testFindNodeToUnreserve() throws Exception {
    CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
    setup(csConf);
    final String user_0 = "user_0";
    final ApplicationAttemptId appAttemptId_0 = TestUtils.getMockApplicationAttemptId(0, 0);
    LeafQueue a = stubLeafQueue((LeafQueue) queues.get(A));
    FiCaSchedulerApp app_0 = new FiCaSchedulerApp(appAttemptId_0, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    String host_1 = "host_1";
    FiCaSchedulerNode node_1 = TestUtils.getMockNode(host_1, DEFAULT_RACK, 0, 8 * GB);
    // Setup resource-requests
    Priority p = TestUtils.createMockPriority(5);
    SchedulerRequestKey priorityMap = toSchedulerKey(p);
    Resource capability = Resources.createResource(2 * GB, 0);
    RMApplicationHistoryWriter writer = mock(RMApplicationHistoryWriter.class);
    SystemMetricsPublisher publisher = mock(SystemMetricsPublisher.class);
    RMContext rmContext = mock(RMContext.class);
    ContainerAllocationExpirer expirer = mock(ContainerAllocationExpirer.class);
    DrainDispatcher drainDispatcher = new DrainDispatcher();
    when(rmContext.getContainerAllocationExpirer()).thenReturn(expirer);
    when(rmContext.getDispatcher()).thenReturn(drainDispatcher);
    when(rmContext.getRMApplicationHistoryWriter()).thenReturn(writer);
    when(rmContext.getSystemMetricsPublisher()).thenReturn(publisher);
    when(rmContext.getYarnConfiguration()).thenReturn(new YarnConfiguration());
    ApplicationAttemptId appAttemptId = BuilderUtils.newApplicationAttemptId(app_0.getApplicationId(), 1);
    ContainerId containerId = BuilderUtils.newContainerId(appAttemptId, 1);
    Container container = TestUtils.getMockContainer(containerId, node_1.getNodeID(), Resources.createResource(2 * GB), priorityMap.getPriority());
    RMContainer rmContainer = new RMContainerImpl(container, SchedulerRequestKey.extractFrom(container), appAttemptId, node_1.getNodeID(), "user", rmContext);
    // nothing reserved
    RMContainer toUnreserveContainer = app_0.findNodeToUnreserve(csContext.getClusterResource(), node_1, priorityMap, capability);
    assertTrue(toUnreserveContainer == null);
    // reserved but scheduler doesn't know about that node.
    app_0.reserve(node_1, priorityMap, rmContainer, container);
    node_1.reserveResource(app_0, priorityMap, rmContainer);
    toUnreserveContainer = app_0.findNodeToUnreserve(csContext.getClusterResource(), node_1, priorityMap, capability);
    assertTrue(toUnreserveContainer == null);
}
Also used : DrainDispatcher(org.apache.hadoop.yarn.event.DrainDispatcher) RMContext(org.apache.hadoop.yarn.server.resourcemanager.RMContext) FiCaSchedulerNode(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode) ContainerAllocationExpirer(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.ContainerAllocationExpirer) Priority(org.apache.hadoop.yarn.api.records.Priority) Resource(org.apache.hadoop.yarn.api.records.Resource) RMApplicationHistoryWriter(org.apache.hadoop.yarn.server.resourcemanager.ahs.RMApplicationHistoryWriter) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) RMContainer(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer) SchedulerRequestKey(org.apache.hadoop.yarn.server.scheduler.SchedulerRequestKey) RMContainer(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer) Container(org.apache.hadoop.yarn.api.records.Container) RMContainerImpl(org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainerImpl) SystemMetricsPublisher(org.apache.hadoop.yarn.server.resourcemanager.metrics.SystemMetricsPublisher) YarnConfiguration(org.apache.hadoop.yarn.conf.YarnConfiguration) ContainerId(org.apache.hadoop.yarn.api.records.ContainerId) FiCaSchedulerApp(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp) ActiveUsersManager(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager) Test(org.junit.Test)

Example 25 with FiCaSchedulerNode

use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode in project hadoop by apache.

the class TestReservations method testAssignToUser.

@Test
public void testAssignToUser() throws Exception {
    CapacitySchedulerConfiguration csConf = new CapacitySchedulerConfiguration();
    setup(csConf);
    // Manipulate queue 'a'
    LeafQueue a = stubLeafQueue((LeafQueue) queues.get(A));
    // Users
    final String user_0 = "user_0";
    // Submit applications
    final ApplicationAttemptId appAttemptId_0 = TestUtils.getMockApplicationAttemptId(0, 0);
    FiCaSchedulerApp app_0 = new FiCaSchedulerApp(appAttemptId_0, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    app_0 = spy(app_0);
    Mockito.doNothing().when(app_0).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    rmContext.getRMApps().put(app_0.getApplicationId(), mock(RMApp.class));
    a.submitApplicationAttempt(app_0, user_0);
    final ApplicationAttemptId appAttemptId_1 = TestUtils.getMockApplicationAttemptId(1, 0);
    FiCaSchedulerApp app_1 = new FiCaSchedulerApp(appAttemptId_1, user_0, a, mock(ActiveUsersManager.class), spyRMContext);
    app_1 = spy(app_1);
    Mockito.doNothing().when(app_1).updateAMContainerDiagnostics(any(AMState.class), any(String.class));
    a.submitApplicationAttempt(app_1, user_0);
    // Setup some nodes
    String host_0 = "host_0";
    FiCaSchedulerNode node_0 = TestUtils.getMockNode(host_0, DEFAULT_RACK, 0, 8 * GB);
    String host_1 = "host_1";
    FiCaSchedulerNode node_1 = TestUtils.getMockNode(host_1, DEFAULT_RACK, 0, 8 * GB);
    String host_2 = "host_2";
    FiCaSchedulerNode node_2 = TestUtils.getMockNode(host_2, DEFAULT_RACK, 0, 8 * GB);
    Map<ApplicationAttemptId, FiCaSchedulerApp> apps = ImmutableMap.of(app_0.getApplicationAttemptId(), app_0, app_1.getApplicationAttemptId(), app_1);
    Map<NodeId, FiCaSchedulerNode> nodes = ImmutableMap.of(node_0.getNodeID(), node_0, node_1.getNodeID(), node_1, node_2.getNodeID(), node_2);
    when(csContext.getNode(node_0.getNodeID())).thenReturn(node_0);
    when(csContext.getNode(node_1.getNodeID())).thenReturn(node_1);
    when(csContext.getNode(node_2.getNodeID())).thenReturn(node_2);
    final int numNodes = 2;
    Resource clusterResource = Resources.createResource(numNodes * (8 * GB));
    when(csContext.getNumClusterNodes()).thenReturn(numNodes);
    // Setup resource-requests
    Priority priorityAM = TestUtils.createMockPriority(1);
    Priority priorityMap = TestUtils.createMockPriority(5);
    Priority priorityReduce = TestUtils.createMockPriority(10);
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 2 * GB, 1, true, priorityAM, recordFactory)));
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 3 * GB, 2, true, priorityMap, recordFactory)));
    app_0.updateResourceRequests(Collections.singletonList(TestUtils.createResourceRequest(ResourceRequest.ANY, 5 * GB, 2, true, priorityReduce, recordFactory)));
    // Start testing...
    // Only AM
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(2 * GB, a.getUsedResources().getMemorySize());
    assertEquals(2 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(2 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(14 * GB, a.getMetrics().getAvailableMB());
    assertEquals(2 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_1.getAllocatedResource().getMemorySize());
    // Only 1 map - simulating reduce
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(5 * GB, a.getUsedResources().getMemorySize());
    assertEquals(5 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(5 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(11 * GB, a.getMetrics().getAvailableMB());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(0 * GB, node_1.getAllocatedResource().getMemorySize());
    // Only 1 map to other node - simulating reduce
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_1, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(8 * GB, a.getUsedResources().getMemorySize());
    assertEquals(8 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(0 * GB, a.getMetrics().getReservedMB());
    assertEquals(8 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(8 * GB, a.getMetrics().getAvailableMB());
    assertEquals(null, node_0.getReservedContainer());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(3 * GB, node_1.getAllocatedResource().getMemorySize());
    // now add in reservations and make sure it continues if config set
    // allocate to queue so that the potential new capacity is greater then
    // absoluteMaxCapacity
    TestUtils.applyResourceCommitRequest(clusterResource, a.assignContainers(clusterResource, node_0, new ResourceLimits(clusterResource), SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
    assertEquals(13 * GB, a.getUsedResources().getMemorySize());
    assertEquals(8 * GB, app_0.getCurrentConsumption().getMemorySize());
    assertEquals(5 * GB, app_0.getCurrentReservation().getMemorySize());
    assertEquals(5 * GB, a.getMetrics().getReservedMB());
    assertEquals(8 * GB, a.getMetrics().getAllocatedMB());
    assertEquals(3 * GB, a.getMetrics().getAvailableMB());
    assertEquals(3 * GB, app_0.getHeadroom().getMemorySize());
    assertEquals(5 * GB, node_0.getAllocatedResource().getMemorySize());
    assertEquals(3 * GB, node_1.getAllocatedResource().getMemorySize());
    // not over the limit
    Resource limit = Resources.createResource(14 * GB, 0);
    ResourceLimits userResourceLimits = new ResourceLimits(clusterResource);
    boolean res = a.canAssignToUser(clusterResource, user_0, limit, app_0, "", userResourceLimits);
    assertTrue(res);
    assertEquals(Resources.none(), userResourceLimits.getAmountNeededUnreserve());
    // set limit so it subtracts reservations and it can continue
    limit = Resources.createResource(12 * GB, 0);
    userResourceLimits = new ResourceLimits(clusterResource);
    res = a.canAssignToUser(clusterResource, user_0, limit, app_0, "", userResourceLimits);
    assertTrue(res);
    // limit set to 12GB, we are using 13GB (8 allocated,  5 reserved), to get under limit
    // we need to unreserve 1GB
    // also note vcore checks not enabled
    assertEquals(Resources.createResource(1 * GB, 4), userResourceLimits.getAmountNeededUnreserve());
    refreshQueuesTurnOffReservationsContLook(a, csConf);
    userResourceLimits = new ResourceLimits(clusterResource);
    // should now return false since feature off
    res = a.canAssignToUser(clusterResource, user_0, limit, app_0, "", userResourceLimits);
    assertFalse(res);
    assertEquals(Resources.none(), userResourceLimits.getAmountNeededUnreserve());
}
Also used : RMApp(org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp) FiCaSchedulerNode(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode) Priority(org.apache.hadoop.yarn.api.records.Priority) AMState(org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt.AMState) Resource(org.apache.hadoop.yarn.api.records.Resource) ApplicationAttemptId(org.apache.hadoop.yarn.api.records.ApplicationAttemptId) FiCaSchedulerApp(org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp) ResourceLimits(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceLimits) NodeId(org.apache.hadoop.yarn.api.records.NodeId) ActiveUsersManager(org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager) Test(org.junit.Test)

Aggregations

FiCaSchedulerNode (org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode)79 FiCaSchedulerApp (org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp)47 Resource (org.apache.hadoop.yarn.api.records.Resource)46 Test (org.junit.Test)39 ResourceLimits (org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceLimits)37 NodeId (org.apache.hadoop.yarn.api.records.NodeId)35 ApplicationAttemptId (org.apache.hadoop.yarn.api.records.ApplicationAttemptId)34 Priority (org.apache.hadoop.yarn.api.records.Priority)34 RMContainer (org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer)21 ActiveUsersManager (org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager)20 ArrayList (java.util.ArrayList)14 RMApp (org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp)11 ResourceRequest (org.apache.hadoop.yarn.api.records.ResourceRequest)10 SchedulerRequestKey (org.apache.hadoop.yarn.server.scheduler.SchedulerRequestKey)9 HashMap (java.util.HashMap)8 YarnConfiguration (org.apache.hadoop.yarn.conf.YarnConfiguration)7 AMState (org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt.AMState)7 Container (org.apache.hadoop.yarn.api.records.Container)6 ContainerId (org.apache.hadoop.yarn.api.records.ContainerId)5 RMContext (org.apache.hadoop.yarn.server.resourcemanager.RMContext)5