use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeRemovedSchedulerEvent in project hadoop by apache.
the class TestOpportunisticContainerAllocatorAMService method testNodeRemovalDuringAllocate.
@Test(timeout = 60000)
public void testNodeRemovalDuringAllocate() throws Exception {
MockNM nm1 = new MockNM("h1:1234", 4096, rm.getResourceTrackerService());
MockNM nm2 = new MockNM("h2:1234", 4096, rm.getResourceTrackerService());
nm1.registerNode();
nm2.registerNode();
OpportunisticContainerAllocatorAMService amservice = (OpportunisticContainerAllocatorAMService) rm.getApplicationMasterService();
RMApp app1 = rm.submitApp(1 * GB, "app", "user", null, "default");
ApplicationAttemptId attemptId = app1.getCurrentAppAttempt().getAppAttemptId();
MockAM am1 = MockRM.launchAndRegisterAM(app1, rm, nm2);
ResourceScheduler scheduler = rm.getResourceScheduler();
RMNode rmNode1 = rm.getRMContext().getRMNodes().get(nm1.getNodeId());
RMNode rmNode2 = rm.getRMContext().getRMNodes().get(nm2.getNodeId());
nm1.nodeHeartbeat(true);
nm2.nodeHeartbeat(true);
((RMNodeImpl) rmNode1).setOpportunisticContainersStatus(getOppurtunisticStatus(-1, 100));
((RMNodeImpl) rmNode2).setOpportunisticContainersStatus(getOppurtunisticStatus(-1, 100));
OpportunisticContainerContext ctxt = ((CapacityScheduler) scheduler).getApplicationAttempt(attemptId).getOpportunisticContainerContext();
// Send add and update node events to AM Service.
amservice.handle(new NodeAddedSchedulerEvent(rmNode1));
amservice.handle(new NodeAddedSchedulerEvent(rmNode2));
amservice.handle(new NodeUpdateSchedulerEvent(rmNode1));
amservice.handle(new NodeUpdateSchedulerEvent(rmNode2));
// Both node 1 and node 2 will be applicable for scheduling.
for (int i = 0; i < 10; i++) {
am1.allocate(Arrays.asList(ResourceRequest.newInstance(Priority.newInstance(1), "*", Resources.createResource(1 * GB), 2)), null);
if (ctxt.getNodeMap().size() == 2) {
break;
}
Thread.sleep(50);
}
Assert.assertEquals(2, ctxt.getNodeMap().size());
// Remove node from scheduler but not from AM Service.
scheduler.handle(new NodeRemovedSchedulerEvent(rmNode1));
// After removal of node 1, only 1 node will be applicable for scheduling.
for (int i = 0; i < 10; i++) {
try {
am1.allocate(Arrays.asList(ResourceRequest.newInstance(Priority.newInstance(1), "*", Resources.createResource(1 * GB), 2)), null);
} catch (Exception e) {
Assert.fail("Allocate request should be handled on node removal");
}
if (ctxt.getNodeMap().size() == 1) {
break;
}
Thread.sleep(50);
}
Assert.assertEquals(1, ctxt.getNodeMap().size());
}
use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeRemovedSchedulerEvent in project hadoop by apache.
the class TestFairScheduler method testRemoveNodeUpdatesRootQueueMetrics.
@Test
public void testRemoveNodeUpdatesRootQueueMetrics() throws IOException {
scheduler.init(conf);
scheduler.start();
scheduler.reinitialize(conf, resourceManager.getRMContext());
assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
assertEquals(0, scheduler.getRootQueueMetrics().getAvailableVirtualCores());
RMNode node1 = MockNodes.newNodeInfo(1, Resources.createResource(1024, 4), 1, "127.0.0.1");
NodeAddedSchedulerEvent addEvent = new NodeAddedSchedulerEvent(node1);
scheduler.handle(addEvent);
assertEquals(1024, scheduler.getRootQueueMetrics().getAvailableMB());
assertEquals(4, scheduler.getRootQueueMetrics().getAvailableVirtualCores());
// update shouldn't change things
scheduler.update();
assertEquals(1024, scheduler.getRootQueueMetrics().getAvailableMB());
assertEquals(4, scheduler.getRootQueueMetrics().getAvailableVirtualCores());
NodeRemovedSchedulerEvent removeEvent = new NodeRemovedSchedulerEvent(node1);
scheduler.handle(removeEvent);
assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
assertEquals(0, scheduler.getRootQueueMetrics().getAvailableVirtualCores());
// update shouldn't change things
scheduler.update();
assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
assertEquals(0, scheduler.getRootQueueMetrics().getAvailableVirtualCores());
}
use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeRemovedSchedulerEvent in project hadoop by apache.
the class TestContinuousScheduling method testWithNodeRemoved.
@Test
public void testWithNodeRemoved() throws Exception {
// Disable continuous scheduling, will invoke continuous
// scheduling once manually
scheduler = new FairScheduler();
conf = super.createConfiguration();
resourceManager = new MockRM(conf);
// TODO: This test should really be using MockRM. For now starting stuff
// that is needed at a bare minimum.
((AsyncDispatcher) resourceManager.getRMContext().getDispatcher()).start();
resourceManager.getRMContext().getStateStore().start();
// to initialize the master key
resourceManager.getRMContext().getContainerTokenSecretManager().rollMasterKey();
scheduler.setRMContext(resourceManager.getRMContext());
Assert.assertTrue("Continuous scheduling should be disabled.", !scheduler.isContinuousSchedulingEnabled());
scheduler.init(conf);
scheduler.start();
// Add two nodes
RMNode node1 = MockNodes.newNodeInfo(1, Resources.createResource(8 * 1024, 8), 1, "127.0.0.1");
NodeAddedSchedulerEvent nodeEvent1 = new NodeAddedSchedulerEvent(node1);
scheduler.handle(nodeEvent1);
RMNode node2 = MockNodes.newNodeInfo(1, Resources.createResource(8 * 1024, 8), 2, "127.0.0.2");
NodeAddedSchedulerEvent nodeEvent2 = new NodeAddedSchedulerEvent(node2);
scheduler.handle(nodeEvent2);
Assert.assertEquals("We should have two alive nodes.", 2, scheduler.getNumClusterNodes());
// Remove one node
NodeRemovedSchedulerEvent removeNode1 = new NodeRemovedSchedulerEvent(node1);
scheduler.handle(removeNode1);
Assert.assertEquals("We should only have one alive node.", 1, scheduler.getNumClusterNodes());
// Invoke the continuous scheduling once
try {
scheduler.continuousSchedulingAttempt();
} catch (Exception e) {
fail("Exception happened when doing continuous scheduling. " + e.toString());
}
}
use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeRemovedSchedulerEvent in project hadoop by apache.
the class TestFairScheduler method testSteadyFairShareWithReloadAndNodeAddRemove.
@Test
public void testSteadyFairShareWithReloadAndNodeAddRemove() throws Exception {
conf.set(FairSchedulerConfiguration.ALLOCATION_FILE, ALLOC_FILE);
PrintWriter out = new PrintWriter(new FileWriter(ALLOC_FILE));
out.println("<?xml version=\"1.0\"?>");
out.println("<allocations>");
out.println("<defaultQueueSchedulingPolicy>fair</defaultQueueSchedulingPolicy>");
out.println("<queue name=\"root\">");
out.println(" <schedulingPolicy>drf</schedulingPolicy>");
out.println(" <queue name=\"child1\">");
out.println(" <weight>1</weight>");
out.println(" </queue>");
out.println(" <queue name=\"child2\">");
out.println(" <weight>1</weight>");
out.println(" </queue>");
out.println("</queue>");
out.println("</allocations>");
out.close();
scheduler.init(conf);
scheduler.start();
scheduler.reinitialize(conf, resourceManager.getRMContext());
// The steady fair share for all queues should be 0
QueueManager queueManager = scheduler.getQueueManager();
assertEquals(0, queueManager.getLeafQueue("child1", false).getSteadyFairShare().getMemorySize());
assertEquals(0, queueManager.getLeafQueue("child2", false).getSteadyFairShare().getMemorySize());
// Add one node
RMNode node1 = MockNodes.newNodeInfo(1, Resources.createResource(6144), 1, "127.0.0.1");
NodeAddedSchedulerEvent nodeEvent1 = new NodeAddedSchedulerEvent(node1);
scheduler.handle(nodeEvent1);
assertEquals(6144, scheduler.getClusterResource().getMemorySize());
// The steady fair shares for all queues should be updated
assertEquals(2048, queueManager.getLeafQueue("child1", false).getSteadyFairShare().getMemorySize());
assertEquals(2048, queueManager.getLeafQueue("child2", false).getSteadyFairShare().getMemorySize());
// Reload the allocation configuration file
out = new PrintWriter(new FileWriter(ALLOC_FILE));
out.println("<?xml version=\"1.0\"?>");
out.println("<allocations>");
out.println("<defaultQueueSchedulingPolicy>fair</defaultQueueSchedulingPolicy>");
out.println("<queue name=\"root\">");
out.println(" <schedulingPolicy>drf</schedulingPolicy>");
out.println(" <queue name=\"child1\">");
out.println(" <weight>1</weight>");
out.println(" </queue>");
out.println(" <queue name=\"child2\">");
out.println(" <weight>2</weight>");
out.println(" </queue>");
out.println(" <queue name=\"child3\">");
out.println(" <weight>2</weight>");
out.println(" </queue>");
out.println("</queue>");
out.println("</allocations>");
out.close();
scheduler.reinitialize(conf, resourceManager.getRMContext());
// The steady fair shares for all queues should be updated
assertEquals(1024, queueManager.getLeafQueue("child1", false).getSteadyFairShare().getMemorySize());
assertEquals(2048, queueManager.getLeafQueue("child2", false).getSteadyFairShare().getMemorySize());
assertEquals(2048, queueManager.getLeafQueue("child3", false).getSteadyFairShare().getMemorySize());
// Remove the node, steady fair shares should back to 0
NodeRemovedSchedulerEvent nodeEvent2 = new NodeRemovedSchedulerEvent(node1);
scheduler.handle(nodeEvent2);
assertEquals(0, scheduler.getClusterResource().getMemorySize());
assertEquals(0, queueManager.getLeafQueue("child1", false).getSteadyFairShare().getMemorySize());
assertEquals(0, queueManager.getLeafQueue("child2", false).getSteadyFairShare().getMemorySize());
}
use of org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeRemovedSchedulerEvent in project hadoop by apache.
the class TestFairScheduler method testAggregateCapacityTracking.
@Test
public void testAggregateCapacityTracking() throws Exception {
scheduler.init(conf);
scheduler.start();
scheduler.reinitialize(conf, resourceManager.getRMContext());
// Add a node
RMNode node1 = MockNodes.newNodeInfo(1, Resources.createResource(1024), 1, "127.0.0.1");
NodeAddedSchedulerEvent nodeEvent1 = new NodeAddedSchedulerEvent(node1);
scheduler.handle(nodeEvent1);
assertEquals(1024, scheduler.getClusterResource().getMemorySize());
// Add another node
RMNode node2 = MockNodes.newNodeInfo(1, Resources.createResource(512), 2, "127.0.0.2");
NodeAddedSchedulerEvent nodeEvent2 = new NodeAddedSchedulerEvent(node2);
scheduler.handle(nodeEvent2);
assertEquals(1536, scheduler.getClusterResource().getMemorySize());
// Remove the first node
NodeRemovedSchedulerEvent nodeEvent3 = new NodeRemovedSchedulerEvent(node1);
scheduler.handle(nodeEvent3);
assertEquals(512, scheduler.getClusterResource().getMemorySize());
}
Aggregations