[
https://issues.apache.org/jira/browse/YARN-10546?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17254453#comment-17254453
]
zhuqi commented on YARN-10546:
------------------------------
[~leftnoteasy] [~tangzhankun] [~Jim_Brennan] [~Tao Yang]
Our cluster now is FairScheduler, it will happened sometimes before YARN-4270 .
When i write a test, i confirmed CS should also handle it. I will fixed it
later.
{code:java}
@Test
@SuppressWarnings("unchecked")
public void testReservationLimitedForSingleApp() 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);
// Setup 5 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_RACK1, 0,
8 * GB);
String host_2 = "host_2";
FiCaSchedulerNode node_2 = TestUtils.getMockNode(host_2, DEFAULT_RACK2, 0,
8 * GB);
String host_3 = "host_3";
FiCaSchedulerNode node_3 = TestUtils.getMockNode(host_3, DEFAULT_RACK3, 0,
8 * GB);
String host_4 = "host_4";
FiCaSchedulerNode node_4 = TestUtils.getMockNode(host_4, DEFAULT_RACK4, 0,
8 * GB);
Map<ApplicationAttemptId, FiCaSchedulerApp> apps = ImmutableMap.of(
app_0.getApplicationAttemptId(), app_0);
Map<NodeId, FiCaSchedulerNode> nodes = ImmutableMap.of(node_0.getNodeID(),
node_0, node_1.getNodeID(), node_1, node_2.getNodeID(), node_2,
node_3.getNodeID(), node_3, node_4.getNodeID(), node_4
);
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);
when(csContext.getNode(node_3.getNodeID())).thenReturn(node_3);
when(csContext.getNode(node_4.getNodeID())).thenReturn(node_4);
cs.getNodeTracker().addNode(node_0);
cs.getNodeTracker().addNode(node_1);
cs.getNodeTracker().addNode(node_2);
cs.getNodeTracker().addNode(node_3);
cs.getNodeTracker().addNode(node_4);
final int numNodes = 5;
Resource clusterResource = Resources.createResource(numNodes * (8 * GB));
when(csContext.getNumClusterNodes()).thenReturn(numNodes);
root.updateClusterResource(clusterResource,
new ResourceLimits(clusterResource));
// Setup resource-requests
Priority priorityAM = TestUtils.createMockPriority(1);
app_0.updateResourceRequests(Collections.singletonList(TestUtils
.createResourceRequest(ResourceRequest.ANY, 5 * GB, 5, true,
priorityAM, recordFactory)));
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_0,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_1,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_2,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_3,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_4,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
app_0.updateResourceRequests(Collections.singletonList(TestUtils
.createResourceRequest(ResourceRequest.ANY, 4 * GB, 8, true,
priorityAM, recordFactory)));
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_0,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_1,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_2,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_3,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
TestUtils.applyResourceCommitRequest(clusterResource,
a.assignContainers(clusterResource, node_4,
new ResourceLimits(clusterResource),
SchedulingMode.RESPECT_PARTITION_EXCLUSIVITY), nodes, apps);
System.out.println(a.getUsedResources());
System.out.println(a.getMetrics().getReservedMB());
System.out.println(app_0.getReservedContainers());
System.out.println(node_0.getReservedContainer());
System.out.println(node_1.getReservedContainer());
System.out.println(node_2.getReservedContainer());
System.out.println(node_3.getReservedContainer());
System.out.println(node_4.getReservedContainer());
}
{code}
> Limit application resource reservation on nodes for non-node/rack specific
> requests shoud be supported in CS.
> -------------------------------------------------------------------------------------------------------------
>
> Key: YARN-10546
> URL: https://issues.apache.org/jira/browse/YARN-10546
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacityscheduler
> Affects Versions: 3.3.0
> Reporter: zhuqi
> Assignee: zhuqi
> Priority: Major
>
> Just as fixed in YARN-4270 about FairScheduler.
> The capacityScheduler should also fixed it.
> It is a big problem in production cluster, when it happended.
> Also we should support fs convert to cs to support it.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
