[
https://issues.apache.org/jira/browse/SLIDER-1224?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16006856#comment-16006856
]
Billie Rinaldi commented on SLIDER-1224:
----------------------------------------
[~kyungwan nam], thanks for the patch! This looks great, so I will close
SLIDER-1104 as a duplicate of this. I took your test from SLIDER-1104 and
expanded it a bit to test this patch, and I found one more thing that needs to
be changed:
https://github.com/apache/incubator-slider/blob/develop/slider-core/src/main/java/org/apache/slider/server/appmaster/state/OutstandingRequestTracker.java#L179
The outcome is set to open when a request is found in openRequests. We should
add a check here that says if the request mayEscalate and isEscalated, the
outcome should be set to escalated. This is similar to what is done when a
request is found in placedRequests, but the mayEscalate property needs to be
checked as well.
Please also include the test in your patch:
{noformat}
diff --git
a/slider-core/src/test/groovy/org/apache/slider/server/appmaster/model/history/TestRoleHistoryOutstandingRequestTracker.groovy
b/slider-core/src/test/groovy/org/apache/slider/server/appmaster/model/history/TestRoleHistoryOutstandingRequestTracker.groovy
index 7be01ad..574ded5 100644
---
a/slider-core/src/test/groovy/org/apache/slider/server/appmaster/model/history/TestRoleHistoryOutstandingRequestTracker.groovy
+++
b/slider-core/src/test/groovy/org/apache/slider/server/appmaster/model/history/TestRoleHistoryOutstandingRequestTracker.groovy
@@ -143,6 +143,52 @@ class TestRoleHistoryOutstandingRequestTracker extends
BaseMockAppStateTest {
}
@Test
+ public void testIssuedEscalatedRequest() throws Throwable {
+ def req1 = tracker.newRequest(host1, 0)
+ def resource = factory.newResource()
+ resource.virtualCores = 1
+ resource.memory = 48;
+ def yarnRequest = req1.buildContainerRequest(resource, role0Status, 0)
+ assert tracker.listPlacedRequests().size() == 1
+ assert tracker.listOpenRequests().size() == 0
+
+ tracker.escalateOutstandingRequests(role0Status.placementTimeoutSeconds *
1000)
+ assert !req1.isEscalated()
+ assert tracker.listPlacedRequests().size() == 1
+ assert tracker.listOpenRequests().size() == 0
+
+ tracker.escalateOutstandingRequests(role0Status.placementTimeoutSeconds *
1000 + 1)
+ assert req1.isEscalated()
+ assert tracker.listPlacedRequests().size() == 0
+ assert tracker.listOpenRequests().size() == 1
+
+ def c1 = factory.newContainer()
+
+ def nodeId = factory.newNodeId()
+ c1.nodeId = nodeId
+ // if request was escalated, container can be allocated to another host
+ // by relaxed placement.
+ nodeId.host = "host9"
+
+ def pri = ContainerPriority.buildPriority(0, false)
+ assert pri > 0
+ c1.setPriority(new MockPriority(pri))
+
+ c1.setResource(resource)
+
+ def issued = req1.issuedRequest
+ assert issued.capability == resource
+ assert issued.priority.priority == c1.getPriority().getPriority()
+ assert req1.resourceRequirementsMatch(resource)
+
+ def allocation = tracker.onContainerAllocated(0, nodeId.host, c1)
+ assert tracker.listPlacedRequests().size() == 0
+ assert tracker.listOpenRequests().size() == 0
+ assert allocation.outcome == ContainerAllocationOutcome.Escalated;
+ assert allocation.origin.is(req1)
+ }
+
+ @Test
public void testResetEntries() throws Throwable {
tracker.newRequest(host1, 0)
tracker.newRequest(host2, 0)
{noformat}
> the outstanding request that has been escalated is failed to cleanup
> --------------------------------------------------------------------
>
> Key: SLIDER-1224
> URL: https://issues.apache.org/jira/browse/SLIDER-1224
> Project: Slider
> Issue Type: Bug
> Affects Versions: Slider 0.92
> Reporter: kyungwan nam
> Assignee: kyungwan nam
> Attachments: SLIDER-1224.001.patch
>
>
> There is a slider app with placement policy is normal ( 0 ).
> If an outstanding request is escalated, the container can be allocated to
> another node, not desired node.
> But, when container is allocated to another node, the outstanding request is
> kept, not to remove
> I think it is the same issue with SLIDER-1104.
> it can still happen with slider-0.92.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
