HDDS-2214. TestSCMContainerPlacementRackAware has an intermittent fai…

[elek]

What changes were proposed in this pull request?

Fixing an intermittent unit test.

What is the problem

For example from the nightly build:

  <testcase name="testNoFallback[8]" classname="org.apache.hadoop.hdds.scm.container.placement.algorithms.TestSCMContainerPlacementRackAware" time="0.014">
            <failure type="java.lang.AssertionError">java.lang.AssertionError
        	at org.junit.Assert.fail(Assert.java:86)
        	at org.junit.Assert.assertTrue(Assert.java:41)
        	at org.junit.Assert.assertTrue(Assert.java:52)
        	at org.apache.hadoop.hdds.scm.container.placement.algorithms.TestSCMContainerPlacementRackAware.testNoFallback(TestSCMContainerPlacementRackAware.java:276)
        	at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        	at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
        	at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
        	at java.lang.reflect.Method.invoke(Method.java:498)
        	at org.junit.runners.model.FrameworkMethod$1.runReflectiveCall(FrameworkMethod.java:47)

The problem is in the testNoFallback:

Let's say we have 11 nodes (from parameter) and we would like to choose 5 nodes (hard coded in the test).

As the first two replicas are chosen from the same rack an all the other from different racks it's not possible, so we except a failure.

But we have an assertion that the success count is at least 3. But this is true only if the first two replicas are placed to the rack1 (5 nodes) or rack2 (5nodes). If the replica is placed to the rack3 (one node) it will fail immediately:

Lucky case when we have success count > 3

 rack1 -- node1 
 rack1 -- node2 -- FIRST replica
 rack1 -- node3 -- SECOND replica
 rack1 -- node4
 rack1 -- node5 
 rack2 -- node6
 rack2 -- node7 -- THIRD replica
 rack2 -- node8
 rack2 -- node9 
 rack2 -- node10
 rack3 -- node11 -- FOURTH replica{code}

The specific case when we have success count == 1, as we can't choose the second replica on rack3 (This is when the test is failing)

 rack1 -- node1 
 rack1 -- node2
 rack1 -- node3
 rack1 -- node4
 rack1 -- node5 
 rack2 -- node6
 rack2 -- node7
 rack2 -- node8
 rack2 -- node9 
 rack2 -- node10
 rack3 -- node11 -- FIRST replica{code}

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-2289

How was this patch tested?

With Intellij you can execute the unit test multiple times (1000x) or until the next failure. Execute it with or without the patch. Usually I got the problem during the first 100 execution.

[elek]

/retest

[adoroszlai]

Copying my +1 from hadoop repo:

Thanks @elek for digging into this intermittent issue.

[adoroszlai]

@ChenSammi could you please review this fix when you have some time?

[xiaoyuyao]

Thanks @elek for fixing this and @adoroszlai for the review. The change LGTM. I will merge it shortly.