KYLIN-4397 Use newLinkedHashMap for deterministic order in AssignmentUtil.java

[cpugputpu]

This PR aims to solve the issue here: https://issues.apache.org/jira/browse/KYLIN-4397

The fix is to use newLinkedHashMap instead of newHashMap when initializing a Map. In this way, the non-deterministic behaviour is eliminated and the test will not suffer from the failure again. The code will be more stable.

[hit-lacus]

I am a little confused at the moment.

From your message in JIRA issue, you found that error :

java.lang.AssertionError: expected:<2> but was:<1>
at org.apache.kylin.stream.coordinator.assign.DefaultAssignerTest.reBalanceTest(DefaultAssignerTest.java:165) when executing assertEquals(2, rsAssignment.size());

You claim that LinkedHashMap has guarantees to its order, and I agreed with it.
So you thought we should change the implementation of the result from HashMap to LinkedHashMap, right? Andresult is the return value of DefaultAssigner#reBalancePlan.

    public Map<Integer, Map<String, List<Partition>>> reBalancePlan(List<ReplicaSet> replicaSets,
            List<StreamingCubeInfo> cubes, List<CubeAssignment> existingAssignments) {
        Map<Integer, Map<String, List<Partition>>> newPlan = Maps.newHashMap();
        if (replicaSets == null || cubes == null || cubes.size() == 0 || replicaSets.size() == 0) {
            return newPlan;
    }

So, maybe you change the implementation of newPlan, right? Not in the AssignmentUtil.java.

Finally , I didn't reproduce this kind of error in my dev env. I felt confused too.

[hit-lacus]

@cpugputpu Hi, if I misunderstand anything, please let me know, thank you in advance.

[hit-lacus]

Looks like it is not necessary code change. Hope author will give more explanation.

[shaofengshi]

+1 LGTM