Ordering of subaggregations in Kibana 6.2 (vs 6.0 and earlier)

[mikeage (Mike Miller)]

Hi,

I have a bunch of records (representing test data) in elastic search with the following fields:

name: foo, bar, etc
outcome: passed, failed

I want to see which tests are the most consistent in a given time (or with a filter). In Kibana 6 and earlier, I was able to do the following:

Vertical bar chart
bucket: x-axis, terms, field:name, order by: term
bucket: split series, terms, field:outcome, order by: term

I got a nice chart with my tests listed left to right, with the stacked bar chart showing how many successes and how many failures each had.

In K6.2 (hosted on AWS's ESaaS, if it matters), it seems like the x-axis sort takes into account the second priority sub-aggregation even though the x-axis split is above it. As a result, my tests are no longer sorted by name, but by a mix of name and outcome.

What's strange is that it seems like the ES query and response are identical (confirmed via a diff tool), but the rendering is significantly different.

How can I get back the 6.0 and earlier behavior?

[I'm afraid I can't post screenshots as it's proprietary data]

[mikeage (Mike Miller)]

It was definitely introduced in 6.2.x; running 6.1.2 locally against my 6.2.2 ES instance, I don't see this problem.

To describe it a bit more clearly, the ordering is:
[for all tests that have at least one passed], tests, sorted by name
[for all tests that don't], tests, sorted by name

So I see something like (on the X axis)
TestA, TestB, TestE, TestF, Test C, Test D, Test G
Where A, B, E, and F have at least one pass, and C,D, and G do not. Within each "grouping" of "outcome", they are sorted.

If I reverse the priorities (putting the split series on top), it seems like it does the same thing, but putting tests with 1 or more failure first (because failure comes before passed alphabetically)? This makes a bit of sense, but the first does not (at least to me...).