Weird behavior for uneven distributed pie charts

[timroes]

If you have a pie chart, that has a very uneven distribution, i.e. very large and very slow slices and spare data on an outer level (at least I think that will be the cause), the chart drawing is behaving unpredictable.

Use the following sample data: data.zip

Installation instructions

unzip data.zip

curl -XPUT 'http://localhost:9200/hotel'

curl -H 'Content-Type: application/x-ndjson' -XPOST 'localhost:9200/hotel/type1/_bulk?pretty' --data-binary @1request.json
curl -H 'Content-Type: application/x-ndjson' -XPOST 'localhost:9200/hotel/type1/_bulk?pretty' --data-binary @2request.json
curl -H 'Content-Type: application/x-ndjson' -XPOST 'localhost:9200/hotel/type1/_bulk?pretty' --data-binary @3request.json

If you now create a chart, with the following configuration:

• Bucket 1: Terms group.keyword (desc, size 5)
• Bucket 2: Filters:
  • city:tokyo
  • city:Osaka
  • city:nara
• Metrics: Count

After hitting apply I sometimes end up with the following chart:

That chart looks not like what I would expect looking at the data:

Actually G1 is the lowest value in the inner circle, but it gets a slice that takes around 99% of the inner circle. It seems the chart does strip away those other slices because they don't have a value on the outer circle. Also if you hover over the inner circle, you see that G1 actually only has a percentage of 0.03% on that chart, but still it's slice is the only visible.

Sometimes a resizing of the chart (e.g. in the below animation, by opening the dev tools and closing them again) helps to get the "correct" result. This is the way we would expect the inner level of the pie chart to look like. The outer level actually vanished because all it's slices having a subpixel size, which is issue #16931:

[flash1293]

@timroes Can't reproduce this for 6.6, 6.7 and master (Chrome Mac). Should this be part of the current stabilizing phase?

[timroes]

Did you try to reproduce it with the above sample data? If so we can close that for now, since it was reliable reproducible using that data.

[markov00]

I've tested locally with the dataset. This is what I see:

I've applied a slight black stroke to each slice to see the 2nd level that is actually all empty except for the G1 group that is 0.3% and has few documents with it.
Seems to the the right representation, what do you think @timroes ?

I've also tested the resizing of the page (with the chrometool open/close and resizing the windows) but it render always the same chart.

I'm closing this because seems to be fixed and/or not reproducible anymore