[Maps] Downsample Rendered features should be random

[anottrott]

Kibana version: 7.2.0

Elasticsearch version: 7.2.0

Describe the bug:
In the maps app there is a 10k limit to the number of features on a layer that will render. This downsampling is working as expected, however, it is not "geospatially random". Downsampled points are selected in geographically correlated regions. This creates unexpected behavior.

Steps to reproduce:
Consider the following Map where 200000 points are randomly distributed across France. At zoom level 7 many more than 10k points are within the map border. When points are downsampled for display on the map they are all contained within a narrow latitude band

Expected behavior:

[elasticmachine]

Pinging @elastic/kibana-gis

[nreese]

@anottrott Have you tried creating a layer using the Grid aggregation source? This sounds like a better solution for your use case. See Plot big data without plotting too much data for more details about using aggregations to plot large data volumes.

The Grid aggregation source uses Elasticsearch GeoTile grid aggregation to group your documents into grids and then calculate metrics for each gridded cell. This is the only way to show results for large data sets that symbolize all data instead of just the first 10000 matching documents.

Then you can use multiple layers to toggle between aggregated views when zoom levels show large portions of the world and individual documents when zoom levels show smaller regions. See Add layers for Elasticsearch data for more details

[thomasneirynck]

thanks for flagging @anottrott

In 7.5, users can now define a sort-order for the features that are queried when mapping ES-documents. I would expect this would likely help with this issue (ie. not showing that strip of documents), as you would be able to define a sort-order that makes sense for your data #47361

It's unlikely that Maps would introduce randomizing results. Rather Maps should handle large datasets gracefully and more "automagically", and not have end-users run into this problem.

As @nreese suggested too in the docs he linked, this is a good use-case for using grid-aggregations. Cluster your results when zoomed out using a grid layer, and when zoomed in, show actual result sets using the es-document source.