Geohex grid aggregation on geo_shape MVP

[craigtaverner]

Elasticsearch already [supports geohex grid aggregations|https://www.elastic.co/guide/en/elasticsearch/reference/current/search-aggregations-bucket-geohexgrid-aggregation.html], but only on geo_point. There is value in supporting this on fields of type geo_shape too. Since the related [geotile|https://www.elastic.co/guide/en/elasticsearch/reference/current/search-aggregations-bucket-geotilegrid-aggregation.html] and [geohash|https://www.elastic.co/guide/en/elasticsearch/reference/current/search-aggregations-bucket-geohashgrid-aggregation.html] aggregations work with both geo_point and geoshapewe can apply similar thinking to thegeohex` aggregation.

The main challenge is that both geohash and geotile work with euclidean space (mercator projection) while geohex is very much more spherical. As a result the underlying maths needs to take this into account. The approach used by the existing geo_point support is to connect the existing 2D euclidean aggregation support (using Lucene's Component2D) to the underlying maths provided by Lucene's spatial3d library. Currently we support geo_point aggregations by implementing only a small part of the Component2D interface. Supporting geo_shape could require completing this support.

Update: This initial plan of using spatial3d for geohex over geo_shape was used in the prototype, but during that work we realised that polygons stored in elasticsearch are indexed with edges assumed to be lines in equirectangular projection. In order to ensure that aggregations match queries, we need to perform the aggregation in equirectangular projection too. Initially this was attempted in the prototype below, maintaining both spherical and equirectangular approaches in parallel. However, this proved to be unnecessarily complex, and performed badly (slow queries), so a new prototype was built with equirectangular-only and a focus on performance with dramatically improved results. This second attempt became the base of the production version.

h3. Sub-tasks:

h4. [Prototype|https://github.com//pull/90026]:

• Complete Component2D support in the H3LatLonGeometry class
• Unit tests of all new supporting code for this Component2D work
• Create infrastructure for supporting geo_shape in geohex grid aggregations
• Some basic test coverage
• Discuss prototype learnings

h4. Make production ready

• Any changes needed based on prototype learnings
• [Create mercator projection version of H3LatLonGeometry|https://github.com/Geohex aggregation on geo_shape field #91956]
• Add full test coverage
• [Benchmarking|https://github.com/Added six geo benchmarks, three geo_grid and three grid aggregations rally-tracks#367]
• Performance enhancements
• [Documentation|https://github.com/Documentation for geohex_grid over geo_shape #92999]

Marketing

• Blog
• Release notes

[elasticsearchmachine]

Pinging @elastic/es-analytics-geo (Team:Analytics)