[Maps] point-to-point source can exceed default search.max_buckets and fail

[nreese]

Point-to-point source fetches the top 100 destination points and then runs a geotile_grid aggregation for each bucket. The geotile_grid aggregation is limited to top 500 buckets. Even with these constraints in place, the request could return 50,000 buckets. This will result in an exception for the request, because the default search.max_buckets is breached.

point-to-point should use composite aggregation to break the request into pieces to ensure the default search.max_buckets is not breached.

With composite aggs, point-to-point could also be expanded to support more than 100 destinations.

[elasticmachine]

Pinging @elastic/kibana-gis

[thomasneirynck]

Originally in separate ticket (#47895), but will paste this here as that is a proposal to address this issue (different from using _scroll)

---

The point-2-point layer uses a grid aggregation nested under a terms aggregation to construct a line from all origins to a single unique destination.

This creates an approximate line to a destination that gets finer and finer as users zoom in.

There are worst-case scenarios where this strains Elasticsearch.

• when zoomed in, the origins are aggregated on a really fine grid. When the origins fall outside the bounds, they are not even visible.
• when there are many nearby destinations, with slightly varying coordinates., These are really just noise when zoomed out.

Consider:

(a) Add a spatial filter so the origins need to fall inside the bounds, and not just the destination. The result is the line would only be visible if both destination and origin are in the bounds
(b) The top level aggregation is not a term-agg on destinations, but a grid-agg itself, using the geo-centroid to approximate the location
These changes would not really impact negatively on the purpose of this layer-type (imo).

(a) Already lines will "blip"-in and out depending on whether the destination is inside the view. This just tightens this restriction.
(b) The exact destination location is not required to preserve the visual effect. Only the aggregate values for the line. Zooming in on the destination will refine the grid.

The advantage is that it cuts down on the size of the aggregration, especially when zoomed in, or when there are many destinations.

[nreese]

closed by elastic/elasticsearch#57042