Optimize sibling aggs together

[nik9000]

If we have sibling aggregations running on the same field there are a few cases where we could collect them with a single aggregation and then split the results. For example, if you run the min and max aggregations as siblings on the same field you could collect a stats aggregation and then split out the results on read. Similarly, if there are two percentiles aggregations as siblings we could combine them into one for collection.

[elasticmachine]

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

[iverase]

FWIW: I observed something similar when developing vector tiles. There are some cases where we are running a GeoGrid aggregation with a sibling GeoBounds aggregation. In some cases even the GeoGrid aggregations has a child GeoCentroid aggregation.

For each aggregation we are having to read and decode the doc value (which for geo_shape might be expensive) so optimising them together would be a big win.

[flash1293]

The most relevant case for the Lens use case is merging separate percentiles aggs on the same field.

e.g.
{
a: { percentiles: { field: x, values: [1,2,3] } },
b: { percentiles: { field: x, values: [4,5,6] } },
}

and
{
a: { percentiles: { field: x, values: [1,2,3,4,5,6] } }
}

should have near-identical performance

[wchaparro]

we discussed this one as a team, adding to our roadmap as its something we believe we want to do, but we need to spend some time investigating a solution and coming up with a plan for this. removing the team-discuss.

[wchaparro]

Closing as not planned, focus is on ES|QL development.