Elasticsearch Version

8.9.0

Installed Plugins

No response

Java Version

bundled

OS Version

all

Problem Description

In #96953 we added support for a much more memory efficient algorithm that relies on the time-ordering and grouping provided by the time-series aggregation in TSDB. However, this work only covered the data nodes, and the reduce phase in the coordinating nodes that merges geo_lines does not take this into account, leading to unexpected truncation of long geo_lines.

Steps to Reproduce

Easiest to reproduce by setting the shard size small, loading very long geolines that will be composed of many data points over many shards, and then setting the geo_line aggregation size to a small size. The final lines will not include the two end points as they should, but be truncated.

Logs (if relevant)

No response