Elasticsearch Geo Point clustering aggregation plugin

This plugin extends Elasticsearch with a geo_point_clustering aggregation, allowing to fetch geo_point documents as clusters of points. It is very similar to what is done with the official geohash_grid aggregation except that final clusters are not bound to the geohash grid.

For example, at zoom level 1 with points across France, geohash_grid agg will output 3 clusters stuck to geohash cells u, e, s, while geo_point_clustering will merge these clusters into one. This is done during the reduce phase.

Contrary to geohash_grid aggregation, buckets keys are a tuple(centroid, geohash cells) instead of geohash cells only, because one cluster can be linked to several geohash cells, due to the cluster merge process during the reduce phase. Centroids are built during the shard collect phase.

Please note that geo_shape data type is not supported.

Usage

Install

Install plugin with: ./bin/elasticsearch-plugin install https://github.com/opendatasoft/elasticsearch-aggregation-geoclustering/releases/download/v8.19.6.0/geopoint-clustering-aggregation-8.19.6.0.zip

Quickstart

Intro

{
  "aggregations": {
    "<aggregation_name>": {
      "geo_point_clustering": {
        "field": "<field_name>",
        "zoom": "<zoom>"
      }
    }
  }
}

Input parameters :

• field: must be of type geo_point
• zoom: mandatory integer parameter between 0 and 25. It represents the zoom level used in the request to aggregate geo points
• radius: radius in pixel. It is used during the reduce phase to merge close clusters. Default to 40
• ratio: ratio used to make a second merging pass during the reduce phase. If the value is 0, no second pass is made. Default to 0
• extent: Extent of the tiles. Default to 256

Real-life example

Create an index:

PUT test
{
  "settings": {
    "number_of_shards": 1,
    "number_of_replicas": 0
  },
  "mappings": {
    "properties": {
      "location": {
        "type": "geo_point"
      }
    }
  }
}

Push some points:

POST test/_bulk?refresh
{"index":{"_id":1}}
{"location":[2.454929, 48.821578]}
{"index":{"_id":2}}
{"location":[2.245858, 48.86914]}
{"index":{"_id":3}}
{"location":[2.240358, 48.863481]}
{"index":{"_id":4}}
{"location":[2.25292, 48.847176]}
{"index":{"_id":5}}
{"location":[2.279111, 48.872383]}
{"index":{"_id":6}}
{"location":[2.336267, 48.822021]}
{"index":{"_id":7}}
{"location":[2.338677, 48.822672]}
{"index":{"_id":8}}
{"location":[2.336643, 48.822493]}
{"index":{"_id":9}}
{"location":[2.438465, 48.84204]}
{"index":{"_id":10}}
{"location":[2.381554, 48.835382]}
{"index":{"_id":11}}
{"location":[2.407744, 48.83733]}
{"index":{"_id":12}}
{"location":[2.34521, 48.849358]}
{"index":{"_id":13}}
{"location":[2.252938, 48.846041]}
{"index":{"_id":14}}
{"location":[2.279715, 48.871775]}
{"index":{"_id":15}}
{"location":[2.380629, 48.879757]}

Perform an aggregation:

POST test/_search?size=0
{
  "aggregations": {
    "clusters": {
      "geo_point_clustering": {
        "field": "location",
        "zoom": 9
      }}}}

Result:

"aggregations" : {
    "clusters" : {
      "buckets" : [
        {
          "geohash_grids" : [
            "u09wn",
            "u09tz",
            "u09ty",
            "u09tx",
            "u09tv",
            "u09tt"
          ],
          "doc_count" : 9,
          "centroid" : {
            "lat" : 48.83695897646248,
            "lon" : 2.380013056099415
          }
        },
        {
          "geohash_grids" : [
            "u09w5",
            "u09tg",
            "u09tf"
          ],
          "doc_count" : 6,
          "centroid" : {
            "lat" : 48.86166598415002,
            "lon" : 2.258483301848173
          }
        }
      ]
    }

Development environment setup

Build

Built with Java 17. Apply spotless code formatting with:

./gradlew spotlessApply

Development Environment Setup

Build the plugin using gradle:

./gradlew build

or

./gradlew assemble  # (to avoid the test suite)

In case you have to upgrade Gradle, you can do it with ./gradlew wrapper --gradle-version x.y.z.

Then the following command will start a dockerized ES and will install the previously built plugin:

docker compose up

You can now check the Elasticsearch instance on localhost:9200 and the plugin version at localhost:9200/_cat/plugins.

Please be careful during development: you'll need to manually rebuild the .zip using ./gradlew build on each code change before running docker compose up up again.

NOTE: In docker-compose.yml you can uncomment the debug env and attach a REMOTE JVM on *:5005 to debug the plugin.