arrow-cluster-layer

A deck.gl CompositeLayer for rendering clustered point data from Apache Arrow tables. Built on top of arrow-supercluster.

Why

Replaces the typical GeoJSON → Supercluster → GeoJsonLayer pipeline with Arrow Table → ArrowClusterEngine → binary ScatterplotLayer. No intermediate JS objects, no GeoJSON serialization. Typed arrays go straight to the GPU.

Install

# pnpm
pnpm add arrow-cluster-layer @deck.gl/core @deck.gl/layers apache-arrow

# npm
npm install arrow-cluster-layer @deck.gl/core @deck.gl/layers apache-arrow

# yarn
yarn add arrow-cluster-layer @deck.gl/core @deck.gl/layers apache-arrow

@deck.gl/core, @deck.gl/layers, and apache-arrow are peer dependencies.

arrow-supercluster is included as a direct dependency — you don't need to install it separately. Its types are re-exported for convenience.

Usage

import { ArrowClusterLayer } from "arrow-cluster-layer";
import type { Table } from "apache-arrow";

// Load your Arrow Table however you like (GeoParquet, IPC, etc.)
const table: Table = /* ... */;

const layer = new ArrowClusterLayer({
  id: "clusters",
  data: table,
  geometryColumn: "geometry",
  idColumn: "id",

  // Clustering
  clusterRadius: 75,
  clusterMaxZoom: 16,
  clusterMinZoom: 0,
  clusterMinPoints: 2,

  // Styling
  primaryColor: [26, 26, 64, 200],
  secondaryColor: [100, 100, 200, 200],
  selectedColor: [255, 140, 0, 230],
  textOpacity: 255,
  pointRadiusMinPixels: 10,
  pointRadiusMaxPixels: 100,

  // Interaction state
  selectedClusterId: null,
  focusedClusterId: null,

  // View
  viewType: "map", // or "globe"

  pickable: true,
});

Props

Data

Prop	Type	Default	Description
`data`	`arrow.Table`	required	Arrow Table with a GeoArrow Point geometry column
`geometryColumn`	`string`	`"geometry"`	Name of the geometry column
`idColumn`	`string`	`"id"`	Name of the ID column

Clustering

Prop	Type	Default	Description
`clusterRadius`	`number`	`75`	Cluster radius in pixels
`clusterMaxZoom`	`number`	`20`	Max zoom level for clustering
`clusterMinZoom`	`number`	`2`	Min zoom level for clustering
`clusterMinPoints`	`number`	`2`	Min points to form a cluster

Styling

Prop	Type	Default	Description
`primaryColor`	`ColorRGBA`	`[26, 26, 64, 200]`	Default fill color
`secondaryColor`	`ColorRGBA`	`[100, 100, 200, 200]`	Focused cluster + descendants color
`selectedColor`	`ColorRGBA`	`[255, 140, 0, 230]`	Selected cluster color
`textOpacity`	`number`	`255`	Cluster label text opacity (0-255)
`pointRadiusMinPixels`	`number`	`10`	Minimum circle radius in pixels
`pointRadiusMaxPixels`	`number`	`100`	Maximum circle radius in pixels

Interaction

Prop	Type	Default	Description
`selectedClusterId`	`number \| null`	`null`	Currently selected cluster ID
`focusedClusterId`	`number \| null`	`null`	Currently focused (hovered) cluster ID
`viewType`	`"map" \| "globe"`	`"map"`	Flips text labels on globe view at low zoom

Picking

The layer returns ArrowClusterPickingInfo from onHover and onClick:

interface ArrowClusterPickingInfo extends PickingInfo {
  isCluster: boolean; // cluster or individual point?
  clusterId: number; // encoded cluster ID or Arrow row index
  pointCount: number; // points in cluster (1 for individual)
  arrowIndices: number[]; // Arrow row indices of all leaf points
  rows: StructRowProxy[]; // materialized Arrow rows
}

Example:

onClick: (info) => {
  const pick = info as ArrowClusterPickingInfo;
  if (pick.isCluster) {
    // Zoom into the cluster
    const zoom = layer.getClusterExpansionZoom(pick.clusterId);
    // ... fly to pick.coordinate at zoom
  } else {
    // Individual point — access the Arrow row
    const row = pick.rows[0];
    console.log(row.id, row.city, row.description);
  }
};

Public Methods

`layer.getClusterExpansionZoom(clusterId) → number`

Returns the zoom level at which a cluster splits into its children. Useful for zoom-on-click behavior.

Rendering

The layer renders two sublayers:

ScatterplotLayer — circles with binary attribute buffers (positions, radii, colors as typed arrays)
TextLayer — cluster count labels with SDF fonts

Radius scales logarithmically with point count: baseSize + (log(count+1) / log(total+1)) * scaleFactor.

Text color is automatically chosen for contrast (black or white) based on the fill color's relative luminance.

Style Helpers

The style computation functions are exported if you need to customize or extend them:

import {
  computeFillColors,
  computeRadii,
  computeTextColors,
  computeTexts,
} from "arrow-cluster-layer";

Re-exports

The engine and its types are re-exported for convenience:

import {
  ArrowClusterEngine,
  type ClusterOutput,
  type ArrowClusterEngineOptions,
} from "arrow-cluster-layer";

License

ISC