Trino DataSketches Plugin

Apache DataSketches probabilistic data structures as Trino SQL functions.

Overview

This plugin provides 65 SQL functions across 8 sketch families for Trino 479, enabling approximate computations on massive datasets:

Sketch Family	Use Case
HLL	Cardinality estimation (count distinct)
Theta	Cardinality with set operations (union, intersect, exclude)
CPC	Compact cardinality estimation
KLL	Quantile approximation (percentiles, ranks, CDF, PMF)
Quantiles	Classic quantile approximation (DoublesSketch)
Frequencies	Frequent items / heavy hitters
Tuple ArrayOfDoubles	Cardinality with associated numeric values
Tuple DoubleSummary	Cardinality with summary statistics

Why Sketches?

Probabilistic data structures (sketches) let you compute approximate answers to queries like “count distinct” or “what’s the 99th percentile” in a single pass over the data, using a fraction of the memory that exact computation requires. They are:

Mergeable — pre-aggregate sketches, then combine them for any time range or dimension
Fast — single-pass, no sorting or shuffling needed
Compact — a sketch of 10 billion values fits in a few KB
Accurate — typical error is 1-3% for cardinality, tighter for quantiles

Compatibility

Component	Version
Trino	479
datasketches-java	9.0.0
Java	21+