tdigest

Overview

Version

Install

You can install tdigest directly. First, make sure the PGDG repository is added and enabled:

pig repo add pgdg -u          # Add PGDG repo and update cache

Install the extension using pig or apt/yum/dnf:

pig install tdigest;          # Install for current active PG version

pig ext install -y tdigest -v 18  # PG 18
pig ext install -y tdigest -v 17  # PG 17
pig ext install -y tdigest -v 16  # PG 16
pig ext install -y tdigest -v 15  # PG 15
pig ext install -y tdigest -v 14  # PG 14

dnf install -y tdigest_18       # PG 18
dnf install -y tdigest_17       # PG 17
dnf install -y tdigest_16       # PG 16
dnf install -y tdigest_15       # PG 15
dnf install -y tdigest_14       # PG 14

apt install -y postgresql-18-tdigest   # PG 18
apt install -y postgresql-17-tdigest   # PG 17
apt install -y postgresql-16-tdigest   # PG 16
apt install -y postgresql-15-tdigest   # PG 15
apt install -y postgresql-14-tdigest   # PG 14

Create Extension:

CREATE EXTENSION tdigest;

Usage

Sources:

• tdigest v1.4.4 README
• v1.4.4 release
• Extension control file

tdigest implements an approximate, mergeable t-digest for online rank statistics such as quantiles, percentile ranks, and trimmed means. It supports parallel aggregation and storing pre-aggregated digests for later rollups.

CREATE EXTENSION tdigest;

Direct Aggregation Functions

Pre-aggregation Functions

Incremental Update Functions

Utility Functions

Parameters

• compression – controls accuracy (higher = more accurate, larger digest). Error is roughly 1/compression.

Examples

-- Instead of: SELECT percentile_cont(0.95) WITHIN GROUP (ORDER BY a) FROM t;
SELECT tdigest_percentile(a, 100, 0.95) FROM t;

-- Multiple percentiles
SELECT tdigest_percentile(a, 100, ARRAY[0.5, 0.95, 0.99]) FROM t;

-- Pre-aggregate for fast repeated queries
CREATE TABLE p AS SELECT a, b, tdigest(c, 100) AS d FROM t GROUP BY a, b;

-- Query pre-aggregated data (~1.5ms vs ~7s for exact)
SELECT a, tdigest_percentile(d, 0.95) FROM p GROUP BY a ORDER BY a;

Caveats

• Results are estimates. Validate the chosen compression against exact percentile_cont results on representative data before setting accuracy targets.
• Higher compression usually improves tail accuracy but increases state size and CPU cost.
• Stored digests can be merged across groups and time windows. Version 1.4.4 fixes combining digests created with different parameters, so use that patch level when heterogeneous states may meet.
• Version 1.4.4 also strengthens text-input parsing and validation and adds PostgreSQL 19 build/test coverage; malformed serialized digests that older builds accepted may now be rejected.