pg_rrf

pg_rrf : Reciprocal rank fusion functions for hybrid search

Overview

Packages

Source

pig build pkg pg_rrf;		# build rpm/deb

Install

Make sure PGDG and PIGSTY repo available:

pig repo add pgsql -u   # add both repo and update cache

Install this extension with pig:

pig install pg_rrf;		# install via package name, for the active PG version

pig install pg_rrf -v 17;   # install for PG 17
pig install pg_rrf -v 16;   # install for PG 16
pig install pg_rrf -v 15;   # install for PG 15
pig install pg_rrf -v 14;   # install for PG 14

Create this extension with:

CREATE EXTENSION pg_rrf;

Usage

Sources: README, v0.0.3 release

pg_rrf provides Reciprocal Rank Fusion functions for hybrid search score fusion. It is focused on combining ranked candidate lists without hand-written FULL OUTER JOIN / COALESCE plumbing.

Core Functions

• rrf(rank_a, rank_b, k)
• rrf3(rank_a, rank_b, rank_c, k)
• rrf_fuse(ids_a bigint[], ids_b bigint[], k int default 60)
• rrfn(ranks bigint[], k int)

The v0.0.3 release explicitly adds rrfn while keeping rrf and rrf3 as compatibility wrappers. The README documents the same score behavior:

• missing ranks are ignored
• ranks <= 0 are ignored
• k <= 0 raises an error

Example

CREATE EXTENSION pg_rrf;

SELECT rrf(1, 2, 60) AS rrf_12;
SELECT rrf3(1, 2, 3, 60) AS rrf_123;
SELECT rrfn(ARRAY[1, 2, 3], 60) AS rrfn_123;
SELECT *
FROM rrf_fuse(ARRAY[10, 20, 30], ARRAY[20, 40], 60)
ORDER BY score DESC;

Hybrid Search Pattern

The upstream README shows rrf_fuse as a replacement for a manual fusion query:

WITH fused AS (
  SELECT *
  FROM rrf_fuse(
    ARRAY(SELECT id FROM docs ORDER BY bm25_score DESC LIMIT 100),
    ARRAY(SELECT id FROM docs ORDER BY embedding <=> :qvec LIMIT 100),
    60
  )
)
SELECT d.*, fused.score
FROM fused
JOIN docs d USING (id)
ORDER BY fused.score DESC
LIMIT 20;

Notes

The README targets PostgreSQL 14-17 and documents Docker-based build and test flows. The extension surface remains intentionally small: score helpers plus rrf_fuse for the common two-list hybrid-search pattern.