Order by Relevance

The score column returned by paradedb.score can be used to sort results by BM25 relevance.

SELECT description, rating, category, paradedb.score(id)
FROM mock_items
WHERE description @@@ 'shoes'
ORDER BY score DESC
LIMIT 5;

Order by Field

The result set can be ordered by any field in ASC or DESC order. By default, Postgres orders by ASC.

SELECT description, rating, category
FROM mock_items
WHERE description @@@ 'shoes'
ORDER BY rating DESC
LIMIT 5;

Tiebreaking

Postgres can ORDER BY multiple columns to break ties in BM25 scores. In the following query, rows with the same score will be sorted by rating in descending order.

SELECT description, rating, category, paradedb.score(id)
FROM mock_items
WHERE category @@@ 'electronics'
ORDER BY score DESC, rating DESC
LIMIT 5;

Fast Ordering

An ORDER BY...LIMIT over a single text, numeric, datetime, or boolean field is automatically “pushed down” to the BM25 index if the ORDER BY field is indexed as fast. This makes these queries significantly faster.

You can verify if an ORDER BY...LIMIT was pushed down by running EXPLAIN on the query. If pushdown occurred, a Custom Scan with a Sort Field will appear in the query plan.

-- Pushdown may not occur over very small tables
-- This forces pushdown
SET enable_indexscan = off;

EXPLAIN SELECT description
FROM mock_items
WHERE description @@@ 'shoes'
ORDER BY rating DESC
LIMIT 5;

Ordering by Text Field

If a fast text field is indexed with the raw normalizer, ORDER BY <text_field> LIMIT can be pushed down.

If the lowercase normalizer is used, then ORDER BY lower(<text_field>) LIMIT (but not ORDER BY <text_field> LIMIT) can be pushed down.

CREATE INDEX search_idx ON mock_items
USING bm25 (id, description, category)
WITH (
    key_field='id',
    text_fields='{
        "category": {"fast": true, "normalizer": "lowercase"}
    }'
);

-- category uses normalizer = lowercase, so lower(category) can be pushed down
EXPLAIN SELECT description, rating, category
FROM mock_items
WHERE description @@@ 'shoes'
ORDER BY lower(category) DESC
LIMIT 5;

Not all ORDER BYs are pushed down. The following queries are not pushed down:

  1. ORDER BY without a LIMIT.

Partial Ordering with Multiple Sort Fields

When using ORDER BY with multiple sort fields, ParadeDB can partially push down the sorting operation. In this case, only the first column is pushed down to the BM25 index, and PostgreSQL handles the additional columns using sort operations.

For example, in the following query with multiple sort fields, sorting by sale_date is pushed down to the BM25 index, while sorting by amount is handled by PostgreSQL:

SELECT description, sale_date, amount, paradedb.score(id) as score
FROM sales
WHERE description @@@ 'laptop'
ORDER BY score, sale_date, amount
LIMIT 10;

You can verify if partial ORDER BY pushdown occurred by running EXPLAIN on the query. The query plan will show a Custom Scan with our ParadeDB scan provider, followed by an appropriate sort operation based on your PostgreSQL version:

  • In PostgreSQL 16+: Often uses an Incremental Sort node which can take advantage of the already-sorted first column
  • In older PostgreSQL: Uses a regular Sort node, but still benefits from our optimized ordering

This feature significantly improves performance when sorting by multiple columns, as the index is used for the first level of sorting, requiring PostgreSQL to perform less work to produce the final ordered results.

Limitations for partial ORDER BY pushdown:

  1. Only the first sort field is pushed down to the BM25 index.
  2. The first sort field must be indexed as a fast field.
  3. A LIMIT clause is still required.