GiST Index

GiST (Generalized Search Tree) is a family of PostgreSQL indexes designed for data that cannot be ordered linearly: geometries, ranges, similarity vectors, full-text. It’s a balanced tree that organises data by hierarchical bounding boxes instead of lexicographic ordering.

How it works #

While a B-tree orders values from “minimum” to “maximum” and performs dichotomous search, GiST groups data into nested regions (bounding boxes). Each tree node represents a region containing all data in its children. When searching for a value, GiST discards entire regions with an overlap comparison — without descending into nodes that can’t contain the result.

This structure can index:

• Geometries: points, polygons, lines (with PostGIS)
• Ranges: int4range, tsrange, daterange and other range types
• Full-text: tsvector vectors for text search
• Similarity: with extensions like pg_trgm for approximate searches

What it’s for #

It solves “spatial” or range queries that a B-tree cannot handle:

• Find all points inside a box or radius
• Find all records whose range overlaps with another range
• Find texts similar to a query, even with typos
• Search by containment: range1 @> range2 or geom1 && geom2

When to use it #

Use it with CREATE INDEX ... USING GIST (column). It’s the natural complement of GIN: GIN for array/JSONB containment, GiST for geometry/range/similarity. On high-churn tables it has write costs similar to GIN — so it should be evaluated case by case.