Property, range and composite indexes

Wiki / Indexes and constraints

On this page

Property indexes

A property index on (label, property) turns MATCH (n:Label {prop: v}) and WHERE n.prop = v from an O(|nodes|) scan into an O(1)/O(log n) lookup.

CREATE INDEX ON :Person(email);     // declare (backfills existing data)
SHOW INDEXES;                       // every index, with its name

The planner uses an index automatically when a pattern/predicate allows it. Confirm with EXPLAIN:

EXPLAIN MATCH (p:Person {email: $e}) RETURN p;
// Match: NodeIndexSeek(:Person .email)      <- index is used

Notes:

  • Creating an index backfills existing matching nodes — every node that carries the label at that moment, under each of its labels — and every later write keeps it current. Only nodes are indexed: there are no relationship property indexes, and an unlabelled node is never indexed. Re-declaring an existing index is accepted but re-scans the label.
  • A built-in label/token index is always on (not declared), making MATCH (n:Label) cost O(label population) rather than O(all nodes); the planner anchors a multi-label pattern on its rarest label.
  • The plain CREATE INDEX index is hash/equality only. For range queries, declare a range index (Property, range and composite indexes). Composite non-unique indexes are supported (see below); composite uniqueness uses node keys, Constraints and shared uniqueness.

Range (ordered) indexes

A range index is ordered, so it accelerates </<=/>/>= predicates (and equality) — the planner uses it when a clause WHERE ranges over the indexed property.

CREATE RANGE INDEX person_age FOR (p:Person) ON (p.age);   // 5.x form; the name is kept
SHOW INDEXES;                                    // the `type` column shows RANGE vs BTREE
EXPLAIN MATCH (p:Person) WHERE p.age >= 18 AND p.age < 65 RETURN p;
// Match: NodeIndexSeekByRange(:Person .age)     <- range index is used

When the query returns its rows sorted by the indexed property and keeps only the first page, the index is read in that order and the scan stops at the page (from version 2.2.1). PROFILE shows it as OrderedRangeScan:

PROFILE MATCH (o:Order) WHERE o.placed >= date('2026-01-01')
RETURN o.ref, o.placed ORDER BY o.placed LIMIT 20;
// OrderedRangeScan (reads the range index in ORDER BY order, stops at the page)

It reads the first twenty matches in index order, plus any that tie with the twentieth, rather than every order since January. It applies to a single node pattern with a range predicate on the sorted property, an ascending sort, LIMIT (and optional SKIP) of at most 4,096 rows, and plain property reads in RETURN and WHERE; other queries sort as before and return the same rows.

STARTS WITH is a range in disguise, and plans the same seek: the prefix desugars to >= prefix AND < successor(prefix) internally, for a literal or a parameter prefix alike.

CREATE RANGE INDEX FOR (p:Person) ON (p.name);
EXPLAIN MATCH (p:Person) WHERE p.name STARTS WITH 'Al' RETURN p;
// Match: NodeIndexSeekByRange(:Person .name)    <- prefix seek

Composite (multi-property) indexes

Both kinds take a property list, giving a composite index over several properties of a label:

CREATE INDEX FOR (p:Person) ON (p.first, p.last);        // composite hash
CREATE RANGE INDEX FOR (e:Event) ON (e.category, e.day); // composite range
  • A composite index covers a node only when it carries the label and every indexed property — a node missing one is simply not indexed.

  • The planner uses a composite hash index when every property is bound by equality (WHERE p.first = $f AND p.last = $l, or the inline form (:Person {first: $f, last: $l})), and prefers it over a single-property index when it binds more properties:

    EXPLAIN MATCH (p:Person) WHERE p.first = 'Ada' AND p.last = 'Lovelace' RETURN p;
    // Match: NodeIndexSeek(:Person (first, last))
    
  • A composite range index serves a leading-prefix equality plus an optional range on the next property (WHERE e.category = 'a' AND e.day >= 5). A node missing any indexed property is not in the index, so from version 1.2.0 the index is used only when the predicate constrains every indexed property, or an existence or node-key constraint guarantees the rest. Earlier versions used it for a prefix alone and could miss nodes that lacked a later property; upgrade if you rely on composite range indexes.

  • SHOW INDEXES lists a composite index with its properties joined in the property column. Composite uniqueness is provided separately by node-key constraints (Constraints and shared uniqueness.

Removing indexes

After the examples, remove indexes by name, DROP INDEX person_age, or by definition: DROP INDEX ON :Person(email), DROP RANGE INDEX FOR (p:Person) ON (p.age) or DROP INDEX FOR (p:Person) ON (p.first, p.last). A constraint-backed index cannot be dropped while its constraint exists.

EXPLAIN and PROFILE · Resident and hybrid storage · Schema and maintenance procedures

Planning a deployment? Review compatibility and licence setup for your instance.