Memgraph Indexes And Constraints

Indexes and Constraints in Memgraph

Indexes

Indexes are NOT created automatically unless explicitly enabled via flags. Adding an index speeds reads but slows writes and uses extra memory.

Index types

Show: SHOW INDEX INFO; or SHOW INDEXES;

Label-property index

CREATE INDEX ON :Person(age);

Creating a label-property index does NOT create a label index - create both if needed. Best performance on high-cardinality properties (unique IDs, names). Avoid on booleans or low-cardinality fields.

Composite index

CREATE INDEX ON :Person(name, occupation);

Follows the leftmost prefix rule:

Put highest-cardinality property first.

Descending index

CREATE INDEX ON :Person(age) WITH CONFIG {"order": "DESC"};

Optimizes ORDER BY prop DESC and top-N queries. ASC and DESC can coexist on the same label+property. Only supported in IN_MEMORY_TRANSACTIONAL.

Dropping without config removes both ASC and DESC variants.

Nested (map) property index

CREATE INDEX ON :Project(delivery.status.due_date);

Must use WHERE clause to query - inline map matching won't use the index:

-- Wrong (compares entire map):
MATCH (p:Project {delivery: {status: {due_date: date("2025-06-04")}}}) RETURN p;

-- Correct:
MATCH (p:Project) WHERE p.delivery.status.due_date = date("2025-06-04") RETURN p;

Edge indexes

Require --storage-properties-on-edges=true.

CREATE EDGE INDEX ON :KNOWS;
CREATE EDGE INDEX ON :KNOWS(since);
CREATE GLOBAL EDGE INDEX ON :(weight);

Point index

CREATE POINT INDEX ON :Location(coords);
MATCH (n:Location) WHERE point.distance(point({x:1, y:1}), n.coords) < 1000 RETURN n;
MATCH (n:Location) WHERE point.withinbbox(n.coords, point({x:0, y:0}), point({x:10, y:10})) RETURN n;

Index hinting

USING INDEX :Person(name)
MATCH (n:Person {name: "Alice", gender: "F"}) RETURN n;

ANALYZE GRAPH

Run once after indexes are created and data is loaded:

ANALYZE GRAPH;
ANALYZE GRAPH ON LABELS :Person, :City;
ANALYZE GRAPH DELETE STATISTICS;

Calculates property value distribution for optimal index and MERGE selection. Statistics persist via snapshots/WAL.

Automatic index creation

Only in IN_MEMORY_TRANSACTIONAL. Flags:

• --storage-automatic-label-index-creation-enabled
• --storage-automatic-edge-type-index-creation-enabled

Concurrent index creation

Label, label-property, composite, and edge indexes are created concurrently. Brief READ ONLY during registration phase (waits for pending writes), then background indexing proceeds while reads and writes resume.

Constraints

Existence constraint

CREATE CONSTRAINT ON (n:Employee) ASSERT EXISTS (n.first_name);
DROP CONSTRAINT ON (n:Employee) ASSERT EXISTS (n.first_name);
SHOW CONSTRAINT INFO;

One label + one property at a time.

Uniqueness constraint

CREATE CONSTRAINT ON (n:Employee) ASSERT n.email IS UNIQUE;
CREATE CONSTRAINT ON (n:Employee) ASSERT n.name, n.address IS UNIQUE;
DROP CONSTRAINT ON (n:Employee) ASSERT n.email IS UNIQUE;

Multi-property: same name OR same address is allowed; same name AND address is forbidden. Uniqueness constraints do NOT create indexes - add them separately.

Data type constraint

CREATE CONSTRAINT ON (n:Person) ASSERT n.name IS TYPED STRING;
CREATE CONSTRAINT ON (n:Person) ASSERT n.age IS TYPED INTEGER;
DROP CONSTRAINT ON (n:Person) ASSERT n.name IS TYPED STRING;

Supported types: NULL, STRING, BOOLEAN, INTEGER, FLOAT, LIST, MAP, DURATION, DATE, LOCALTIME, LOCALDATETIME, ZONEDDATETIME, ENUM, POINT.

Only one type constraint per label-property pair.

Drop all

DROP ALL CONSTRAINTS;

Constraint behavior

Constraints are checked optimistically on commit. In multi-query explicit transactions, intermediate violations are allowed as long as the final state satisfies all constraints.

Enums

CREATE ENUM Status VALUES { Good, Okay, Bad };
SHOW ENUMS;
ALTER ENUM Status ADD VALUE Excellent;
ALTER ENUM Status UPDATE VALUE Bad TO Poor;

Use in queries:

CREATE (:Machine {status: Status::Good});
MATCH (n:Machine) WHERE n.status = Status::Bad RETURN n;
RETURN ToEnum("Status", "Good");
RETURN ToEnum("Status::Okay");

Less memory and faster comparison than string properties.

Storage access

Index/constraint/enum DDL requires exclusive (unique) access - briefly blocks other queries during execution. Normal Cypher queries use shared access.

Timeout for write access during index creation:

SET DATABASE SETTING 'storage.access_timeout_sec' TO '30';

Key rules

1. Indexes are not auto-created unless flags are enabled (IN_MEMORY_TRANSACTIONAL only)
2. Uniqueness constraints do NOT create indexes - add them separately
3. Label-property index does NOT imply a label index
4. Composite index follows the leftmost prefix rule
5. Nested map indexes require WHERE clause, not inline map matching
6. Edge indexes require --storage-properties-on-edges=true
7. Descending indexes only work in IN_MEMORY_TRANSACTIONAL (see memgraph-database-configuration)
8. Run ANALYZE GRAPH once after data load for optimal index selection
9. DROP INDEX without config drops both ASC and DESC variants
10. Constraints are checked on commit (optimistic)