UUID

Generate a new random UUID (version 4).

Syntax

UUID()

Description

## Overview Generates a new random UUID (Universally Unique Identifier) in the standard 8-4-4-4-12 hexadecimal format. Each call returns a fresh, independent value. Use this function to create surrogate primary keys, correlation IDs, idempotency tokens, or any situation that needs a globally unique identifier without coordination. UUID is a dialect-neutral alias for GEN_RANDOM_UUID and UUID_GENERATE_V4 and produces the same value format and randomness guarantees. ## Behavior - Non-deterministic: every call returns a different value, even within the same row or the same statement. - Uses a cryptographically secure random source. - Returns a 36-character lowercase hyphenated string (version 4, variant 10xx). - Never returns NULL. - Side effect free (does not consult a sequence, file, or external service). ## Compatibility - Dialect-neutral name. PG-compat callers may prefer GEN_RANDOM_UUID; the function UUID_GENERATE_V4 is available for code ported from engines that use that name.

Examples

-- Generate a single UUID
SELECT UUID();

-- Stamp a request log with a correlation id
SELECT UUID() AS request_id, 'process_start' AS event;

-- Freeze a UUID in a CTE so the same value is reused in multiple projections
WITH id AS (SELECT UUID() AS v) SELECT id.v, id.v FROM id;

-- Validate freshly generated UUIDs round-trip through IS_UUID
SELECT IS_UUID(UUID()) AS ok;  -- true

-- Use as primary key on insert
INSERT INTO ops.audit.events (event_id, actor, occurred_at)
VALUES (UUID(), CURRENT_USER(), CURRENT_TIMESTAMP);

Pitfalls

• UUID() is non-deterministic and is re-evaluated on every reference. `SELECT UUID() AS a, UUID() AS b` produces two different values; wrap the call in a CTE or subquery if you need the same value in multiple projections.
• Repeated references within the same row do not return the same value. `SELECT UUID(), UUID()` yields two distinct UUIDs, which is often the intent but can surprise readers coming from deterministic scalar functions.
• UUID() values are not time-ordered. Using them as a clustering key can cause poor data layout and fragmented indexes. Consider timestamp-prefixed identifiers for write-heavy time-series workloads.
• UUIDs are 36 characters long. Downstream columns that store them as VARCHAR(36) will fail if the engine ever emits a format without hyphens (32 characters).

See Also

• gen_random_uuid
• uuid_generate_v4
• is_uuid