UUIDs in MySQL

How to use UUIDs as primary keys in MySQL — storage strategies, performance tradeoffs, and UUID versions explained.

VillageSQL is a drop-in replacement for MySQL with extensions.

All examples in this guide work on VillageSQL. Install Now →

Auto-increment integers are simple until they’re not: they leak row counts, fail across distributed systems, and break when you merge databases or shard. UUIDs solve all of that, but MySQL’s default handling makes them slow to insert at scale. Here’s why that happens and how to fix it.

The MySQL Default: UUID() as CHAR(36)

MySQL’s built-in UUID() function returns a v1 UUID as a 36-character string like 550e8400-e29b-41d4-a716-446655440000. Storing that in a CHAR(36) column works, but it’s inefficient:

• 36 bytes instead of 16 for the same data
• String comparison is slower than binary comparison
• Random UUIDs cause B-tree page splits on every insert (more on this below)

-- This works but has problems at scale
CREATE TABLE orders (
    order_id CHAR(36) PRIMARY KEY DEFAULT (UUID()),
    user_id  INT NOT NULL
);

The Standard Workaround: BINARY(16)

MySQL provides UUID_TO_BIN() and BIN_TO_UUID() to convert between string and compact binary form. Using BINARY(16) cuts storage in half and speeds up comparisons:

CREATE TABLE orders (
    order_id BINARY(16) PRIMARY KEY,
    user_id  INT NOT NULL
);

-- Store: convert string UUID to binary
INSERT INTO orders (order_id, user_id)
VALUES (UUID_TO_BIN(UUID()), 101);

-- Retrieve: convert back to readable format
SELECT BIN_TO_UUID(order_id), user_id FROM orders;

UUID_TO_BIN() takes an optional second argument (1) that reorders the timestamp bits so v1 UUIDs sort chronologically — a partial fix for the insert performance problem.

The Remaining Problem: Random Inserts Fragment Indexes

MySQL’s InnoDB uses a clustered index, meaning rows are physically stored in primary key order. When you insert a row with a random UUID, InnoDB has to find where it goes in the middle of the index and potentially split a page to make room. Under sustained load, this causes write amplification and index fragmentation.

UUID v7 fixes this by embedding a Unix timestamp in the high bits of the identifier. New UUIDs are always larger than existing ones, so inserts always append to the rightmost leaf of the index — no splits.

With VillageSQL: Native UUID Type and All Versions

VillageSQL’s vsql_uuid extension adds a native UUID column type with 16-byte binary storage and no manual conversion, plus generation functions for UUID v1 through v7.

INSTALL EXTENSION vsql_uuid;

-- Native UUID type — no UUID_TO_BIN/BIN_TO_UUID needed
CREATE TABLE orders (
    order_id UUID PRIMARY KEY,
    user_id  INT NOT NULL,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

-- UUID_V7 inserts always append to the index — no page splits
INSERT INTO orders (order_id, user_id) VALUES (UUID_V7(), 101);
INSERT INTO orders (order_id, user_id) VALUES (UUID_V7(), 102);

-- Sorts chronologically by creation order
SELECT user_id, created_at FROM orders ORDER BY order_id;

-- Inspect the timestamp embedded in a v7 UUID (as datetime or Unix epoch)
SELECT UUID_TIMESTAMP(order_id), UUID_EPOCH(order_id) FROM orders LIMIT 1;

Choosing a UUID version

Version	Use when
UUID_V4()	You need random IDs and don’t care about insert performance
UUID_V7()	New rows, primary keys, high insert volume — recommended default
UUID_V6()	Legacy v1 systems that need sortable output
UUID_V1()	Compatibility with systems that expect v1 format
UUID_V3() / UUID_V5()	Deterministic IDs derived from a name (v3 = MD5, v5 = SHA-1)

UUID_VERSION(uuid) returns the version number of any UUID — useful for validating IDs arriving from external systems. UUID_COMPARE(uuid1, uuid2) returns -1, 0, or 1 for explicit ordering logic. For a broader look at when to use UUIDs versus other primary key approaches, see Choosing a Primary Key Strategy.

These introspection functions are deterministic, so — unlike the generators (UUID_V4(), UUID_V7(), and friends, whose output changes on every call) — they can appear in generated columns and CHECK constraints. That lets you derive or validate stored values from a UUID at the schema level:

CREATE TABLE audit_log (
  id uuid NOT NULL,
  ver INT AS (UUID_VERSION(id)) STORED,  -- derive the version into a column
  CHECK (UUID_VERSION(id) IN (4, 7))     -- accept only v4 or v7 IDs
);

UUID vs. Auto-Increment

	AUTO_INCREMENT	UUID (BINARY/native)
Distributed safe	No — requires coordination	Yes
Leaks row count	Yes	No
Storage	4–8 bytes	16 bytes
Insert performance	Fast (always appends)	Fast with v7, slow with v4
Human readable	Yes	No (binary)
Merge-safe	No	Yes

For most single-database applications, AUTO_INCREMENT is fine. Reach for UUIDs when you need globally unique IDs across systems, when merging data from multiple sources, or when you don’t want to expose sequential row counts.

Frequently Asked Questions

Can I use UUID_V7 as a DEFAULT?

Not currently. Non-deterministic functions can’t be used in DEFAULT expressions in MySQL. Generate the UUID in your INSERT statement.

Can I backfill existing rows with UUIDs?

Yes. Add a UUID column, populate it with UUID_V7() in an UPDATE, then make it the primary key.

Is UUID_V7 a standard?

Yes — defined in RFC 9562, which supersedes RFC 4122.

Troubleshooting

Problem	Solution
FUNCTION UUID_V7 does not exist	Run INSTALL EXTENSION vsql_uuid
DEFAULT (UUID_V7()) rejected	Remove DEFAULT; call UUID_V7() in each INSERT
Extension 'vsql_uuid' is already installed	Already loaded — no action needed

See also

• Choosing a Primary Key Strategy in MySQL — comparing UUIDs against AUTO_INCREMENT and other strategies
• Generating Random Data in MySQL — the randomness behind UUID v4
• How InnoDB Stores Data — why random UUID primary keys cause page fragmentation