MariaDB Replication Bug Reproduction

This repo demonstrates how a MariaDB migration that adds a new UUID primary key can silently corrupt replication and eventually trigger:

ERROR 1032: Can't find record in '<table>'

This happens because MariaDB executes DDL as SQL on both master and replica, causing uuid() defaults to generate different PKs on each server.

Setup

Start master + replica:

docker compose up -d

Check replication:

docker exec -it replica mariadb -uroot -proot -e "SHOW SLAVE STATUS\G"

Should show:

Slave_IO_Running: Yes
Slave_SQL_Running: Yes

Reproduce the Issue

Create demo db on the replica

docker exec -it replica mariadb -uroot -proot -e "CREATE DATABASE IF NOT EXISTS demo;"

1. Run initial migration + seed

DB_HOST=127.0.0.1 DB_PORT=3306 npm run initial:migrate
DB_HOST=127.0.0.1 DB_PORT=3306 npm run seed

2. Apply the problematic migration

DB_HOST=127.0.0.1 DB_PORT=3306 npm run bad:migrate

This migration:

• Drops the old PK
• Adds id with DEFAULT uuid()
• Makes id the new PK
• → master + replica now have different UUIDs

3. Trigger replication failure

DB_HOST=127.0.0.1 DB_PORT=3306 npm run trigger-failure

Verify the Break

docker exec -it replica mariadb -uroot -proot -e "SHOW SLAVE STATUS\G"

Expected:

Last_SQL_Errno: 1032
Last_SQL_Error: ... Can't find record ...
Slave_SQL_Running: No