Overview
Runs one mariadbd whose whole data directory is on a replicated volume. The block is a singleton that runs where the volume's primary is, behind a stable VIP. DRBD replicates synchronously, so MariaDB's own replication and Galera are not used. InnoDB syncs its redo log on every commit and keeps the doublewrite buffer on, so every commit a client sees acknowledged survives the crash that a failover is.
Compare db/postgres, which replicates with PostgreSQL's own streaming replication across one volume per replica.
- Shape
- Singleton on a replicated volume
- Reached at
- VIP
- Node package
pkgs.mariadb
Fetched on demand: the first time a block of this type is placed on a node, its agent fetches pkgs.mariadb, built from the node's own nixpkgs so the NixOS binary cache serves it. That first placement needs network access (or a binary cache on the LAN) and waits for the download; later placements on that node start at once.
Example
apiVersion: expanse.io/v1 kind: Block metadata: name: appdb namespace: default spec: type: db/mariadb replicas: 1 strategy: kind: SINGLETON resources: requests: cpu: 500m memory: 1Gi storage: - name: appdb-data size: 50Gi replication: 3 mountPath: /var/lib/mariadb config: port: 13306 # never 3306: the VIP holder binds that rootPassword: change-me-root database: app user: app password: change-me-app bufferPool: 512M network: ports: - name: mysql port: 3306 target_port: 13306 protocol: tcp expose: EXPOSE_VIP health_check: readiness: type: PROBE_TCP port: 13306 period_seconds: 2
$ expanse ctl block apply -f mariadb.yaml --dry-run # validate only $ expanse ctl block apply -f mariadb.yaml
Configuration
Keys go under spec.config and are checked against the type's JSON Schema when the block is applied. A default is what the block uses when the key is unset.
| Key | Default | Meaning |
|---|---|---|
rootPassword | required | Password for root@localhost and root@'%', at least 8 characters; applied on every start. Sealed: kept out of the stored block (block secrets). |
database | none | Database created on start if it is missing. |
user | none | Application account with all privileges on database; needs password and database. |
password | none | Password for user; applied on every start. Sealed: kept out of the stored block (block secrets). |
bufferPool | 128M | innodb_buffer_pool_size; keep it inside the memory request. |
maxConnections | 151 | max_connections. |
port | 13306 | mariadbd's listen port, the target_port of 3306. |
Using it
$ mariadb -h <VIP> -P 3306 -u app -p app
A Go DSN: app:<password>@tcp(<VIP>:3306)/app. Changing rootPassword or password and redeploying rotates them. The block deletes every other root account and every anonymous account on each start.
Connections are not encrypted and root@'%' can log in from anywhere that reaches the VIP, so keep the VIP on a network you trust and choose a strong rootPassword.
When a node fails
When the serving node dies, the block, its volume's primary and its VIP move together to a survivor, and InnoDB runs crash recovery on the same data. Open connections break and in-flight transactions roll back; a transaction whose COMMIT returned is on the volume. When a connection drops before COMMIT returns, the transaction may or may not have committed, so retries must be idempotent. In Expanse's VM test the VIP was on a survivor 53 s after the serving node was killed.
Limits
- No TLS, and one application account and database.
- No read replicas: one block is one
mariadbd, replicated by DRBD. - Backups: a crash-consistent volume snapshot is a valid restore point (see
util/restic-backup); for logical dumps runmariadb-dumpagainst the VIP.
Source
- docs/MARIADB.md: the Expanse guide this page follows.
- nix/blocks/db/mariadb: the type's identity and JSON Schema.