Creating a database
An engine, a version, a login. Four things are fixed the moment it exists, and each for a reason.
cubeship db create pg --engine postgres
cubeship db create pg --engine postgres --version 16 --username shop --database shop
cubeship db create cache --engine redis
cubeship db engines # what this release can runOr New database on the Databases screen, which offers the same fields and generates the password for you.
Provisioning runs detached from the request: the row appears as
provisioning, the image is pulled, the container started, and the
status follows. A database that failed to come up is failed with the
reason on it, and Start retries it.
The fields
| Field | Default | Notes |
|---|---|---|
name | — | a slug, unique on the instance; it becomes cubeship-db-<name> |
engine | — | postgres, mysql, mariadb, redis, mongodb |
version | the newest offered | see Engines for the list |
username | cubeship | Redis has one login, default, and refuses another; MySQL and MariaDB refuse root |
password | generated | 24 characters when left empty, so a weak one is never the default |
database | the name | the database created on first start; Redis has none |
The default login is cubeship rather than the engine's own:
postgres and root are the two names every scanner on the internet
tries first.
Fixed after creation
- The name. It is the container's, which every attached app resolves.
- The engine and the version. A data directory written by one major version is not readable by another. A database that "changed version" would be a container that will not start, with the only copy of the data inside a directory it will not read. Running a new version is a second database and the engine's own migration tools — or a backup restored into it.
The password can be rotated while the database is running. Cubeship changes the engine login and then stores the new value.
You can also change the description and ceiling.
Where the data lives
A host directory under the data directory, <data dir>/datastores/<id>,
mounted into the container. Anything in a container's writable layer is
destroyed when its configuration changes — publishing a port is such a
change — so the data is outside it, and survives.
Through an agent
create_datastore, list_datastore_engines and the rest are in the
tool reference. No tool reads or sets a
password, and none exposes a database: an agent can provision one and
attach an app without ever holding the credential.