Should MariaDB be a provider or stay a flavor

[lefred]

Open question:

Currently, MariaDB is a flavor. But with the new concept of providers, should MariaDB be considered as such?

[renecannao]

MariaDB should stay a flavor. Here is how I would draw the lines in dbdeployer's architecture:

Provider = fundamentally different engine, different binaries, different wire protocol or runtime model. Today: mysql and postgresql. The ProxySQL integration already uses this concept - backendProvider is either mysql or postgresql.

Flavor = variant within a provider. Same process model, same lifecycle, mostly compatible binaries. mysql, percona, mariadb, pxc, ndb, tidb are all flavors under the MySQL provider. The differences are binary names (mariadbd-safe vs mysqld_safe), capability flags, and package layout - but they all run the same way: unpack, init, start a single mysqld process.

Topology = deployment pattern. single, master-slave, group, galera, pxc, ndb, innodb-cluster. These are orthogonal to flavor - you can (in principle) deploy replication with mysql, percona, or mariadb. The topology determines node arrangement and configuration, not the engine.

MariaDB ticks all the flavor boxes: single-process mysqld, same sandbox lifecycle (start/stop/kill), MySQL-compatible protocol. The only template-level change needed is binary name resolution, which is exactly what a flavor handles.

The one edge case is TiDB - it has a fundamentally different process model (tidb + tikv + pd), so you could argue it is a separate provider. But since it speaks the MySQL protocol and dbdeployer already treats it as a flavor, that is a pragmatic boundary worth keeping unless there is a reason to split it out.