We use an init container in Kubernetes for each pod that requires a Camunda schema.
That init container contains the liquibase runner and a bunch of the SQL files that Camunda requires, plus of course, the liquibase descriptor that ensures the correct SQL files are run in the right order.
In the meantime, Camunda has started publishing liquibase descriptors for any Camunda versions > 7.16 IIRC. So you could also base off of that. Back when we started, we basically handcrafted the same files.
In order to share the setup between components, we made sure that our shared library contained all of the descriptors and SQL files, and that upgrading to a newer version of the shared library also gave you the correct matching setup for liquibase. That part was a combination of Maven dependency management and copying resources into the docker working directory when building the init container. This is obviously only needed for re-use and quite custom to your setup with Maven and how you use Docker. The general idea would be the same with slightly different setup: build a container that matches the version of Camunda in the main container and package all the scripts into it and run them in order with liquibase before you start the main container.