Throwing out ideas, would a MariaDB started in a docker container be better? There is a Python SDK for docker to programmatically start containers

Next step: Experimenting by Robert. Larger size of downloadable is a drawback, promoting the latest MariaDB 10.5 is an advantage. What are other considerations?

As far as I understand, using the SDK would require users to have docker installed or install it and go through a image building process when the kernel starts. This might eat a chunk of the easiness of use we want to promote for the project and a part of the users might already have an existing installation of mariadb and they'll have to go through the docker steps anyway. I don't know if this is the right move, can cvicentiu correct me if I'm wrong or maybe state some advantages for going with docker on this?

Complexity seems to be a disadvantage!

Let's demote this to nice-to-have. We don't want to delay the release because of this. (We can still consider doing it later, because of the advantage of promoting MariaDB 10.5).

Because it was getting annoying to test things as I kept forgetting to start the server, I implemented this feature to make my development process more comfortable. It is currently implemented with Python subprocess framework, but it is isolated in the architecture of the kernel so it can easily be swapped later with some other solution that is more efficient, covers more platforms, etc.

Need some input before closing this issue. What does kaj mean when he says "Notification"?
At the moment the kernel just issues a message in the logs (which is quite visible because in the use cases when this feature is needed, the user is running jupyter from a terminal window). Does this sound ok to you?