Manage one database per git branch

[angererc]

Hi,

We would like to find a friction-less way to manage a separate database per git branch.

Eg imagine the master branch, accompanied by a master DB and sqitch config and plan to set up the master DB. Now I start working on a feature X and start a new GIT feature branch X. I don't want to get in the way of my fellow developers, therefore I also create a new DB X. When I'm done implementing X, I want to merge back into master and delete the X DB.

Now how do I best reconfigure sqitch in the X branch to use the new database X? If I add a target, then merging the changes back into master will have this left-over X target of the (now defunct) X branch, right? If I change the original target to point to the X DB, then merging into master will override the master target. I somehow want to temporarily, during the lifetime of a branch, override a target's database URI or (alternatively) create a temporary target that doesn't get merged back. (I hope this makes semse:) Any thoughts?

[theory]

Most engines support environment variables and defaults that specify database connection information. Postgres, for example, will fall back on your local username as the database name and username. So if you set a target like this:

sqitch add target local-dev db:pg:localhost/

Then for a user named bob it would connect to db:pg:localhost/bob and for a user named sue it would connect to db:pg:localhost/sue. All you have to do is remember to use it each time

sqitch deploy local-dev

Or set $SQITCH_TARGET to that target:

export SQITCH_TARGET=local-dev
sqitch deploy # uses local-dev

Of course, if you're setting environment variables, you can just do that generally, esp if youre engine doesn't natively support the flexibility you require. For example, you could use a tool like direnv and commit an .envrc file to the root of your project with something like this:

export SQITCH_TARGET=db:pg://localhost/myproj_$(git branch --show-current)

Then every time you check out a new branch, as long as you've told direnv to trust the file, it will automatically reset the environment variable with the branch name as part of the database name.

HTH!

[angererc]

Thank you for your answer. I think your direnv suggestion will be workable for us, we'll test this out. I still think it would be nice to have such a feature built into sqitch (git-branch-aware DB targets) so one can tie the dev database to the git feature branch you are working on. But I also understand that that's probably not easy to do in a VCS independent way, therefore the env variable workaround seems like a good solution.

[theory]

That's a fair request. I've long wanted better integration with VCS software, but aside from sqitch checkout for Git only, I've not had the chance and there hasn't been much demand TBH. I could perhaps see some additional branch awareness that, say, acted as an environment variable for the branch name to use in target or some such, but it would require a fair bit of thinking through, since branches can be local or remote or in different remotes, but even some sort of simply support for environment variable interpolation in Config::GitLike would go a long way, especially for CI/CD pipelines that tend to set an environment variable to a branch name.