ListenBrainz release process changes

Over the past year, I’ve made a major change in how we think about releases for ListenBrainz.

Before

There were no real expectations around deployment after pull requests got merged. There were two branches – master and production. The HEAD of production is what the site was running at any moment. People made pull requests based off master, stuff got merged into master and eventually got released.

Each release required merging master into production. However, there may have been hotfixes to production in the meanwhile as well. So what we ended up doing to bring master and production into sync was:

  1. Merge master into production.
  2. Deploy the production branch.
  3. Merge production back into master.

The time between releases could be months! The commit difference between master and production was often in the order of hundreds. This made releases painful and we had to be extra careful. If there was a schema change involved, that made things even more complicated.

After

The change that we made is merging stuff into master as often as possible. And as soon as it’s merged, release the changes into production. This set up a decent pipeline where each release consisted of a small number of pull requests, making it much easier to release stuff.

Over the last 3 months, we’ve done 33 releases (almost 3 deploys a week), which account for ~400 commits.

There have been a number of advantages to this approach. The deployer doesn’t feel as concerned releasing 1 small pull request vs releasing 100 huge pull requests merged into master a few months ago. It also sets up a quick feedback loop for the developer who gets to see their changes working in real life as soon as their pull request gets merged.

Given our engineering capacity, this new method of releasing stuff often has really helped in making ListenBrainz a more active project.

Links to this note