In the case that a new change is so dramatic that it is hard to anticipate all of the potential side-effects, here is a protocol for rolling out these sensitive changes in a way that:
- Minimizes adverse impact on end users.
- Maximizes our responsiveness to these changes.
- Prepare a normal release.
- Prepare a rollback release.
- Roll the normal release out.
- In case of emergency, roll back.
Simply follow the steps in the publishing guide.
Follow the steps in the publishing guide with a different context:
Instead of creating a version branch off of the main branch, create a version branch off of the latest release. It is customary that this release increments the patch version number.
Ensure the rollback release has been built, and downloaded locally, fully ready to deploy with immediacy.
For a sensitive release, initially roll out to only 1% of Chrome users (since Chrome allows incremental rollout).
Monitor Sentry for any recognizable error logs.
Gradually increase the rollout percentage.
If a problem is detected, publish the roll-back release to 100% of users, identify the issue, fix it, and repeat this process with a new release.
This protocol is a worst-case scenario, just a way to be incredibly careful about our most sensitive possible changes.