A run of deployments through late March and early April was spent on the least visible and most important thing an API does, which is stay up.
The database connection gained resilience and proper diagnostics, so a dropped connection reconnects and says what happened rather than leaving the process running and answering everything with an error. Until then, the failure mode was an API that was technically alive and useless, which is the hardest kind to notice from outside.
A configuration fault went with it: the bundled production build was not finding its environment file, because it resolved the path relative to the wrong directory. That is the class of bug that works perfectly in development and fails only once deployed, and it had been costing deployments.
The leaderboard was also corrected to include every contributor rather than only those with at least one note, which had been quietly excluding most of the people on it.
Everything in this release
- Improved
The database connection recovers from a drop and reports what happened.
- Fixed
The production build now finds its environment configuration.
It resolved the path relative to the wrong directory, so it worked in development and failed once deployed.
- Fixed
The leaderboard includes every contributor, not only those with a note.