bmersereau tightens the screws on Mike's file storage
A small, pointed backend cleanup that closes a cluster of reliability complaints around how Mike handles uploaded documents.
Mike stores its documents in Cloudflare R2, an object-storage service similar to Amazon S3. bmersereau noticed the backend was spinning up a fresh storage client for every single file operation - wasteful, and it meant the underlying network connections couldn't be reused. The fix consolidates this into a single shared client that's created once and held onto.
The more interesting half of the change is observability. Until now, when R2 failed - say, during an outage - the server would quietly return nothing, with no trace in the logs. The team would only learn about it from users. bmersereau's PR makes those failures visible to whoever is running the system. The pull request is pitched at upstream and claims to resolve five separate tracked issues about storage reliability, which suggests this is the consolidation point for a chorus of complaints.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?