“After a bit of sleuthing, it ended up that Supabase was taking a database backup that took the database fully offline every night, at midnight.”
Val Town’s scheduled jobs died every night at midnight because its database host took the whole thing offline to back it up. It took a week to get Supabase to stop. Then auto-resize failed on a Sunday and left the database read-only at 95% full, with an admin UI that crashed along with it. The local dev CLI runs 11 Docker containers and never stayed working for more than a day. A managed database exists to make backups boring. This one made them an outage.