engineering1 min de lecture
Friday Five #4: The cleanup job that never ran
Five short engineering notes from the week: a cleanup job that never ran, a streaming bug, a security count that only goes down, a slow push gate, and one question.
Rob Vasquez·
Five notes from the week. Short on purpose.
- Broke: My laptop kept slowing down. The file watcher had grown to 35 GB of memory, crawling 214 old worktrees inside one repo. I had already scheduled an hourly cleanup for exactly that. It had never worked once. The scheduler doesn't load my shell profile, so it couldn't find the tool, and it exited silently every hour.
- Found: A streaming endpoint I built this week ended a cut-off response with a normal "done" event, so a client would treat half an answer as complete. Client disconnects also left traces open. A second reviewer caught both. Neither shows up on the happy path.
- Changed: The security scanner now fails a push on any new finding, and a ratchet file lowers the allowed count after every fix. Moving the payment webhooks onto one shared receiver took that count from 27 to 16. The number can only go down.
- Using: My push gate takes about ten minutes, so another session can land first and my push loses the race. The rule now: fetch, rebase, rerun the gate suite, push again right away.
- Asking: When did you last check that a job you scheduled actually did something?