Course
llm-zoomcamp
Question
I'm logging user feedback (thumbs up/down) that arrives sometime after the original interaction is stored — does it matter which serverless database I pick for the bonus deployment points?
Answer
It can, more than you'd expect. If your write pattern is an INSERT when the interaction happens, then an UPDATE later when someone clicks feedback, check your database's real write-consistency window before committing to it — not every serverless option handles that second step the same way. BigQuery, for example, keeps a freshly inserted row un-updatable for up to ~90 minutes (its streaming buffer), which silently breaks a delayed feedback UPDATE unless you restructure feedback into a separate table to work around it. A serverless Postgres option like Neon supports both INSERT and a later UPDATE natively, no special casing needed, and it still scales to zero the same way. Worth checking this specifically if your monitoring/feedback flow has this two-step write pattern — "serverless" doesn't mean every option supports the same access patterns.
Checklist
Course
llm-zoomcamp
Question
I'm logging user feedback (thumbs up/down) that arrives sometime after the original interaction is stored — does it matter which serverless database I pick for the bonus deployment points?
Answer
It can, more than you'd expect. If your write pattern is an INSERT when the interaction happens, then an UPDATE later when someone clicks feedback, check your database's real write-consistency window before committing to it — not every serverless option handles that second step the same way. BigQuery, for example, keeps a freshly inserted row un-updatable for up to ~90 minutes (its streaming buffer), which silently breaks a delayed feedback UPDATE unless you restructure feedback into a separate table to work around it. A serverless Postgres option like Neon supports both INSERT and a later UPDATE natively, no special casing needed, and it still scales to zero the same way. Worth checking this specifically if your monitoring/feedback flow has this two-step write pattern — "serverless" doesn't mean every option supports the same access patterns.
Checklist