Diagnose a 429 spike
Introduction
When callers see 429 from a pool, Metrics plus Logs usually answer why in under
two minutes. You do not need to export NDJSON first.
Steps
- Open the pool's Metrics tab and set the range to the incident window (clamped to your plan retention).
- Check Outcomes. A jump in
overloadedorqueue_timeoutmeans the concurrency limit or queue budget filled. A jump inrejected_authorrejected_ipis not capacity. - Open Top keys (overload) and click a noisy key to jump into Logs with that filter.
- Confirm Live queue depth and effective limit on Metrics. If depth is high while the limit looks right, raise queue timeout or concurrency only after you know which callers are burning slots.
- If you only care about allowance spend, open Billing instead. Metrics is admit/queue/reject proof; Billing is request allowance.