Skip to content- Sources: primary
- Summary: A background data-cleanup job throttled on replica lag but not primary load, so the primary hit its connection limit, and missing timeouts plus a token-creation retry loop delayed recovery. The report also says GitHub paused Central US Azure ramps on 2026-09-11 over uneven zone capacity and missed its 2026-09-30 target to deploy every production service there, with 120 of 160 services moved by 2026-10-07.
- Why it matters: The failure pattern of lag-only throttling, no fast timeouts, and retries on failing writes is a checklist item for anyone running batch jobs against a shared primary.