GitHub’s October review describes a database-routing machine that retained records for retired servers. Elected leader on October 7, it published bad write destinations that validation accepted. Correcting leadership restored writes, but maintenance ignored a pause and triggered another election; the unsafe machine could return. Removing it ended the recurrence. Planned fixes span retirement cleanup, onboarding, route validation and enforcement of the pause.

Other incidents isolate scarce-resource problems. Enterprise Importer’s oversized queries were replaced with smaller fixed-size batches. A slow lookup was moved before a write so it no longer held an unused connection. Synchronized reconnections saturated network or server capacity; admission limits, jitter and available headroom mattered separately. On October 5, moving runner-assignment work to a healthy region overloaded that region too. Failover needed capacity for both transferred work and the recovery surge.

github october