GitHub Outage Report: Network Saturation Caused 8-Hour Service Disruption
RealGeneKim · x · 2026-08-19
GitHub released a root cause analysis for the August 17 outage. The incident lasted approximately 7 hours and 47 minutes, impacting Issues, PRs, APIs, Actions, and Copilot. The primary cause was network saturation on load balancers in the Central US datacenter due to a traffic spike, compounded by an Istio sidecar pod failing to auto-scale correctly. A latent retry bug in VS Code, triggered by delayed replies, amplified traffic to the Copilot Token Service by roughly 10x. Most services recovered on the same day.
More from Infra
- Brex data: over half of the fastest-growing startups sell AI infra, not AI products — FinanceYF5 · 2026-08-19
- Neon: over 80% of new databases are created by AI agents; Supabase tops 60% — FinanceYF5 · 2026-08-19
- Over 80% of new databases are created by AI agents, Neon says — FinanceYF5 · 2026-08-19
- Startups split workloads: closed models for reasoning, open models 5-20x cheaper for bulk tasks — FinanceYF5 · 2026-08-19
- Time from closed API to open-model compute shrank from 23 months to 5 — FinanceYF5 · 2026-08-19
- The open-model stack: every layer from GPU clouds to hosted fine-tuning sells 'less to worry about' — FinanceYF5 · 2026-08-19