GitHub publishes the root cause for its 7 hour 47 minute outage: an Istio sidecar scaling gap, HAProxy flow exhaustion, and a VS Code retry storm

Sources: GitHub Status incidents API, a rolling feed, GitHub Status incident page, the durable citation
Summary: GitHub published the root cause analysis for the 2026-08-17 outage, which ran 13:28 to 21:15 UTC with web and API error rates near 20 percent and archive and raw content near 50 percent. GitHub attributes the trigger to network saturation on Central US load balancers, originating in an Istio sidecar pod that hit its concurrency limit and failed to autoscale because the scaling policy watched host service limits and not sidecar limits, which cascaded until four HAProxy nodes exhausted their flow limits and degraded the gateway authentication path, and pausing HAProxy on those nodes produced immediate broad recovery. A latent retry bug in VS Code amplified Copilot token traffic roughly tenfold, taking the Copilot Token Service from 7 to 9 thousand requests per second to 70 to 100 thousand.
Why it matters: The length of the outage came from client and gateway retry amplification rather than from the original fault, which is the transferable lesson for anyone running retrying clients behind a shared gateway.
Follow-up: This closes the analysis GitHub promised on 2026-08-17. Track the further root cause analysis GitHub promised for the 2026-08-18 runner-group incident.