On August 17, GitHub was down for nearly eight hours.
Issues, Pull Requests, Actions, the API, and Copilot all failed in succession.
The next day, GitHub posted its root cause analysis (RCA) on its official status page.
Seven hours and 47 minutes: what went down
Cassidy Williams, senior director of Developer Experience at GitHub, announced on X at 7:31 p.m. KST on August 18, 2026 that "the RCA for yesterday's outage is out."
According to the official GitHub status page incident report, the outage ran from 13:28 to 21:15 UTC on August 17, about 7 hours and 47 minutes.
At the peak, web and API error rates reached about 20%, and archive and raw file download error rates reached about 50%.
Authentication systems based on SAML and OIDC were affected as well.
It started with one sidecar failing to scale
The official RCA pointed to network saturation in a US Central data center.
The trigger was that Istio sidecar pods used in the service mesh hit their concurrency limit, but a misconfigured autoscaling policy kept them from scaling properly.
That bottleneck exhausted load balancer capacity, and "optimistic retry" logic added more load to the internal load balancers, so the failure cascaded.
GitHub also named VS Code's retry behavior, which amplified Copilot traffic, as one of the causes.
Prevention steps and practical takeaways
In the RCA, GitHub listed these steps: fixing the autoscaling policy for service mesh sidecars, auditing all Istio request and concurrency limits, reviewing retry behavior across gateways and clients, improving VS Code's Copilot retry logic, and strengthening load balancer monitoring and failover safeguards.
If your team runs a sidecar-proxy service mesh such as Istio, this is a good case for checking whether your autoscaling limits actually hold up during real traffic spikes.
It's also a reminder, when designing retry policies, that client-side "optimistic retries" can make load worse during an outage.
The 13th incident this month: when will the cause be explained?
This was the 13th separate incident GitHub has had in August 2026 (spread across nine days).
When developers asked whether there's a deeper architecture or distributed-systems problem, Williams only replied that "we publish monthly availability reports covering big and small issues and our response." There's still no separate explanation of a structural cause for the repeated outages.
The monthly availability report covering all of August is due in September.