GitHub's administration published a detailed incident report on the outage that paralyzed the developer platform for nearly eight hours. Service access issues began on August 17, 2026, at 16:28 MSK, and were fully resolved only at 00:15 MSK on August 18. Thus, the total outage duration was 7 hours 47 minutes, during which users reported a rise in errors across various platform services — from the API to CI/CD pipelines and Copilot.

Root cause: Istio autoscaling failure

According to the report, the root cause was network overload on the load balancers at the company's central office in the US. The trigger was reaching the parallel execution limit of the Istio sidecar — a service mesh proxy component. Under normal operation, the system was supposed to automatically increase throughput as these limits were exhausted, but this did not happen due to a misconfigured autoscaling policy: only the host service's state was monitored, not that of the load balancer itself. As a result, throughput was not scaled up, and requests began to accumulate.

Tenfold traffic due to a VS Code bug

The situation was sharply worsened by client behavior: as the service degraded, clients began sending massive numbers of retry requests, which ultimately took the load balancer down. According to engineers, the incident helped uncover a long-standing bug in the Visual Studio Code editor — programs began generating retry requests, causing traffic to the Copilot Token Service to grow tenfold. This led to a significant delay in restoring this particular component.

How engineers stopped the avalanche

To stabilize the system, specialists temporarily reduced the number of gateway retry attempts by changing the code and configured the load balancers to reject incoming requests to the Copilot Token Service with a 403 error code. The phased recovery looked like this: the GitHub Actions service was up by 18:03 MSK on August 17, most other services by 19:36 MSK the same day, and the Copilot Token Service was fully restored only at 0:02 MSK on August 18, after which the incident was officially closed at 00:15 MSK.

Contradictory data

There are inconsistencies in the wording around the incident that are worth noting. In headlines and brief summaries, the outage duration is often rounded to 'eight hours,' whereas the report itself cites the exact figure of 7 hours 47 minutes. In addition, open sources for 2026 mention other GitHub outages (specifically, incidents in June 2026, as well as reports of access issues in certain regions), which are not part of the August 17–18 event under consideration. When reading the news, it is important not to conflate these separate episodes into a single continuous outage.

What was promised to be fixed

The company announced a series of corrective measures: fix the errors in the autoscaling policy, review the limits on the number of retry attempts, audit the Istio parallel access settings, and resolve the VS Code behavior issues that caused traffic to the Copilot Token Service to multiply. These steps are aimed at preventing a repeat of the cascading outage when service mesh limits are reached.