GitHub Actions reliability falls below enterprise benchmark

GitHub Actions, a widely-used continuous integration and delivery service, has fallen below the 99.9% availability level typically associated with enterprise cloud services. In the 90 days leading up to August 19, Actions experienced 99.33% uptime, which translates to about 14.5 hours of downtime. This is significantly lower than the 99.99% availability of GitHub's core Git Operations service.

The difference underscores how reliability fluctuates across various components of the GitHub ecosystem. Despite the repository infrastructure remaining stable, Actions has encountered repeated disruptions that impact automated workflows used for building, testing, and deploying software. A major disruption on August 6 lasted over nine hours, during which 71% of workflow runs faced infrastructure failures, and 75% experienced delays exceeding five minutes.

This incident, caused by an internal service's capacity limitations, extended beyond GitHub-hosted runner users to include self-hosted runners. The root cause was a routine deployment that triggered a cascading failure across multiple clusters and downstream systems. Recovery efforts were complicated by a backlog and a job assignment flaw.

GitHub has since acknowledged the incident's deviation from its availability commitments and is accelerating its shift of the Actions architecture to Microsoft Azure. Prior to this, another outage on August 17 affected a wide range of GitHub services, including Actions, for seven hours and 47 minutes. This outage, primarily due to network saturation in GitHub's Central US data center, saw web and API error rates reach approximately 20%, and archive and raw-content downloads experience error rates of around 50%.

The incident originated when an Istio sidecar pod hit its concurrency limit and failed to scale correctly due to a policy monitoring the host service instead of the sidecar limits. The failure amplified as retry mechanisms were triggered, further straining GitHub's infrastructure. These issues highlight the risks of concentration in software delivery pipelines, where reliance on a single hosted CI/CD platform can lead to delayed releases, interrupted security patches, and failed automated deployments even when the customer's own infrastructure is healthy.