GitHub reports three service disruptions in May
GitHub recorded three incidents in May that affected service performance, spanning its Issues, Actions, and Microsoft Teams integration products. The company detailed the failures, root causes, and remediation steps in its monthly availability report.
Issues attachment upload failures
On May 1, from 22:09 UTC to 23:13 UTC, users could not upload attachments to the Issues service. A new feature had introduced a custom header on all client-side HTTP requests, which triggered CORS errors against the upload provider. Roughly 130,000 users were affected for about 45 minutes.
Engineers rolled back the feature flag at 22:56 UTC to restore service. To avoid a repeat, GitHub is adding metrics to monitor client-side request changes during rollout. An updated version of the feature, informed by the incident, is now running in production.
Delayed job starts for Ubuntu-24 runners
From approximately 09:45 UTC to 14:45 UTC on May 28, GitHub Actions users saw delayed job starts for workflows in public repositories using Ubuntu-24 standard hosted runners. The cause was a misconfiguration in backend caching after a failover, which duplicated job assignments and reduced capacity in the affected runner pools. An estimated 19.7% of Ubuntu-24 hosted runner jobs in public repos were delayed. Other runner types and private repository workflows were unaffected.
The backend cache was fixed by 12:45 UTC, and the pools were scaled up to clear the job queue by 14:45 UTC. GitHub says it is strengthening failover resiliency and validation to reduce the odds of similar issues.
Microsoft Teams integration outage
Between 08:10 UTC and 16:00 UTC on May 30, the Microsoft Teams GitHub integration experienced a complete outage. The service returned a 100% error rate on all functionality except link previews and could not process requests or deliver notifications.
The failure was traced to an authentication issue with GitHub's downstream provider. Although monitoring was active, alerting thresholds were not sensitive enough to trigger a timely response, delaying detection. Once engaged, the team worked with the provider to fix the authentication problem, but slow provider response times extended the incident to nearly eight hours.
Service was restored with the provider's help. GitHub is now working to adopt more durable authentication methods to lower the risk of recurrence.
