Outage on Editors
major- Started
- Feb 16, 12:32 PM EST
- Resolved
- Feb 16, 12:32 PM EST
- Duration
- —
SuperAnnotate's updates (2)
- resolved · Feb 16, 12:32 PM EST
An outage affecting the platform's editors occurred from 17:32 to 17:45 UTC. The issue has been identified and resolved, and all services are now operating normally.
- postmortem · Mar 13, 1:27 AM EDT
### **Summary** On February 16 2026, a CloudFront configuration update introduced request-processing logic at the viewer-request stage for the production distribution. During deployment, a timing overlap occurred between the activation of the viewer-request function and the completion of CloudFront distribution propagation and cache invalidation. Because CloudFront updates propagate asynchronously across edge locations, there was a short period where the new request-processing logic was active while some edge locations were still serving previously cached content. During this window, certain requests received HTTP 301 redirect responses from CloudFront instead of being served normally. The behavior resolved automatically once CloudFront propagation and cache invalidation completed across the edge network. ### **Impact** The production editor briefly returned unexpected redirects for some requests. Scope: A single CloudFront distribution serving the production editor environment. Observed behavior: Several hundred requests were processed during the incident window.Many of those requests returned HTTP 301 redirect responses. Application navigation and static asset loading were temporarily affected, which resulted in redirected or interrupted user sessions. ### **Timeline** * February 16, 17:32 UTC – Unexpected redirect responses observed shortly after the CloudFront configuration update. * February 16, 17:45 UTC – CloudFront distribution propagation and cache invalidation completed. Request handling returned to normal. ### **Root Cause** Expected behavior: The viewer-request function was designed to normalize certain incoming request attributes and apply redirects only when specific routing conditions were met. Requests that did not match those conditions were expected to pass through CloudFront normally so that cached content or static assets could be served as usual. Actual behavior: During the production rollout, the viewer-request function became active while CloudFront was still completing distribution propagation and cache invalidation across edge locations. This created a brief period where the updated request-processing logic was running while some edge locations were still serving previously cached content. During this transition window, the function evaluated requests against this temporary state and returned HTTP 301 responses for certain requests. Once CloudFront propagation and cache invalidation completed and all edge locations reached a consistent state, request handling returned to normal. ### **Resolution** No rollback or configuration change was required. Once CloudFront distribution propagation and cache invalidation completed across all edge locations, the distribution returned to a consistent state and redirect responses were no longer generated. ### **Pipeline Improvements Implemented** Following the incident, adjustments were made to the deployment process to prevent similar timing conditions during CloudFront updates. The deployment pipeline now includes additional verification steps to confirm that CloudFront configuration changes have fully propagated and that cache invalidation operations have completed across the distribution before deployments are considered finished. Additional automated validation checks were also introduced to verify expected request behavior through the production distribution endpoint once the deployment completes. These improvements ensure that new request-processing logic is activated only after the distribution has reached a consistent state across all edge locations, reducing the risk of temporary inconsistencies during CloudFront propagation.
Updates are SuperAnnotate's own words from its status page (status: postmortem); we never add an inferred cause. The vendor's incident page ↗