Document Understanding Outage on GXP East US Region

Update timeline

1. investigating Aug 10, 2026, 02:11 PM UTC

   We are investigating degraded performance impacting Front-End Services in Document Understanding across GXP East US Region. Impact: Users may notice timeouts when accessing the UI in GXP US. Next update: Additional updates will be provided as more information becomes available.

2. resolved Aug 10, 2026, 02:25 PM UTC

   The issue has been resolved and the performance of the Front-End Service in Document Understanding has returned to expected levels after degraded performance impacted the UI in the GXP East US Region. Impact: No ongoing user impact.

3. postmortem Aug 19, 2026, 06:14 AM UTC

   ## Customer impact Between August 10, 2026 at 13:21 UTC and August 10, 2026 at 14:03 UTC, a subset of customers experienced degraded performance and timeouts when accessing the Document Understanding user interface, which provides the design-time experience, in the Delayed US region. Document processing automations were not affected. ## Root cause During a manual deployment from a pre-existing build of the service behind the Document Understanding user interface to the Delayed US region, our deployment process reassigned the version identifier for the service being deployed. The Document Understanding interface requires that its supporting resources are available under the same version identifier as the deployed service. Because the deployment process changed this identifier, the interface could not locate the required resources, causing it to become inaccessible or time out. ## Detection We became aware of the issue moments after the deployment was finished, through manual verification as part of the manual deployment checklist. An automated alert shortly followed, triggered at 13:28 UTC on August 10, 2026. ## Response After identifying the reason for the failure of the manual deployment, our engineering team started to deploy a corrected update with the required resources available. At the same time, the operations team was engaged to perform a manual rollback of the deployment. The rollback was completed at 14:03 UTC, restoring access to the design-time experience. At 14:11 UTC, we published a public status update indicating degraded performance and possible timeouts in the Document Understanding user interface for the Delayed US region. By 14:15 UTC, the corrected update had completed and the interface was confirmed as deployed with a correct new version and working. At 14:25 UTC, the incident was marked resolved and the public status page was updated to confirm that Document Understanding interface performance had returned to expected levels. ## Follow up 1. We are reducing the time required to restore a previous version of the service, so that recovery from a failed deployment is faster. 2. We are making improvements to the manual deployment process to prevent a future situation, by validating the required resources' existence as a prerequisite.