US - Document Understanding - Partial Outage

Final update

## Customer impact Between August 21, 2026 at 11:05 am UTC and August 21, 2026 at 12:18 pm UTC, a subset of customers experienced failed Document Understanding operations, including document classification, extraction, and digitization. The estimated duration of the partial outage was 48 minutes. The impact was limited to customers using Document Understanding in the US region. ## Root cause The incident was caused by the Document Understanding storage database failover configuration entering a broken state during a preventive database scale-up operation. The scale-up was initiated after the database approached its storage limit. During the operation, the secondary database could not be scaled, the attempted removal from the failover configuration failed, and the database platform provider had to break the replication link. This left the failover configuration in a temporary unavailable state, causing the storage and runtime services that depend on that database to fail requests. ## Detection The incident was detected through an automated alert for the Document Understanding services, which was acknowledged on August 21, 2026 at 12:10 pm UTC. ## Response Before the customer-impacti

1. Postmortem · Aug 27, 03:46 UTC
   ## Customer impact Between August 21, 2026 at 11:05 am UTC and August 21, 2026 at 12:18 pm UTC, a subset of customers experienced failed Document Understanding operations, including document classification, extraction, and digitization. The estimated duration of the partial outage was 48 minutes. The impact was limited to customers using Document Understanding in the US region. ## Root cause The incident was caused by the Document Understanding storage database failover configuration entering a broken state during a preventive database scale-up operation. The scale-up was initiated after the database approached its storage limit. During the operation, the secondary database could not be scaled, the attempted removal from the failover configuration failed, and the database platform provider had to break the replication link. This left the failover configuration in a temporary unavailable state, causing the storage and runtime services that depend on that database to fail requests. ## Detection The incident was detected through an automated alert for the Document Understanding services, which was acknowledged on August 21, 2026 at 12:10 pm UTC. ## Response Before the customer-impacting incident was declared, the primary database scale-up was completed and support from the database platform provider was engaged for the secondary database issue. After the failover configuration became broken, we explored redirecting the service connection but could not identify a safe, immediate
2. Resolved · Aug 21, 13:36 UTC
   The issue impacting document classification and extraction for Document Understanding in the US region has been resolved. Following a period of monitoring, service is confirmed healthy and operating normally.
3. Monitoring · Aug 21, 12:37 UTC
   A fix has been implemented for the issue impacting document classification and extraction for Document Understanding in US, and we are currently monitoring the results.