We migrated the transcribe service to GKE, and to save costs, we used a shared read-only disk to store model files. However, the language detection model was outdated, and when the code attempted to update it, the update failed due to the read-only disk. This resulted in some speech having incorrect language detection.
Resolved
Major outage
Started 22 days agoLasted about 1 hour
Affected
OpusClip
Operational from 12:31 AM to 1:19 AM
Updates
- ResolvedResolved
- InvestigatingInvestigating