On July 31, 2026, the Cartesia API returned elevated 503 errors across API traffic. Customers may have experienced failed API requests and reconnects during the incident.
Our automated monitoring detected the issue within 4 minutes, and we published an incident to our public status page within 11 minutes. We mitigated the issue within 13 minutes by shifting traffic to a backup content delivery path, and fully restored normal routing within approximately 1 hour.
All times are Pacific Time (PT) on July 31, 2026.
| Time | Event |
|---|---|
| 6:29 PM | A change to the routing code for our primary CDN was deployed to production. This change caused elevated errors in production due to use of unsupported JS features in the runtime our CDN provides. |
| 6:32 PM | Our automated monitoring registered elevated error rates on our primary CDN, and our internal API health checks began failing shortly after, alerting our on-call team. |
| 6:33 PM | The elevated errors triggered a retry storm from some clients, which resulted in our CDN throttling all of our traffic. |
| 6:40 PM | We published an incident to our public status page. |
| 6:42 PM | We shifted traffic to a backup CDN, which mitigated the customer-facing impact. |
| 7:14 PM | We reverted the routing change and began gradually moving traffic back to the primary CDN. |
| 7:29 PM | All traffic was restored to the primary CDN. |
| 7:34 PM | We confirmed normal operation and marked the incident as resolved on our status page. |
A recent routing update used unsupported JavaScript syntax in our primary content delivery path. This change passed testing but failed in production due to a mismatch between the JS runtime we use for testing and the JS runtime our CDN runs our routing logic on. The resulting client reconnects increased request volume enough to trigger throttling, which contributed to elevated 503 responses for other customers.
We are taking the following steps to reduce the likelihood and impact of similar incidents: