Nearly all API services were interrupted for 12 minutes, between 21:58 and 22:10 UTC.
The outage was causing by an attempt to test OVHcloud (our hosting provider)'s Load Balancer product to improve Zyno's reliability in the event of a single server outage. Upon deploying a load balancer instance in our public cloud account, OVHcloud's control plane automatically deployed a "gateway" instance with the same IP address as our dedicated routing infrastructure. As a result, outbound traffic from our services back to users was directed to the wrong gateway and subsequently dropped. Normal traffic resumed as soon as we deleted the IP address from the conflicting gateway.