On September 4, 2026, between 18:26 and 18:47 UTC, a subset of requests to the E2B US production environment failed. We were gradually migrating production ingress traffic to a new cluster, reaching 15% of traffic, and the new ingress did not have routes configured for all of our production domains. Requests on those domains that were routed to the new ingress were rejected with HTTP 404 before reaching the sandbox API or running sandboxes. Up to 15% of requests were affected, and running sandboxes were not stopped or restarted. We detected the issue at 18:43 UTC and rolled traffic back to the existing ingress at 18:46 UTC, after which errors stopped, with a small number of residual failures reported until about 18:50 UTC. We have since added routing for all production domains to the new ingress, added monitoring and alerting for requests rejected at the ingress, and changed the ingress to return HTTP 503 instead of a generic 404 for unrouted hostnames so that routing failures are immediately distinguishable from missing resources.