- Shipped
- August 5, 2026 at 1:46 AM UTC
- Author
- Kamo
- Commit
- ca6160c
The /colibri-ws/<jvb-id>/ location builds its upstream from a capture group. nginx resolves such names at request time and only when a resolver is configured, so every request died with "no resolver defined to resolve **************** and returned 502. The bridge channel silently fell back to SCTP, so it was never noticed. Move the start script out of a printf in the Dockerfile into nginx/start.sh and have it write conf.d/00-resolver.conf from the pod's own /etc/resolv.conf, so the cluster DNS address is not hardcoded. Verified end to end: the route now returns 405 through meet.kamocrm.com, identical to hitting the bridge's 9090 directly. Needs the jvb-k1m1 Service added in KlusterServices 1a1a7dd, and JVB_WS_DOMAIN so the advertised URL is not wss://localhost:8443/.