Probe the bridge's real health, not a port that is always open

FixKlusterServices
Shipped
29 Agosti 2026, 02:57 UTC
Author
Kamo
Commit
9112205

kamo-meet let one person into a meeting and then ended it the moment a second joined. The bridge had been hard-unhealthy for 41 hours: an in-place container restart on 2026-08-27 came up with "Failed to bind single-port", and ice4j builds that harvester ONCE at startup and never retries, so the process can never recover on its own. One participant hid it completely -- jicofo allocates no bridge for a lone occupant. The second join triggered **************** hit "no operational bridges", and jicofo tore down BOTH participants and stopped the conference. That is the whole reported symptom. Nothing restarted the pod because the probes were a bare TCP connect to 9090. Jetty binds 9090 regardless of ICE state, so the probe reported healthy for the entire 41 hours while every two-person meeting died. The probe could not observe the only failure that matters. It now reads /about/health, which returns 500 for exactly this state, so the restart -- the only recovery available -- actually happens. It has to exec curl against 127.0.0.1: the pod is hostNetwork and the node DNATs <hostIP>:8080 to Traefik, so an httpGet probe gets Traefik's 404 rather than the bridge's health. Verified both ways before settling on exec. JVB_OHTTP_SERVER_PORT="-1" went with it. It claimed to close 8080 over a CockroachDB conflict, but CRDB is retired and the image never honoured the variable -- 8080 listened the whole time -- so it only misled.

All changes

Je, unaona nini kuhusu usafiri?

Kila moja ya hizi updates ardhi katika nafasi yako ya kazi moja kwa moja. Kuanza bure na kuangalia kukua wiki baada ya wiki.

Kuwa Huru MileleMtazamo wa bei