Two of our Public, CPU-basic (free) Spaces have been returning intermittent HTTP 500 / 503 responses from the HF edge proxy since ~2026-07-19 (UTC), even though both containers are healthy and RUNNING. This blocks our field operators.
Affected Spaces (same org, different stacks):
- korea-standards-association/barrier-free-dashboard (Streamlit, commit 3addae7c)
-
- korea-standards-association/bf-field-checklist (FastAPI/Uvicorn, commit 5694b596)
- Observations: Both report stage=RUNNING on the correct commits. Container logs are clean (single startup, no crash loop/traceback/OOM); the apps’ own health checks and page requests return 200 OK internally (checklist GET /healthz & GET / → 200; Uvicorn on 0.0.0.0:7860; Streamlit on
- Already tried (no effect): restart_space() on both; restart_space(factory_reboot=True) on the checklist (rebuilt cleanly to RUNNING in ~34s, but the edge 500 persists).
Sample failing request IDs (HTTP 500, 2026-07-19 ~14:0x UTC): dashboard /_stcore/health: IAJIMZ, vCYVIi, cd4hvE - checklist /healthz: SdUfco, 4XDGMt, j1o-yY, IkgDQc
Timeline: normal until ~2026-07-18 08:06 UTC; 500s noticed 2026-07-19.
Request: Please investigate the edge->container upstream routing (x-proxied-host: NULL / 500 flapping) for these two Spaces. Containers are healthy and answer internally; the failure is in the proxy layer.