Project: nangman-api (production, asia-southeast1-eqsg3a Metal, 1 replica) and nangman-api-dev (same region)

Observed: 2026-09-30 07:40–07:50 UTC, from Seoul, Korea

Symptom

• GET https://nangman-api-production.up.railway.app/api/health: 90+ probes, about 8% receive zero bytes (curl --max-time 5 and 60 both time out; http_code 000).
• TCP + TLS complete normally (time_appconnect ~0.17s, remote 69.46.46.115, server: railway-hikari, x-railway-edge: sin1), then no response at all.
• Same rate over HTTP/1.1 and HTTP/2. Successful requests take ~0.25–0.3s.
• nangman-api-dev (separate service, same region) shows the same ~7% no-response rate → not app-specific.
• Application logs show no errors, no restarts, no crashes in the last 2 days. Sleep is off. Health check path not configured.
• status.railway.com shows all operational.

Impact

• Our admin UI makes several API calls per screen; one hung call stalls the whole page for minutes.

Ask

• Please check the sin1 edge → Metal host path for dropped requests for these services. Request IDs cannot be captured because the hung requests never return headers.

Repro

for i in $(seq 1 30); do curl -s -o /dev/null -w "%{http_code} %{time_total}s tls=%{time_appconnect}s\n" --max-time 5 https://nangman-api-production.up.railway.app/api/health; done