Custom domain returned edge 404 "Application not found" for ~12 min while deployment stayed running (2026-09-30)

Question

## Problem and desired outcome

My production API at `server.forexizer.app` returned intermittent `404 "Application not found"` responses from the Railway edge for about 12 minutes on 2026-09-30 (roughly 07:36–07:48 UTC), even though the service's deployment was not changed or redeployed. Uptime monitoring opened an incident and at least one mobile app user saw failed requests. It has recovered and is not happening now. I'd like Railway to check what happened to routing for this service in that window and tell me whether anything on my side should change to prevent it.

## Resources and timing

- Project: Forexizer (`0f7a554e-52e2-4a8b-b0be-565633006138`)
- Environment: production (`9aa49b52-c6dc-4542-9b7d-96e7a1b05906`)
- Service: Forexizer_Backend (`b1dd9135-4e55-42be-82ca-f782bfe920ba`)
- Custom domain: `server.forexizer.app` (CNAME → `fyl7d1im.up.railway.app`)
- Running deployment: `7988a659-eba7-465c-9b55-44188412c25f`, status SUCCESS, created 2026-09-28 12:38:51 UTC. It is the same deployment before, during and after the failure window; there is no separate failing deployment.
- First failure seen: 2026-09-30 07:36:17 UTC. Last failure seen: 07:43:23 UTC (client), 07:42:24 UTC timeout (monitor, Australia). Fully recovered by 07:48:30 UTC.
- Current state (observed 2026-09-30 09:01 UTC): `GET https://server.forexizer.app/api/health` → 200 in 0.26s.

## Evidence and attempts

**External uptime monitor (Better Stack), checking `https://server.forexizer.app/api/health`:**

```
07:36:17 UTC  status 404 from Asia (5.223.73.226)
07:36:19 UTC  status 404 from Europe (168.119.246.194)
07:36:29 UTC  status 404 from Australia (45.79.236.121)
              body: {"status":"error","code":404,"message":"Application not found","request_id":"S5X0k3HrSqyHxxt2xtoGcA"}
07:37:15 UTC  body: {"status":"error","code":404,"message":"Application not found","request_id":"YQy-SN0vR6WR7dSrwUFZXw"}
07:42:14 UTC  recovered (Europe)
07:42:16 UTC  recovered (Asia)
07:42:24 UTC  timeout (no headers received) from Australia
07:48:30 UTC  recovered (Australia)
```

**Mobile app client (Sentry breadcrumbs, iOS, one user):** several requests hung ~15 s and then received the same edge 404:

```
07:42:47.7 → 07:43:02.9 UTC  GET  /api/exchange-rates?baseCurrency=USD&targetCurrency=CAD  → 404
              body: {"code":404,"message":"Application not found","request_id":"52BuJO8GTGyN8WllpHNmDw","status":"error"}
07:42:48.0 → 07:43:03.2 UTC  POST /api/calculations → 404
07:43:23 UTC                 two more POST /api/calculations → 404
```

Other requests in the same window from the same client returned 200/201, e.g. `GET /api/exchange-rates?GBP→CAD` → 200 at 07:42:47.98 UTC.

**Railway logs for Forexizer_Backend (exported from the dashboard):** the app logged incoming requests during the window (request lines stamped 07:42:39–07:43:24 UTC, e.g. `GET /currencies`, `GET /exchange-rates`, `POST /calculations`). All log lines from 02:49–08:48 UTC carry the same deployment ID, and there are no startup, crash, or exit messages. The app does not log `/api/health` requests, and no requests were logged between the 07:00 cron line and 07:42:39, so the logs don't show whether monitor requests reached the container between 07:36 and 07:42.

**MongoDB service in the same project:** WiredTiger checkpoint messages continue about every 60 s throughout the window.

**Checks performed:**

- `railway deployment list` (production): latest deployment is `7988a659…` from 2026-09-28; no deployment on 2026-09-30.
- Backend git repo: no commits on 2026-09-30.
- Railway status page: no incident listed for this time.
- Current response headers include `server: railway-hikari` and `x-railway-edge: cdg1`.

No configuration changes were made during or after the incident.

## Why contact support?

The AI assistant (Claude Code) suggested contacting Railway support, and I agreed. Its reason: the `404 "Application not found"` body comes from Railway's edge rather than my app, the deployment was never changed, and neither of us can see edge routing logs for the request IDs above. I'm asking Railway to look up those request IDs and explain why the edge couldn't route to this service in that window.

## Uncertainty and constraints

Observed: edge 404s with "Application not found" from three monitor regions and from a real client; an unchanged deployment; app request logs within the window; healthy database logs.

Hypothesis (not confirmed): a routing or service-discovery issue at the edge rather than an application failure. Evidence that complicates this: some requests hung ~15 s before the 404, and the Australia monitor saw a timeout at 07:42:24 UTC, which could also fit the container being briefly slow or unreachable. The app does not log health checks, so I can't confirm the container was reachable between 07:36 and 07:42.

I'm not asking for any changes to resources, only a diagnosis and advice.

Prepared with Railway's support template (v1).