That error says your cached build is corrupted, You can temporarily disable the Build Cache by add NO_CACHE=1 to the Service Variables and remove it once the deployment start working again

Set NO_CACHE=1 and it does disable cache in one build pass (all steps succeed with no errors), but the deploy still fails — it looks like the builder runs two parallel build passes per deploy, and the second one still hits the same corrupted cache ref (7fnxf205i5vfm8f14vjk7hy9h) and fails, taking the whole deployment down with it. Service: fred-web (b8b1155c-647b-43a3-9917-7f442b31527d), builder: builder-fhdqfr.

Set NO_CACHE=1 and it does disable cache in one build pass (all steps succeed with no errors), but the deploy still fails — it looks like the builder runs two parallel build passes per deploy, and the second one still hits the same corrupted cache ref (7fnxf205i5vfm8f14vjk7hy9h) and fails, taking the whole deployment down with it. Service: fred-web (b8b1155c-647b-43a3-9917-7f442b31527d), builder: builder-fhdqfr.

Try to deploy latest commit by clicking on your service, press CTRL+K, then deploy latest commit. If that fails, try to create a new service entirely to avoid this caching issue.

Thanks! Unfortunately my repo isn't connected to GitHub on this service — I deploy via railway up (CLI upload), so there's no 'latest commit' for Railway to redeploy from. I'd rather avoid creating a new service for now (would mean migrating the custom domain and re-setting env vars) unless there's no other option. Still stuck on the same cache ref (7fnxf205i5vfm8f14vjk7hy9h) even with NO_CACHE=1 — is there a way for someone on the Railway side to manually clear/reset the build cache for this service?

I really dont know what to do. 😭

Can you try run railway up again on your local machine?

Just tried again — still the exact same error, same cache ref: failed to compute cache key: failed to calculate checksum of ref 7fnxf205i5vfm8f14vjk7hy9h::...: not found. Tried this many times over the past few hours (with and without NO_CACHE=1), always the same ref. At this point it really looks like this specific ref is permanently corrupted on builder builder-fhdqfr and needs a manual clear on your end — is that something support can do directly?

Update: I tried creating a brand new service (fresh builder, no shared history) and it hit the exact same failure pattern on the very first build attempt — failed to compute cache key: failed to calculate checksum of ref ::...: not found, failing on random COPY steps each time. Different service, different builder (builder-vdeatz this time, vs builder-fhdqfr on the original), different cache ref, but identical error signature. This tells me it's not a corrupted cache tied to one specific service — something more systemic in the build pipeline for this project/account. I've also noticed the build logs consistently show what look like two overlapping/duplicate build passes per deploy attempt (duplicated 'FROM' and 'load build context' lines interleaved), and one of them seems to fail computing its cache key against the other's in-progress cache writes — could this be a race condition on your builder infrastructure rather than an actually-corrupted cache entry? Tried 5+ times on the new service alone (with and without NO_CACHE=1), same result every time. Would appreciate someone looking at this directly — project comfortable-manifestation (19eb9b53-8d57-4311-b8a9-a09fabfcb3d9), services fred-web (b8b1155c-647b-43a3-9917-7f442b31527d) and fred-web-v2 (24e6e315-0c68-4f2e-b42c-e24aa3dcb170).

Following up — I've now upgraded to Pro specifically to get direct support on this, but the system won't let me open a Private Thread since there's already this open thread for the same service. Could someone move this over to a private channel, or have a Railway engineer with builder/infra access take a look here directly?

Summary of the issue: every deploy on fred-web fails during build with the same cache-corruption error, on essentially every attempt since 2026-09-15 — over 24h now, 20+ attempts:

failed to solve: failed to compute cache key: failed to calculate checksum of ref ::...: not found

What I've tried:

Plain retries, many times.

NO_CACHE=1 (per your earlier suggestion) — disables caching for one build pass (visible as 🚂 Caching Disabled in the logs, and that pass completes cleanly), but the deploy still fails. Logs consistently show two overlapping/duplicate build passes per attempt — one succeeds, but a second redundant pass hits the corrupted cache ref and fails, taking the whole deployment down.

Renaming the Dockerfile path in railway.json to force a different build definition — same failure.

Toggling builder Dockerfile ↔ Nixpacks — no effect (config-as-code resets it).

Created a brand-new service (fred-web-v2) specifically to rule out a service-specific cause. Failed with the exact same error signature on its very first attempt — different builder (builder-vdeatz vs builder-fhdqfr), different cache ref, identical failure pattern. This points to something systemic in the build pipeline for this project, not one corrupted cache.

IDs:

Project: comfortable-manifestation (19eb9b53-8d57-4311-b8a9-a09fabfcb3d9)

Services: fred-web (b8b1155c-647b-43a3-9917-7f442b31527d), fred-web-v2 (24e6e315-0c68-4f2e-b42c-e24aa3dcb170)

Builders seen: builder-fhdqfr, builder-vdeatz

This is blocking a real production bug fix for active users. Would really appreciate someone with builder/infra access looking into it directly, plus a rough timeline. Happy to share specific deployment IDs from any of the 20+ failed attempts.

Following up — it's now been over 24 hours since I posted this in the Private Thread (and over 48h total counting the original public thread), with no response from anyone on the Railway team.

To be clear on what I need: not another suggestion to retry, clear cache, or check the community — I've already exhausted every client-side workaround (25+ deploy attempts across two separate services, NO_CACHE=1, rebuilding the service from scratch). This needs an engineer with actual builder/infra access to look at the corrupted cache ref directly.

I'm on the Pro plan specifically for this kind of support. Could someone confirm this is being looked at, and give me a rough timeline? This is blocking a real fix for a live app in active use.