Railway Support Ticket — persistent brotli corruption fetching deno.land/std from the build environment

Persistent brotli corruption fetching deno.land/std in Docker builds — 6 failures today, same file every time

Environment: staging. Services ft-advance (cron, worst hit) and

proud-vision, both building the same railway-worker/Dockerfile

(root = repo root, denoland/deno:2.8.1, RUN deno cache … fetches ~290

remote modules).

The error — byte-identical in all 6 failed builds

error: Import 'https://deno.land/std@0.132.0/node/internal/streams/from.mjs' failed.
    0: brotli error
    at https://deno.land/std@0.132.0/node/internal/streams/duplex.mjs:23:19
[ERRO] [9/9] RUN deno cache main.ts forwardTestJob.ts dispatcher.ts
Build Failed: ... exit code: 1

Always this one file; the ~290 sibling downloads in the same builds succeed.

Full logs: railway logs --build <deployment-id> (IDs below).

Build matrix (2026-08-07, UTC)

| Service | Deployment | Commit | Time | deno fetch | Result |

|---|---|---|---|---|---|

| ft-advance | d0b5db71 | 8190340 | 13:14 | fresh | FAIL (brotli) |

| ft-advance | 5a2cd2ec | 8190340 | 13:33 | fresh (manual retry) | OK |

| proud-vision | 68cf9400 | d328852 | 13:35 | fresh | OK |

| ft-advance | 679b7dc8 | d328852 | 13:35 | fresh | FAIL (brotli) |

| proud-vision | 28574da9 | eaefbd1 | 14:13 | fresh | OK |

| ft-advance | d68e4458 | eaefbd1 | 14:13 | fresh | FAIL (brotli) |

| proud-vision | 2c8ee4e7 | 4687229 | 14:25 | fresh (Dockerfile edit broke cache) | FAIL (brotli) |

| ft-advance | 18181bbb | 4687229 | 14:25 | fresh | FAIL (brotli) |

| proud-vision | 8cc887c8 | ef441c2 | 14:27 | layer-cached, 0 downloads | OK |

| ft-advance | 9567cf5c | ef441c2 | 14:27 | fresh | FAIL (brotli) |

Reading: identical Dockerfile + commits on both services, so not commit

content. ft-advance fails 5/6 fresh fetches; proud-vision 1/3 — and its

final "green" downloaded nothing (layer cache masked it). Correlation is

with which builder/egress path each service's build lands on. Fetching the

URL from outside Railway is clean (curl with Accept-Encoding: br: HTTP

200, decodes to the correct 918-byte module) — a cache/edge on your egress

path is serving a corrupted brotli object for this one file.

No retry path

The dashboard retry control (used for the 13:33 success) is no longer

offered on the failed deployments, and `railway redeploy --service

ft-advance --yes` refuses: "The latest deployment … cannot be redeployed …

currently building, deploying, or was removed" (latest = FAILED 9567cf5c;

nothing was building).

Asks

1. Flush/invalidate the build-env cache or egress path serving the corrupt

   object for https://deno.land/std@0.132.0/node/internal/streams/from.mjs,

   or reassign the affected builder (ft-advance seems pinned to the bad one).

2. Restore a retry/redeploy control for failed deployments (dashboard and

   CLI both refuse, above).

3. Confirm if this is a known deno.land CDN incident on your side.

Impact: the cron service is stuck on the 13:33 build; we cannot ship newer

commits to it.