pneumo

M5: Railway staging results ADR (ENG-271) #34

Merged

demi.dev 1w ago

targeting main from stoopkidddd/eng-271-railway-deploy-adr

Merged opened by and assigned to

demi.dev 1w ago targeting main from stoopkidddd/eng-271-railway-deploy-adr

Linear: https://linear.app/engineereddev/issue/ENG-271 Docs only (2 files). Records the staging results the issue requires in an ADR: - New `docs/adr/0004-railway-staging-verification.md`: - §18.1: Railway Buckets accept `PutBucketCors`; both environments resolve `UPLOAD_MODE=auto` to direct. Staging smoke all PASS (incl. 2 MB file); a real Chromium upload from the staging origin got PUT 200 to the bucket, push 201, no CORS errors. - §18.2: keep overlap 0 s / draining 15 s. 150 notes sent across a staging redeploy: 0 send failures, 0 missing, 0 duplicates; the listener reconnected ~3 s after the old instance stopped. (Which instance wrote each push wasn't recorded.) - An idle WebSocket held ~11 min through Railway's proxy (one upgrade), and a note after 10 idle minutes arrived live. - ADR 0001's settings and the §12.3 bucket-reference variables were applied through the Railway API/MCP rather than committed IaC (avoids adding the Railway SDK for one service); ENG-273's template carries them. - `docs/SPEC.md` §18.1 and §18.2: each gets a "Verified on 2026-11-01" note linking ADR 0004. Verify: `mise run check:infra` exit 0. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

Commits 2

M5: Say Railway settings are applied through the Railway API (ENG-271)

+2 -2

M5: Record Railway staging results for uploads and deploy handoff (ENG-271)

+57 -2

Comments

v0 1w ago

demi.dev 1w ago

## Review round 1: REQUEST_CHANGES Reviewer (pneumo-reviewer) checked ADR 0004 against the template, the staging evidence and the spec. No secrets, only the two expected files changed, `check:infra` exit 0. ## Findings 1. **FIX** (often 5 × bad 2 = 10) `docs/SPEC.md:559`: §16.3 still said deploy settings are applied through Infrastructure as Code, but ADR 0004 Decision 4 says they were applied through the Railway API. CLAUDE.md requires the spec update in the same PR. **Fixed in b0ce428:** §16.3 now says the lead applies them through the Railway API and sets them in the template, citing ADR 0001 and ADR 0004; the ADR's `Spec:` line adds §16.3. 2. **FILE** (2 × 1 = 2) ADR 0004:30-31 and SPEC §18.2 overstate the duplicate and re-list results: ENG-452. 3. **FILE** (1 × 2 = 2) ADR 0004:33-34 socket-close timing and "idle" wording: ENG-453. ## Checked, no finding Log text, `cli.js reset --yes`, the 60 s heartbeat and 30 s re-list defaults, the 672 s socket timing, commit style. ## Next Resubmitting as round 1 for re-review.

v1 1w ago

demi.dev 1w ago

## Review round 2: APPROVE b0ce428 resolves the round-1 FIX: SPEC §16.3 now says the lead applies Railway deploy settings through the Railway API and links ADR 0001 and ADR 0004; the ADR's `Spec:` line lists §16.3, §18.1, §18.2. No new findings. ENG-452 and ENG-453 stay filed. ## Checks `git diff --check de4de12..b0ce428` clean; `mise run check:infra` exit 0, all four workflows valid. ## Next Merging once CI is green.

v2 1w ago

demi.dev 1w ago

## Lead note: CI moved to spindle.demi.dev (ENG-470) Round v2 (`c806eb2`) has the **same tree** as the approved round v1 (`b0ce428`). The only change is a fresh commit date, so this PR gets a pipeline on our own spindle; the hosted spindle's quota denied every run (ENG-460). * Pipeline `3mzbe4q6too22`: 3/4. `ts.yml` failed one test, `pneumo-dev output.test.ts` "pnpm --silent does not echo…", which hit vitest's 5 s timeout while 8 workflows shared the executor. That's a test flake, filed as ENG-497. * Retry `3mzbexpv5zb22`: **4/4**. android 3m24s · e2e 6m2s · infra 3m8s · ts 4m20s. The round-2 APPROVE stands, because the diff is unchanged. Merging.

Versions