[distroinfinity]

Why

All CLI users saw Distro TV — offline demo (backend not reachable) instead of channel content. Root cause: the prod Postgres (Neon free-tier) exhausted its account compute-time quota (HTTP 402) → compute force-suspended (Jun 19) + branch archived (Jun 20) → API DB connections timed out → /me/content/next 500'd → CLI fell back to the demo fixture. Contributing factors: the uncached /health SELECT 1 kept the Neon compute pinned active 24/7 (burning the quota), and no alerting meant the outage was silent for ~4 days.

What

Prod Postgres moved to Railway Postgres (co-located in the API's Railway project, private networking) — already restored out-of-band (Phase 0). This PR is the supporting code + docs (Phase 1):

• railway DB target — env.dbTarget accepts local | neon | railway; the dev-safety guard assertEnvSafe now refuses any non-local target in dev (was neon-only). neon/local keep working.
• dbUrlForTarget + pool options — extracted, testable URL resolver; explicit postgres() options (max: 10, connect_timeout: 10, idle_timeout: 30) so idle connections close.
• migrate runner — resolves any remote target and guards its direct-run so test imports don't trigger a live migration.
• cached probeDb — 10s success cache mirroring probeRedis, killing the per-poll SELECT 1 storm (only success is cached; failures re-probe and surface fast).
• docs — gitbook + CLAUDE.md now say Railway Postgres; incident recorded; alerting follow-up tracked.

Spec + plan: docs/superpowers/specs|plans/2026-06-23-prod-db-railway-migration-*.md

Validation

• 40/40 API unit tests, typecheck, build all green.
• New migrate.js with DB_TARGET=railway run against the real Railway Postgres → [migrate] up to date (target=remote).
• Branch server booted with DB_TARGET=railway → /health 200 ok, and /me/content/next served real news + markets (no demo fixture).
• Live prod user flow (device bearer → /me/content/next) returns 5 news + 5 ticker, real headlines + live quotes.

After merge (manual, gated on this deploying first)

Flip Railway DB_TARGET=railway, then verify /health is ok and deploy logs show [migrate] up to date (target=remote). neon still resolves identically, so the deploy is safe before the flip.

Follow-ups (non-blocking)

• /health degraded alerting (the actual "never silent again" fix).
• Align drizzle.config.ts remote fallback to DATABASE_URL_UNPOOLED ?? DATABASE_URL (pre-existing, off the prod path).

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
devdrip	Ready	Preview, Comment	Jun 23, 2026 12:25pm