ADR-003: Jobs & Scheduling — Vercel Cron + Supabase Jobs Table
Status
Accepted
Context
Our ingestion workers need a reliable nudge every few minutes. Supabase’s pg_cron isn’t guaranteed in every project yet, and we wanted scheduling that fits naturally with the Vercel deploy pipeline.
Options considered
- Vercel Cron → Next API → Edge Functions — Hosted schedule, secret headers, and easy observability in Vercel.
- GitHub Actions — Works, but slower feedback and harder to secure with secrets.
- pg_cron + pg_net — Attractive once it’s generally available, but not portable today.
Decision
Use Vercel Cron to ping /api/cron/rss and /api/cron/x. The handlers (see web/src/app/api/cron/rss/route.ts) validate the shared secret, then call the appropriate Supabase Edge Function with the service-role key. Every run writes a log row into public.jobs so we can trace failures.
Consequences
Upside
- Configuration lives alongside other Vercel project settings.
- We can replay a job by curling the cron endpoint locally.
- The
jobstable doubles as an observability feed; seesupabase/functions/rss-fetch/index.ts.
Trade-offs
- We now depend on Vercel uptime for ingestion triggers.
- When we eventually migrate to pg_cron we’ll need to update the ADR and remove the HTTP middleman.