Deployment

How changes go from developer machine to production.

Per-project deployment

OMS (shopify-esim-system)

• Hosted: Vercel (main production), with Cloudflare as DNS
• Branch → environment:
  • main → production at oms.gstarlink.com
  • Feature branches → preview deployments automatically
• Environment variables configured in Vercel dashboard
• Database migrations via Supabase CLI (supabase db push)

Handbook (gstarlink-handbook)

• Hosted: Cloudflare Pages
• Two separate Pages projects:
  • gstarlink-handbook-internal — builds with npm run build:internal, deploys to handbook.gstarlink.com, protected by Cloudflare Access
  • gstarlink-handbook-public — builds with npm run build:public (runs export script first), deploys to docs.gstarlink.com, unauthenticated
• GitHub Actions runs both builds on push to main

eSIM Hub (esim-hub)

• Status: WIP, target Cloudflare Pages or Vercel
• Will follow same pattern as OMS

Chatbot (gstarlink-chatbot)

• Cloudflare Workers
• Deployed via wrangler deploy
• Changes to main branch auto-deploy via GitHub Actions

Environment separation

Current state: no staging environment. Every change goes straight to production with Vercel preview deployments as the safety net.

Target state:

• main branch → production
• staging branch → staging (connected to a separate Supabase project)
• Feature branches → Vercel previews

Setting up staging is on the tech backlog.

Pre-deploy checklist

Before merging to main:

• Tests pass locally
• PR reviewed by at least one other person
• Database migrations (if any) are additive only — no destructive changes without a planned rollback
• Environment variables updated if new secrets introduced
• Changelog or ADR written if this is a significant change

Rollback

Vercel keeps previous deployments — rollback is one-click from the dashboard. Database migrations are harder — keep them additive to avoid needing data-level rollback.

Monitoring

• Vercel logs for OMS errors
• Supabase logs for database issues
• Cloudflare analytics for handbook and chatbot
• Shopify webhook delivery logs (for retail order ingestion issues)

TODO: set up a unified alerting channel (e.g. Slack #alerts) that all systems post to.