Building on Cloudflare Workers: Why Edge-First Matters

2 min read

The entire content pipeline runs on Cloudflare. Not because it’s trendy — because it’s the right tool for a globally distributed content system.

The stack

Component	Cloudflare Service	Purpose
Site	Workers + Astro	SSR blog + API
Database	D1	Subscribers, posts, jobs
Cache	KV	Sessions, rate limits
Storage	R2	Media, OG images, archives
Queues	Cloudflare Queues	Fan-out distribution
Email	Email Services	Newsletter + confirmations
AI	Workers AI (optional)	Embeddings for search

Cold starts don’t exist here

Lambda cold starts: 500ms–5s. Workers cold starts: <5ms.

When a visitor hits a blog page, the Worker is already warm. The Astro SSR renders in milliseconds. D1 queries resolve locally in the same data center. No round-trip to us-east-1.

D1: SQLite at the edge

D1 is SQLite replicated globally. For a content pipeline, this means:

Subscriber lookups are local
Job state queries are fast
No connection pooling headaches
Built-in migrations via Wrangler

The trade-off: eventual consistency across regions. For a blog and newsletter, that’s fine. Subscriber count doesn’t need to be real-time accurate.

The $5/month budget

For a solo creator, cost matters:

Workers: 100k requests/day free
D1: 5GB storage, 5M reads/day free
KV: 100k reads/day free
R2: 10GB storage free
Queues: 1M operations/day free

The paid plan ($5/month Workers Paid) unlocks higher limits and production features. That’s the entire infrastructure cost.

What about Vercel/Netlify?

They’re great for frontends. But the content pipeline needs queues, a database, cron triggers, and email. On Vercel, you’d add:

PlanetScale/Neon for database ($20+/month)
Upstash for Redis ($10+/month)
Resend for email ($20+/month)
QStash for queues ($10+/month)

Cloudflare bundles all of this. The DX trade-off is real (Wrangler isn’t as polished as Vercel CLI), but the cost and latency advantages are undeniable.

Lessons from production

Use wrangler dev locally — miniflare simulates all bindings
Migrate early — D1 migrations are immutable, plan the schema upfront
Queue everything — fan-out pattern with per-platform queues is cleaner than one mega-worker
Cache at the edge — KV for hot data, D1 for persistent state

The edge-first architecture isn’t just about latency. It’s about simplicity. One platform, one bill, one deployment target.