Edge Databases: Where the Hype Meets Reality

Edge databases promise low latency from anywhere. The reality is more nuanced. Here's where they actually win and where they cost more than they save.

"Edge database" became a buzzword in 2024. The pitch: your data is at every Cloudflare PoP, queries are <10ms from anywhere, the world is wonderful.

The reality: edge databases are great for read-heavy public-data use cases and bad for almost everything else.

What I tried

Three projects, three providers:

• Cloudflare D1 (SQLite at the edge)
• Turso (LibSQL, distributed)
• Neon with read replicas in multiple regions

Where edge wins

• Read-mostly workloads with global users
• Public data that doesn't have sensitive write semantics
• Low write rate - eventual consistency is acceptable

A docs search index, a public product catalog, a pricing page - perfect.

Where it doesn't

• Writes from anywhere. All these systems funnel writes to a primary. Your write latency is not edge - it's "round trip to wherever the primary is."
• Strong consistency across writes and reads. Read your own write? You'll need read-from-primary or sticky routing.
• Anything transactional in the multi-row sense. Cross-row consistency is dicey at the edge.

Numbers from a real client

A SaaS dashboard: started on Neon multi-region, migrated to single-region after 4 weeks. Why?

• Reads were already cached at the CDN; the database read latency wasn't the bottleneck
• Writes (filling the dashboard data) were a small fraction of traffic; the cross-region complexity wasn't worth it
• Operational complexity (which region is "primary"?) cost more eng-time than the feature delivered

My current take

For most apps, a single Postgres in a smart region + good caching beats edge databases. The latency saved by edge reads is usually masked by the latency you didn't optimize elsewhere.

Edge databases are the right answer for ~10% of read-mostly, geo-sensitive workloads. They're the wrong answer for ~90% of the projects that adopt them.

Don't pay the operational tax for the marketing benefit.

References

• Cloudflare D1
• Turso
• Neon