2025 Team
Aurora — a programmable edge cache
A globally distributed cache with per-request invalidation rules, cutting p99 latency by 62% across 30+ regions.
- Role
- Tech lead · 4 engineers
- Year
- 2025
- Links
- Case write-up ↗GitHub ↗
- −62%
- p99 latency
- 30+
- regions
- 99.99%
- availability
The problem
Our origin was buckling under read-heavy traffic with long invalidation tails. Off-the-shelf CDNs gave us caching but no programmable control over what to invalidate, when, and where.
Approach
We built a cache layer on Cloudflare Workers with Durable Objects as the coordination primitive. Invalidation became a first-class, rule-driven operation expressed in a small DSL, evaluated at the edge in Rust compiled to WASM.
Outcome
p99 dropped 62%, origin load fell by an order of magnitude, and product teams could express cache policy as code reviewed in the same PR as the feature.
Stack
RustWASMCloudflare WorkersDurable ObjectsOpenTelemetry
Comments
Loading comments…