US consumer platform

50 million images. A much lighter origin.

Bhavin moved repeated image delivery off origin on a high-traffic platform, while keeping a clear path for updates and invalidation.

50M+ / dayImages served through the delivery pipeline

Work delivered by Bhavin in his engineering role at an anonymized consumer platform. Measurements refer to the image delivery workload.

Repeated image traffic was wearing out the origin.

A high-traffic consumer platform was serving tens of millions of images a day. Most of those requests were for content that had not changed, and they still landed on origin.

Move the repeat work off the origin, and keep updates explicit.

Bhavin rebuilt delivery as a read-through cache with Cloudflare Workers, R2, and D1. The Worker normalized requests and managed the read path. Edge cache served hot content, R2 held durable copies, and origin stayed the source of truth — and the place missing variants were generated.

Canonical keys kept variants consistent. Conditional requests avoided resending unchanged image bodies. A signed invalidation path handled replacements and deletions, with D1 tracking the relevant object keys.

The measured result

The pipeline served more than 50 million images a day. CPU on the image path at origin fell from about 90% to under 15%.

Cached requests measured about 250 ms at p50 and 670 ms at p95. Requests that missed both the edge cache and R2 reached roughly 1.9–2.1 seconds at p95. Those numbers are from this workload, not a general platform benchmark.

The design also had to deal with an image changing, a variant going missing, or a location holding stale content. Those paths were made explicit so the system stayed understandable in production.

Developer Labs brings the same approach to client delivery problems: put repeated work where it belongs, and make the update and invalidation path visible.

Bhavin wrote a longer engineering account on his personal site: 50 million images. A much lighter origin.

THE NEXT GOOD QUESTION

What could your
business do better?

Bring the messy workflow. The scattered data. The idea you've been putting off.
Let's find the right place to start.

Discuss your workflow A focused conversation. A practical next step.