video1.0<iframe src="https://www.loom.com/embed/cfa8efdbb7d74a3bbeb718ed08e65c7d" frameborder="0" width="3840" height="2880" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>28803840Loomhttps://www.loom.com28803840https://cdn.loom.com/sessions/thumbnails/cfa8efdbb7d74a3bbeb718ed08e65c7d-09eb29c7c96810f5.gif1767.538This Loom explains how Depict moved production data processing from Kubernetes to the Cloudflare edge, focusing on their Shopify to merchandising sync and what nearly broke it. Depict describes a non-incremental Kubernetes approach that took up to 6 hours for product changes and even led to a $123,000 Cloudflare mistake on the way to meeting a two minute sync target. In the key deep dive, they stream and sort Shopify collection membership data using Durable Objects to handle out-of-order results, but initial billing reached $121,399 before cutting costs to about $6,000 per month by reducing work based on the updated DAT timestamp. They also share additional scaling lessons around Shopify Bulk Operations and Durable Object queueing, and conclude that while Cloudflare improves isolation and throughput, debugging and observability are harder and limits can be painful.