Nitro

Our monolithic Nuxt app kept crashing at the 2GB memory limit. We doubled to 4GB — twice the cost, same underlying problem. After five days of failed optimization attempts (streaming compression, Redis polling, SWR caching), we split the monolith into three standalone services. Each got its own 2GB allocation, total cost dropped, and the crashes stopped.

How a seemingly correct cache key using url.hostname causes cache fragmentation between SSR and client requests, dramatically reducing cache hit rates in Nuxt/Nitro applications.