Your "web" service shows very low CPU and memory usage, so it is not resource-constrained on our side. However, the logs show repeated 502 errors from the Shopify Storefront GraphQL API during SSR, which is likely the primary driver of your slow p99 times since each SSR render is blocked waiting on those external API responses. The service is also hitting our 500 logs/sec rate limit with 1,782 messages dropped in a single burst. You can use our HTTP log filtering (e.g. @totalDuration:>1000) in the Observability tab to identify which request paths are slowest and compare totalDuration vs upstreamRqDuration to confirm whether latency is in your app's SSR or in the network path - more details in our performance troubleshooting guide. If your users are primarily in Europe, also confirm your service is deployed to the EU West region under service settings to minimize edge-to-origin latency.

our assumption at the moment is some misunderstanding of your hosting infrastructure. it seems the bad p99 is often on non AMS edges. since we have only AMS active why would be any of our application served via other edges? i can also see a good amount of cache headers (seen screenshot). is cache something we can configure?

is this something you re working on? https://nextjs.org/blog/nextjs-across-platforms

i'd appreciate you not making private conversations public without my consent. thank you