We're seeing ~20ms overhead per GraphQL request when using 16+ services. The overhead comes from creating a new ManagedRuntime with fresh layers for every request.
Why we need per-request runtime: DataLoader scope isolation and our ConnectRpcService reads RequestContext (accessToken, signal) when the layer is built, not when methods are called. The token gets captured in a closure. I tried couple of diffrent solution but my effect-ts knowlage is still limited and I cannot figurated out a better way, any suggestions much appreciated.