I see this as a neat way of exposing a graphql interface but a pattern with a number of limitations that are traded off against the goal of reducing the network latency. Definitely useful for users who want a single datasource and only want to expose access through GraphQL (cannot scale db resources indepedently of the api access layer). Wondering if the graphql engine could make use of the postgres stats collector for optimizing graphql resolver execution ex) resolver dispatch scheduling or pre-execution query optimization?