@jeremydmiller @khalidabuhakmeh not a clue about AoT for it in this instance haha, wasn’t a jab at it not supporting it, more ignorance on how entities were generated and everything else involved in the runtime behaviour for marten. The cold start trade off is what it is for runtime generation, I don’t really care if the first request is longer on start up because ???? yeah no shit its cold start of course its going to be slow. AoT was good for catching any devs trying to be too clever with reflection and getting them to stop doing dumb things causing real perf issues on hot paths.

I’ve tried the pipeline docs a few times and like you said it works with a deployment pipeline, so maybe I misunderstood how to have it generate models in a similar way to when running the app normally for dev and the models generating as required. It’s on the list to revisit and have more time to focus on it so normal generation works fine for now.

The project is complicated as there’s a scenario requirement to support local storage for isolated environments, so having data access in an entirely different namespace and project maintains a level of abstraction required to support both and there’s no way to shift the responsibility on to users to maintain a postgres instance. So knowing it is possible to shove my square peg into a triangle hole and have things in another project means we can keep going with this approach for now.