I think I want to run different tasks with slightly different Kubernetes templates in one run, e.g. if machine resources could vary between tasks (GPU or not) or each task has a different entrypoint in the kubernetes template. Is using dependent flows the way to go there? I assume each dependent flow could have its own

run_config

then. Is that use case normal or am I thinking about it wrong? Just to double check that I'm not attempting anything totally unidiomatic here.

Hey @Ben, This seems idiomatic for your use case as currently task runs cannot be given resource configurations individually. So breaking these into singular flows with their own

run_configs

looks like the best way to accomplish what you’re after. I’d say just be cognizant of your run resource allocation and flow concurrency to ensure you’re not overloading your infrastructure unknowingly.

I see, thanks for the answer! I'll play around with the approach a bit and see how far it gets me. 🙂 Configuring such things per-task would seem useful to have at some point for Flows with computationally heterogenous tasks, so to speak. But I don't know enough about Prefect so far to comptetently assess whether that fits with your design goals for the system. Thanks a lot for the quick response!