When exactly do I need a

RunConfig

object? Do I need one if I’m already specifying an Executor?

They are separate concepts, but I think in this case you may not need one. Common use cases would be like setting the logger to debug

KubernetesRun(env={"PREFECT___LOGGING___LEVEL": "DEBUG"})

or adding other env variables (though I guess in your case you could do it in the executor. You can also provide an image to KubernetesRun so that it’ll match the Dask cluster? Like if you start to use

pandas

or

numpy

in your Flow, I think you’ll need to match versions between the client and workers.

So, when I specify a pod template (image) on DaskExecutor, KubernetesRun is still happening and using a different image?

I think the image on the DaskExecutor is for the scheduler and workers, but the KubernetesRun image for the client.