Hi @Louis_Frolio,

Thanks for your complete answer! I really appreciate that.

I've two questions:

1. For setting a driver node as on-demand and my workers as spot instances. How is it possible to do? Because If I would to create a new compute pool, I couldn't decide which is spot compute and which not. As you can see below:

Maybe, do I need to create two different pools one for the driver and one for the workers and create one all on-demand and the other pool all spot-instances?

Can you provide me the yaml boilerplate for writing it in a resource file, we never did it!

2. When is it suggested to use Spot Instances? When the job execution is more than xx minutes or..? I read something around and I'm thinking that our job maybe doesn't need spot instances at all!