In general some of the workloads just needs a certain dependencies and doesn't care where they are executed (vmx or aws for example). So I think it's possible to extend the labels with multiple driver/definitions. That's useful for the second (or third, ...) fallback to use the resource outside of the pool or smaller machines in case huge ones are not available.
Not sure about the way to implement that - it will be hard to store multiple definitions as one label (json could work, but it's a bit tricky) or store definitions separately and link to the label (which introduces an issue with label atomic versioning and complication of creating/updating labels)...
In general some of the workloads just needs a certain dependencies and doesn't care where they are executed (vmx or aws for example). So I think it's possible to extend the labels with multiple driver/definitions. That's useful for the second (or third, ...) fallback to use the resource outside of the pool or smaller machines in case huge ones are not available.
Not sure about the way to implement that - it will be hard to store multiple definitions as one label (json could work, but it's a bit tricky) or store definitions separately and link to the label (which introduces an issue with label atomic versioning and complication of creating/updating labels)...