Comment 3 for bug 1746615

> We should never be relying on placement code directly on the compute node. (This includes introspecting os-traits.) We should always be asking the placement service.

While this statement might be correct, it isn't aligned with how we've discussed os-traits in the past, now how we've made use of the 'fields.ResourceClass.VCPU' (and others) fields in practice.

The idea has been that the os-traits package itself is the authority and which version of that something has available to it is the controlling factor. This does present the challenge you describe but the mechanics of fixing it so that placement becomes the authority for resource classes or traits becomes weird: where do the constants that are used in "client" code come from?

It's not always the case that the client is retrieving something from placement. Sometimes it just wants to say "I have _this_ resource class and _these_ traits".

What's your expectation there?