If you are using a hardware raid controller then you can afford a spare.
If you data is important to you then that would probably be wise.
One of the early controversies I had to deal with was an architect that had a strong bias against software raid and said the company should move to this at some juncture. Except there are some advantages to a raid controller like detection and faulting of bad media. We would have needed to invest writing failure detection software as well as features to manage this. In the end, there can still be human intervention involved to determine why drives split and which is the authoritative mirror. I know, I had a lot of software raid before and it wasn't fun.
Our case, I argued eventually we should drive the secondary drive and move to a different model for local data store and how we guarantee nodes. ie, fast fail and replacement.
All of these things really dependent on your environment, the cost of productivity and ultimately how much you want to spend to be available. Personally, I would never use the stuff again, but some people really like it. I'm not sure they have used it at scale.