Currently the cache mode we use is hard coded to be write-through. We had it at none, and the default before. And it always seems someone will come up with some storage where our current default does not work well. We have a request to allow users to configure it.

I believe there are 3 possible levels at which it makes sense to configure:
1. At a cluster level, a single configuration option that applies to the entire cluster. This will work if the cluster contains a homogeneous storage and the global option will be suitable for this case.
2. At a storage class level, a configuration per storage class level (maybe even in the storage profile?) This suitable if the cluster contains multiple types of storage with different requirements. This is a good mix between little configuration, and some flexibility.
3. At the DV level. Maybe the default is fine for most general cases in your cluster, but a few images require something different. If you can set it at the DV level these cases can be handled. This one might be overkill if the other two are implemented.

This configuration seems very analogous to the fsOverhead configuration we currently have. We can set it at a global level and a storage class level.