Shadow device-like configuration?

Hi there,

I am evaluating options to deploy a high-performance cluster and take advantage of instance storage available on AWS; the corresponding concept in Aerospike is shadow devices.

What I would ideally like to achieve is:

  • storing 3 replica copies for all data, one of them on the NVMe instance store
  • reads prefer routing through the NVMe-powered storage process
  • bonus: somehow tell FoundationDB that the copy stored on the NVMe instance is in a “null” zone which is never fault tolerant

Can something like this be achieved, or this is not an idea which could be translated in something useful in FoundationDB clusters?

I am exploring various options, so thanks in advance for any input!