Hey folks,
My understanding is that a ZFS special device becomes a single point of failure for your dataset.
Therefore, I believe you should mirror it.
Wouldn’t 2 NVME’s suffice?
I see postings where people are saying that they use more than 2 devices, but don’t see how that’s necessary.
Advice?
Yep, if the main pool is redundant, then the special device should be as well, or the whole pool dies.
You can make a mirror deeper, by attaching extra drives to it, so like 3 deep will have the capacity of just one device, but any two of them can fail at the same time, or one can fail, and while waiting for a replacement, another can fail, and you still have a working pool.
I don’t think you can raid special devices (not sure that is even on a roadmap or not)