Also, a hostId as required by zfs.

And account for the odd-host-out. :/  See if you can spot it.

The layer above will have to make sure that it is.

There will be no staging grid and the other two nodes in this grid went
elsewhere.

Define testing grid and production grid with a function they can share.

There is not going to be a staging grid.
There will be a testing grid and a production grid.
The testing grid will be loosey goosey.
The production grid will have incremental rollouts.

Also, comments.

This results in less repetition, less abstraction leakage, and provides more
power to the private-storage module (by exposing meaningful values instead of
Tahoe-LAFS config strings).

This is a bricked machine in the data center where running NixOS is hard.