Thank you all for the input. You all touched on various things I've thought about.
A few things:
1 - This proposal is made with the primary goal being to make nodes searchable by this information. This isn't really possible (in a performant way) with our freeform JSON fields like properties or extra. Right now, CID is working on a spec that would
allow bulk node actions -- so think about the intersection of these ideas: "power down every machine in DC 17, row 5, rack 9" or "set all machines in DC 12 room 1 to retirement".
2 - I am concerned that there's too much variety in what people want, so far on this thread we've seen suggestions ranging from location info (as I mentioned) to serial/model numbers, to purchase date information.
3 - I do not expect this information to be immediately useful for scheduling. It's not a part of my current proposal but I'm sure someone else could take the next steps; it's just not a use case I find valuable. I'll also warn folks: the HA model of nova-fronted-ironic
is failure isolation; targetting to specific servers takes away any HA in these cases (e.g. if customer X always targets rack Y for deployments; that customer will experience 100% downtime if the nova-compute managing rack Y is down/under update).
I will likely pose this as a discussion item for PTG. Please participate when that time comes (and keep chatting here).
Thanks,
JayF