Should we make sure the primary os down here? I would want the mirror to operate like the primary at this point I think.

Should we inherit the subjects from the old primary here?

We can't assume we will be able to reach the primary stream - this is something that will likely be done during a outage, the primary is dead or so down that its unreachable etc and people now are doing recovery playbooks.

So I dont think we can assume its reachable, ditto about being caught up - thats assuming the primary is reachable.

After this the primary Is basically unusable because it cant again become a mirror of this one.

Lots of moving parts in making this right and I think thats all tooling based moving parts, lots of confirming user intentions, asking if its available or not doing checks where the tooling can maybe even fetching the subjects but handling cases where its not possible etc

Yep, I think our consensus is that the server can't reliably judge what the right thing to do is, after all the mirror origin could be online in our domain, online but in a different domain, offline-ish/flaky, offline never to return. The best we can do is just surface the promotion as a primitive operation in the server and try our best to surface it in the tools in a way that makes sense to operators for different playbooks.

Yes, the main use case for this is IMHO is DR scenarios (specifically 2 sites DR), where we don't have any solution right now.

IMHO this isn't meant for 'automated' multi-site fault tolerance: you must use a stretch cluster (or multiple clusters with cross sourcing) for that, but for disaster recovery where the administrators makes the decision to make what was the mirror at the DR site into the new primary, the stream being mirrored is going to be unreachable/down at that point. Let them control adding the subjects as part of that DR operation (the subjects could be a little bit different anyways depending on how the applications running at the DR site are configured). Just like it's going to be an administrative operation when they decide to do the reverse when the primary site is back up. They will indeed likely have playbooks for handling disasters that will involve many administrative operations, with NATS being just a small part of it.

@derekcollison If you're happy with this consensus then we can merge it for experimenting in preview.2 this week.

Might be good to add in same subjects from upstream source if we know them. Could be opt-in and fail if we can't reach it?

I think it'd be better to investigate possibilities like that in a separate PR, I suspect we'd have to issue a stream info to find those and a stream update doesn't feel like the right place to do that.