Is OSPF backbone required for GRE?

Question

tunneling

Colleagues! Tell me, I'm assembling a network on OSPF, there is no common segment between marches, not a single one. GRE only. There are four marches now, there will be more, the connection as a whole is a star (to the core), but there are also links between marches separately (somewhere the path is shorter, somewhere there is a backup connection). From each march to the core there are at least two tunnels, sometimes more (the number of providers on one side * the number of providers on the other side). Who has a similar config, how does ospf behave when such links are placed in one area and the provider fails at the end of the star, in the core?
So, in which area should I place the tunnels (in this case, the tunnel = subnet, because each tunnel has its own transport network) between the core and the ends of the star? The documentation (what we managed to find on similar instances) says that this is a backbone - and the links should be in the backbone, but it's not shared (which contradicts the essence of the backbone), it's ptp. Create a separate area for each link group?

Answer 1

Is OSPF backbone required for GRE?

Backbone (if we are talking about area 0) is always needed. There are rare schemes in the wild with one non-backbone zone, but this is some kind of non-conventional logic.

Who has a similar config, how does ospf behave when such links are placed in one area and the provider fails at the end of the star, in the core?

I can only hypothesize about your situation. If all links are added to one zone (area 0), the dropped link will lead to the generation of a new version of type 1 lsa and the operation of SPF algorithms on all routers in the zone (we will bypass Incremental SPF for now). Thus, one router can influence all others. Here, of course, one can argue at what frequency of the link flap what will be the load on other routers, but it seems to me reasonable to lay the maximum tensile strength into the scheme, that is, place the links in different zones.
On the other hand, you mentioned direct links between regional routers, which, in my opinion, complicates the scheme with several zones (virtual link) and is an argument in favor of one global zone (area 0).
Generally speaking, why not try it on a lab, at least a virtual one?
PS I re-read my answer and realized that in addition to the obvious advice to check on the lab, there is little use in it.