Is it possible to have both direct iBGP peering between PEs and between PEs-RR ?
I am looking for a solution for a subset PEs who can continue communication amongst themselves, even if they don't have reachability with the RR for route updates. If direct peering and RR can coexist, then RR to be preferred as the best source for route updates and direct peeering update as the next alternative, in case PE losts its connectivity with the RR.
Any hint/suggestion would be of great help.
no routing loops should occur as BGP advertisements reflected by RR servers contain the cluster-list attribute and BGP originator attribute
direct BGP peering will provide a direct non-reflected advertisement that will be chosen over advertisements with a non null cluster list.
As noted by Laurent by using the cluster-id you can provide redundancy at the BGP RR server level with two RRS serving the same set of client PE nodes.
This is the typical setup in SP networks.
Eventually, different RRS are used for family ipv4 unicast and VPNv4 and also the VPNV4 space can be partitioned in multiple sub spaces served by different RRS (as a way to get better scalability)
Hope to help
Yes it will work but the update received from the RR may not be the one selected as best (but it doesn't really matter anyway)
You could also provide redundancy at the RR level which will maintain constitancy in your BGP PE config. It will also provide backup for all your PE's.