I made a lab that simulates an ISP connected to another local-area Metro ISP that acts as the infrastructure for the endpoints.
There is equally connection in both sides from the Metro network to the core device of the main ISP which creates redundancy.
Now, I ran into a big problem when I simulated that one of the Metro PE’s that connects to the ISP’s PE fail over and get stuck:
- the interface for the ISP still in UP state
- the Metro PE stuck and can’t forward traffic
- the ISP’s core device still learns his best BGP routes through the fault path because the ISP’s PE is directly connected to the whole Metro subnets as it acts with L2VPN service using VPLS.
Therefore a tracking object isn’t working, or at least can’t make it work as the RIB is populated with the BGP redistributed routes from the fault PE side.
Could any1 suggest a solution for such a scenario? maybe a sample configuration for a working tracking object?
If the ISP’s PE’s interface goes into DOWN state then everything is fine, but if something will happen to the Metro’s PE’s Interface (like making bad configuration for a subset of VLANs) then all those costumers will still be redistributed as they are configured on the ISP’s PE’s interface and therefore the bad configuration on the Metro, they will lose connection for a long time until some1 will fix the problem in a real-life scenario.
I totally have forgotten about that feature but maybe conditional advertisement might help me solve this problem if I will engineer my network in a way that only if I have a route through the Metro cloud, only then I will advertise my networks through the BGP.
Thanks you very much.