[c-nsp] VRF randomly, stably install only one from multiple route sets

[Oliver Boehmer (oboehmer)]

>> I don't think so. For eBGP paths, we don't
>> compare the router-id and stick with the
>> "oldest" path. But for iBGP we can't do
>> this and perform a deterministic decision
>> which will, given all other attributes like
>> localpref, as-path, igp-metric/etc. being
>> equal, compare the router-id.
> 
> I see. Is it even technologically feasible?
> I wonder whether Cisco could implement an
> IOS option to make iBGP to behave like eBGP
> in that respect? Just out of curiosity.

not sure, it sounds dangerous as path selection should be deterministic
within iBGP..
 
>> Why do you need this? Maybe there are
>> alternatives?
> 
> I now face a very similar scenario, where PE1
> becomes unstable, stops exporting R1, central
> VRF-C switches to R2. Then PE1 would flap, its
> unstable routes R1 taking over R2. Then again,
> PE2 may also become unstable. Fortunately,
> both R1 and R2 (almost) never become unstable
> at the same time. The definitive fix is to
> stabilize those PE routers, and we've been
> working on this with Cisco for months, without
> an acceptable work-around. The scenario described
> in the previous mail could provide an acceptable,
> if temporary, stabilization method. Of course,
> alternative suggestions are appreciated.

You might want to check the "import" option to ibgp-multipath. This way
we'll import more paths into the VRF and we can fail-over to alternate
paths without waiting for the periodic import-scan.

	oli