On Wed, Nov 11, 2015 at 10:20 AM, Paul Jakma <[email protected]> wrote:

> On Wed, 11 Nov 2015, Daniel Walton wrote:
>
> but quagga is doing this check earlier then I could see you getting into a
>> situation similar to the one you describe below.  So then you would need a
>> bestpath knob that controls where in the decision algorithm this comparison
>> happens. There are already so many knobs in bgp...this might be more
>> trouble than it is worth (and confusing for the end user).
>>
>
> Agreed on BGP already having way too many knobs to twiddle. :)
>
> How about a followup patch that provides a knob to disable this comparison
>> in bestpath?
>>
>
> Hmm, that still allows for operators to mismatch it and have subtle probs?
> It took a long time for BGP engineers and academics to understand
> oscillation and figure out how to provably prevent it. Most operators
> probably would struggle to even recognise it if it happens, never mind
> debug.
>
> I wonder could we add a capability to describe the routing decision steps
> and their ordering, e.g. as a list of code-points (each code-point being
> some recognised decision process), and refuse to connect when there is
> divergence in the min(length(their list),length(our list)) initial set of
> that list. (i.e. it doesn't matter if one speaker implements additional
> tie-breakers, as long as both are agreed on the sequence before that).
>
> That would let changes to the decision ordering be made and rolled out in
> a planned fashion, without risk of undetected weird problems, at least.


That is a much much bigger can of worms that would have to go through IETF
which can take a really long time.

Donald and I were chatting about this....I think we are overcomplicating
this knob:

   - We've had this behavior for 2 years and have not had any customers hit
   any issues with it
   - On the off chance that the stars aligned and a prefix oscillated back
   and forth between A--B one could always set local-pref, origin, etc (any
   attribute that is compared earlier than cluster-list length) to stabilize
   that prefix. I think the odds of that happening would be slim though, just
   because the IGP cost to the nexthop would likely cause A to always choose
   the E path.

So we can avoid adding a new knob and we already have a way to work-around
this corner case scenario.

Daniel
_______________________________________________
Quagga-dev mailing list
[email protected]
https://lists.quagga.net/mailman/listinfo/quagga-dev

Reply via email to