Tim,

> On Sep 22, 2022, at 8:02 PM, Tim Evens (tievens) <[email protected]> wrote:
> 
> When I wrote route-refresh, that was from the router/sender side, not BMP 
> receiver sending a route-refresh. BMP is still write-only.
>  
> There are use-cases for re-syncing the RIB to the BMP receiver.  Some of the 
> methods today are to clear the peer, request a route-refresh in, or to reset 
> the BMP feed. This is a bit intrusive to the router (sender) and BMP 
> receiver. Having to go to the router to request a refresh to the BMP server, 
> IMO, is still okay.  The problem is that control knobs for refresh are at the 
> BGP peer level, not BMP.  For Adj-RIB-In Pre-Policy, it makes sense to do 
> that on the peer, but for Post-Policy, Adj-RIB-Out and Local-RIB, it doesn’t 
> really make sense.  Having a knob to initiate BMP side refresh would be very 
> nice and hopefully less intrusive.  

Thanks.  This is now clear.

> Regardless of the BMP server needing a refresh, a BMP server does not know 
> when someone has requested route-refresh IN at the peer level (maybe it was 
> requested for policy change, …) Instead, the BMP receiver, blindly, receives 
> a boat load of updates with no awareness that it was a new RIB dump or 
> controlled refresh. In this case, it would appear to be churn.  IMO, there is 
> value in having a message to signal the receiver that a refresh is coming, 
> followed by an EoR.  A PEER_UP could be used for this, but that is intrusive 
> and misuse of PEER_UP.

I think my concerns are in terms of incremental impact on existing BMP 
receivers.  Two thoughts come to mind about encoding:

1. A new BMP message type is used to signal a refresh for the appropriate 
logical view, and similarly that the end of rib markers that this thread was 
originally discussing would happen. :-)

2. Deliberate mis-use of peer-up.  One option is to add a new informational TLV 
to cover that this is a refresh.  For this point, the question to ask of 
receivers is what they would do if they received a peer-up when they already 
had state covering a session that is up.

A quick glance (but not thorough audit) of RFC 7854 is that it seems silent on 
what would happen if you got a redundant peer-up.

While I'd be generally inclined for the new message option, I could sort of see 
cases where existing BMP collectors might be happier using existing peer-up 
triggers for things.  But I could similarly see where some might consider it a 
flavor of state machine violation and may not be happy with it.

-- Jeff

_______________________________________________
GROW mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/grow

Reply via email to