Hello,
I am wondering what is the best approach to put an upstream eBGP session in maintenance in order to minimize the packets loss.
OPTION 1: shutdown the BGP session
This is the hard/easy way! If the eBGP peer hasn't got a fallback route in its RIB (which is usually the case when using a route-reflector), then it needs to learn a new "best route". But how does this happen?
Answer 1:
- The router first and immediately removes the former route from its RIB as it's now invalid (session down)
- Then it informs the route-reflector that this route is invalid
- Then it gets a new best route from the route-reflector
Answer 2:
- The router keeps the former route in its RIB as long as it doesn't have completed the process of getting a new one - It informs the RR that this route has been withdrawn
- Then it gets a new best route from the RR
- It withdraws the former route from its RIB and inserts the new one
OPTION 2: stop the outgoing prefixes advertisement
Does this change anything compared to option 1 or is it exactly the same in terms of RIB/FIB update procedure and time left without a valid route (which means packet loss)?
OPTION 3: BGP graceful shutdown (set community 65535:0) if the peer implements RFC8326.
Does this option reduce the packet loss duration?
When the upstream router receives the BGP community, it will set the local-preference for the prefix to a low value (like 0). Then the router will inform the RR of the new local-pref
The RR will compute a new best route and send it back to the router
The router will then update its RIB.
I assume that, with this option 3, the upstream router always has a valid route in its RIB so there should be nearly no packet loss (except in case of AS nodes routing inconsistency for a short while during the convergence).
With options 1 and 2, the router does not always have a valid route, therefore there is packet loss for a while. Option 3 seems to be the best.
Note: of course there is also the opposite direction but I only mentioned the egress BGP advertisements in order to simplify the picture.
QUESTION 1: in the context of option 1, which answer is correct: answer 1 or 2?
QUESTION 2: is my assessment of the impact for each scenario correct? Any remark or hint?
Thanks a lot!
