Remove service IP from routes if all endpoints are unavailable

[shane-davidson]

Not sure if I am doing something wrong or it is just by design but I am trying to run loxi using BGP ECMP but when all endpoints for a LB are down the routes are not being deleted.

We are wanting to run multiple instances of Loxi using the same service IP that can failover based on the endpoints being down and using BGP routes instead of using DNS for multi-AZ / multi-cloud failover.

Is it possible for the routes to be removed when all endpoints are down? If not is it possible to make this configurable?

I am running Loxi in-cluster and happy to provide my configs if required :)

[inhogog2]

Hi @shane-davidson, thanks for the great point. As you mentioned, even when all endpoints become inactive, the route is still being advertised via BGP. I appreciate you bringing this up.
A proper fix inside LoxiLB will require additional development work, so it may take some time. If this is time-sensitive on your side, feel free to reach out separately.
In the meantime, there is a manual workaround: you can use a BGP policy to stop advertising the specific route. Since each LoxiLB pod runs its own GoBGP instance, applying the policy directly to GoBGP can temporarily resolve the issue.
Below are the example commands along with a reference link to the official documentation:
https://github.com/osrg/gobgp/blob/master/docs/sources/policy.md

gobgp -p 50052 policy prefix add block 20.20.20.1/32 32..32
gobgp -p 50052 policy statement stmt-block add condition prefix block any
gobgp -p 50052 policy statement stmt-block add action reject
gobgp -p 50052 policy add pol-block stmt-block
gobgp -p 50052 global policy export add pol-block default accept

Thanks again for your feedback — I’ll update you once the feature is officially supported.