Anas EL BARKANI

@fedepaol @karampok shouldn't actually be the BGP graceful shutdown feature that should be implemented to avoid the limitations described in MetalLB doc ? Example for Juniper https://www.juniper.net/documentation/us/en/software/junos/cli-reference/topics/ref/statement/graceful-shutdown-edit-protocols-bgp.html

@karampok the limitation I had in mind is when you remove a node from the cluster for maintenance or just for down scaling. Shouldn't be there a sort of signaling...

> In all cases the BGP peer will not have anymore routes towards the node. Existing tcp connection will break Well this is what we want to avoid, breaking existing...

> Now for trafficPolicyCluster you could argue that TCP connection from external client A to pod/EP B in k8s, which established through node X but reaches pod/EP B in node...