v0.80.1

Fixed

• A burst of route-server client session downs no longer withdraws slowly. With rs_control_communities enabled, the default for route_server_client, the RIB checked each departing session's withdrawals separately for every remaining client, so concurrent downs queued behind one another. In the scale benchmark's flap-storm cell (700 clients, 400,400 prefixes, 50 clients down at once), withdraw p50 fell from about 3.5 s on v0.80.0 to about 0.25 s. Withdrawals now share the update-group path unless a withdrawn route carries a control community for the client's route-server ASN; those keep the per-client check, so a client a route was suppressed toward is still not sent its withdrawal.