Related: T1311 (conntrack flush hanging with conntrackd, resolved)

When flush-connections is enabled, the WLB daemon runs conntrack --delete and conntrack -F expect on startup and on every interface state change. This destroys all conntrack entries system-wide, including unrelated NAT translations (hairpin SNAT, DNAT port forwards) and established connections that have nothing to do with WLB.

Any existing TCP connection or NAT mapping is killed, causing brief but total loss of connectivity for all services, not just WLB-managed traffic. The flush should be scoped to connections on WLB-managed interfaces only.