The ovn-nbctl daemon in kube-ovn-controller need to keep a long connection, the default inactivity_probe will drop the connection.

ovn 2.12 use a new incremental process engine that increase the performance of ovn-controller. However we meet inconsistency between ovn-sb and ovs several times, so we pick the recompute patch from upstream and periodically invoke it to keep consistent

Sometimes nbctl-daemon may hang and not response anymore and all request to ovn-nb will failed. We need to check nbctl-daemon repeatedly and use liveness prob to kill the controller that can not connect to ovn-nb for a period.

Refer to https://github.com/ovn-org/ovn-kubernetes/pull/986