I'd suggest you check the Pool level setting Action on Service Down, the default is None;
"None – established connections (and NEW connections through persistence?) continue to be used as long as data is sent and received. Connection management relies on TCP for flows on both sides of the proxy; LTM does not intervene and persistence is maintained. Consider using this if your servers may refuse new connections (including the one used by the assigned Health Monitor) but will continue to service existing ones."
I'd suggest you try Reselect (although I think there are some implications that I can't remember). OneConnect also seems to help as well.