Hi Marco, On Thu, Nov 15, 2012 at 11:29:03AM +0100, Marco Corte wrote: > Il 12/11/2012 17:59, Marco Corte ha scritto: > > Hi all again. > > I was able to reproduce the behaviour on a normal server, so it is > unrelated to the virtualization. > Sorry for the long description.
This description is much more understandable than the first one :-) > Normal situation: > servers A1-A4 in status NOLB - L7OKC/404 > servers A5-A8 in status UP - L7OK/200 > > Action: > I disconnect the nework cable > Effect: > all servers in status DOWN - L4TOUT > > Action: > I reconnect the network cable and I wait for the checks > Effect: > servers A1-A4 in status DOWN - L7STS/404 > servers A5-A8 in status UP - L7OK/200 > > Action: > I enable one of the "DOWN" servers putting the check page in place say on > A4 > Effect: > servers A1-A3 in status DOWN - L7STS/404 > servers A4-A8 in status UP - L7OK/200 > > Action: > I remove the check page from A4 > Effect: > servers A1-A3 in status DOWN - L7STS/404 > server A4 in status NOLB - L7OKC/404 > servers A5-A8 in status UP - L7OK/200 > > It seems that the status does not recover from DOWN to NOLB directly, > but the transition DOWN -> UP -> NOLB works. > > I did not find nothing specific to this behaviour in the changelog for > version 1.4.22. Any ideas? It is the intended behaviour : the state machine is that after switching from NOLB to DOWN, the server remains down until an UP status is seen. Keep in mind that this is made for people who have to start/stop their servers multiple times when deploying. So they first say they want no new visitors on the servers, and once the servers are down, we want to ensure that nobody will bo to these servers even if they return a 404. So what you describe is the right and intended behaviour. Regards, Willy
