On 20 January 2016 at 13:33, Gordon Sim <[email protected]> wrote: > On 01/20/2016 12:53 PM, Rob Godfrey wrote: > >> There is no single point where "it is fully configured for use" in the >> case >> of HA a virtual host may transition from a REPLICA state to a MASTER state >> (or vice versa) at any time, >> > > In the case of transferring from MASTER to REPLICA, simply dropping the > underlying tcp connection would be better (since it would be more likely to > cause the client to try an alternative url).
Potentially - the issue then is that the client doesn't see any message as to why the connection is being dropped, so they don't know what they are doing wrong (if they have forgotten to properly configure their client to round robin amongst potential brokers). > > > similarly virtual hosts may be stopped and >> started by management actions. The broker architecture separates the >> ports >> (which accept incoming connections) from the virtual hosts... until the >> virtual host is chosen, the broker cannot know whether it can service the >> request. >> > > Understood. > > As you say, it is unfortunate that we don't have alternative >> error codes for auth failure and other transient connection issues... but >> not much we can do there. Would it be possible to add some way to have >> the >> client keep retrying even if it gets a 320? >> > > It would certainly be possible, yes. It would have to be a configurable > option though. Yes - that would make sense. Similarly I guess the Java Broker could have an option to simply disconnect the TCP connection rather than say why it is doing so. Thanks, Rob > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
