On Wed, 2009-11-25 at 13:16 -0800, Luke Kanies wrote: > On Nov 25, 2009, at 12:38 PM, Markus Roberts wrote: > > > One possibility we're overlooking here (I'm not making any claims > > apart from the fact that it's a distinct solution) is to bind a run > > to a server on the initial exchange (e.g. do a redirect from the > > generic "puppetmaster pool" URL to an equivalent but more specific > > "the particular puppetmaster who's handling you for this run" URL). > > Session based web services sometimes use this technique. > > > I'm amenable but I've no idea how common/supportable this is. Is this > often how load balancers work? I'd expect that if someone wants to > throw up an F5 in front of their masters that the F5 would be the only > route through to the masters, and I'd (somewhat naïvely) expect there > not to be another route to the masters.
Webapps usually get around that with 'sticky' loadbalancing - essentially, the loadbalancer can be told to look for a cookie and/or request parameter in the request and then makes sure that requests with the same cookie value always get routed to the same server. In the Java world, that's what the infamous jsessionid request parameter and cookie are for. David -- You received this message because you are subscribed to the Google Groups "Puppet Developers" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/puppet-dev?hl=en.
