Il giorno 21/nov/2011, alle ore 13:35, Gunnlaugur Thor Briem ha scritto: > Roberto De Ioris <roberto@...> writes: >> no, if your load-balancer uses round-robin it will connect to the >> reloading instance and will wait until it is ready. But maybe you can >> "mask" this behavior using the "zerg dance" (i posted it in the previous >> mail) where you start a second instance before shutting down the first. > > "Zerg dance" doesn't sound like an activity I would ever want to indulge in. > :)
only if you are a terran :) > > More seriously, I read it and sure, it probably works, but I'd rather just > script this outside of uwsgi: pluck the instance out of the load-balanced > pool, > SIGHUP it, and put it back in. That fits better with our upgrade practice of > taking the instance out of production and sanity-testing it through a separate > nginx. I was just hoping the plucking-out and putting-back-in could be handled > transparently in the uwsgi protocol --- that would have been so very neat! this kind of transparent-auto-scaling behaviour, is accomplished by the subscription system and fastrouter. It works very well (already used in production in different mass-hosting server) but adds another layer to the infrastructure: nginx -> uwsgi fastrouter -> uwsgi nodes you may want to start looking at it, but i suggest you to wait til 1.0 as the fastrouterer (and the subscription system) are getting an extreme boost in performance and usability in the upcoming release (currently the 'unsubscribe' system is broken, but will be fixed soon) -- Roberto De Ioris http://unbit.it JID: [email protected] _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
