handling client connections during a rolling upgrade of the grid infrastructure
probably this topic has been discussed before but I could not find anything suitable.
I would like to know how you handle client connections during a rolling upgrade or while applying a (rolling) patch of the grid infrastructure.
When we want to apply a patch on our RAC (containing 2 or more nodes) we face the problem that we have to kill partly very old client connections and disturb the running applications. Before patching one node we relocate the services. Of course this works fine and new connections do not connect to the node we currently want to patch anymore but there are still the long-term sessions (partly a few month old). I think this is an issue of the middleware or the connecting applications!? After relocating my services all client connections should leave the node as soon