I haven't looked at the pound code to see how modular it's parts are 
but I had this idea for people who don't want a server that is for 
example spewing 503 errors replying to clients.

I have a way to issue "override" commands to pound to manually flag a 
backend as down. For scheduled downtime or having a watchdog script hit 
the webservice and disable it when it gets a reply it doesn't like this 
would be kind of handy. I'd assume a tcp port for injection of commands.

Has anyone mentioned something like this before? What would you see as 
downsides to this sort of thing?