sensible configuration of max_connections

Hi All,

What's a sensible way to pick the number to use for max_connections?

I'm looking after a reasonable size multi-tenant cluster, where the
master handles all the load and there's a slave in case of hardware
failure in the master.
The machine is used to host what I suspect are mainly django
applications, so lots of short web requests, not sure how much, if any,
django's orm does connection pooling.

I arbitrarily picked 1000 for max_connections and haven't had any
problems, but onboarding another app that handles a few million requests
per day  on Monday and thinking about potential problems related to the
number of available connections.

What's "too much" for max_connections? What happens when you set it to
high? What factors affect that number?

cheers,

Chris