If the memcached server goes down for whatever reason, yet comes back straight away (server restart etc) the client
socket will raise Broken pipe errors. I didn't notice any supported away to handle this, and I didn't want to messy
up my code with trying to reconnect, so I've added the ability for an auto reconnect when a Broken pipe error is raised.
There may be other IOError types it should catch too.

As for the unit test, I'm not sure how to write it as it'd require a prompt for the user to manually to restart their
memcached server or something?

Date: 2006-12-07 23:12
Sender: Eric Hodel

I've looked at your patch, and it won't work as is.

Errno::EPIPE is a subclass of SystemCallError.  Rescuing exceptions
by message name is fragile, and not to be done.

I've tried an alternate implementation, but I've not yet gotten
something acceptable.

I am working on attempting to reconnect once, then failing.