Add read/write timeout to connection pool

[dropbox]

Add read/write timeout to connection pool

[pattyshack]

Yeah, that's sound very reasonable; we probably should set the timeout unconditionally, even when timeout is non-positive, to avoid the issue where write has a timeout but read doesn't (and vice versa). I would love to have contribution from non-dropboxers!

The reason we haven't found the need for timeout is because our client always talk to memcached co-located in the same data center (i.e., network partitioning is not a huge issue). (in general) We do not run memcache on AWS.

[pattyshack]

o, i just noticed Conn supports SetWriteDeadline & SetReadDeadline. We should just use that instead of resetting on each read / write call

[pattyshack]

(obviously, i'm bad at reading since that's exactly what you proposed ...)

[pattyshack]

we have timeouts!