[Blue sky] Single session, multiple connections

[Daniel Lundin]

Decoupling session and connection/transport is an attractive design.

Since a connections is ineherently coupled with data distribution
(token range), it would be cool to use the "closest" open connection
to the data you're operating on..

mysession.get('foo')

With a connection pool with N open connections to the cluster, it'd be
pretty sweet to get the one where the 'foo' token lives.

Just a thought.

/d

[Eric Evans]

Indeed. I'm not sure how'd you go about this cleanly though. Through
`describe_ring' you can obtain a correlation between token ranges and
endpoints, but without replicating the partitioner logic you wouldn't
be able to match keys to nearest tokens. I dunno, maybe the
partitioners are stable enough that that'd be OK.

--
Eric Evans
john.er...@gmail.com