Move Operation from Keyspace to CassandraClient

[hz]

Move operateWithFailover and operateWithFailoverSingleIteration from KeyspaceImpl to CassandraClientImpl. Leave keyspace with a clear skelton to package the thrift api.
If the users don't want to use KeyspaceImpl, they can use CassandraClientImpl still.

[zznate]

Within hector, the read/write operations all happen in Keyspace, CassandraClient encapsulates the connection to an individual Cassandra instance. Failover in this context is within the data operation on the inherently distributed keyspace, so those methods live there. In other words, we fail over to another client from within a keyspace operation, not vice versa.

If you choose to get a handle to the Cassandra.Client thrift stuff from CassandraClient, then you are loosing the functionality in Keyspace and to a large degree are on your own anyway.

[rantav]

@hz your point is in place and this is a design issue I've been uncomfortable with from the beginning, but I agree with @zznate that the solution you suggest isn't going to work, at least no as things are now.

If you want to use Keyspace without failover you have 1 choice currently, to use FAIL_FAST.
Still you'd incur the (small) overhead of creating a client etc, but logically I believe it gets you what you need. I agree that this is not as clean, from the API perspective, but logically it's the same.

What we can do, though is maybe to implement two Keyspace implementations - one SimpleKeyspace and one FailoverCapableKeyspace and let the clients instantiate whichever Keyspace flavor they prefer.

I see this as a code enhancement, not a real feature. Let us know what you think.

Also, if you need this badly, we love contributions, just fork the project, make your changes and ping us for a code review.

[hz]

I had forked one before I created this issue. Let me think more about it.

[hz]

So, before any result is reached, I close this issue firstly. :)

[hz]

I added one interface: HectorRequest, to package the failover and pooling features.
The implements HectorRequestImpl implements the operation and failover.
The third class is HectorThriftRequest which extends HectorRequestImpl. It package all the thrift api in it.
Keyspace now initiate with a hector request. Each failover operation is run as request.thrift_api(parameter).
But you do some tricks on getColumn. I don't understand its logic.
I am not investigate the failover logic you used. So the code have many many mistakes. Please take some time to browse my code and decide if you can merge some into your trunk.
Thanks.

[rantav]

Hi @hz, I'm sorry I had browsed http://github.com/hz/hector and couldn't find your changes. Could you post here (or email the dev mailing list from http://wiki.github.com/rantav/hector/mailing-lists) a link to your commits (something like http://github.com/rantav/hector/commit/4321125f79e9c3198ea2116527cd07546d737483)

Thanks!