Make the call to cassandra.describe_keyspace fault tolerant

[rantav]

The call to cassandra.describe_keyspace is at CassandraClientImpl.getKeyspace()
This is a "meta" call, which is invoked once per keyspace before any other operation on that keyspace is performed.
This call, if unsuccessful, will not use the failover policy used by all other read/write operations.
It should.

[hz]

I don't know how to merge my modifications to my fork. I pasted my solution into gist.github.com/hz, 360995 is a interface, 360996 is request featured with failover and pooling. 360997 is the wrapper for thrift api. 30999 is the revised keyspaceimpl. you do some tricks on getColumn. I can not solve it.
After the clean wrapper of thrift api, this issue will be solved.

[rantav]

Sorry, it's pretty hard to see diffs in gists. It's preferable that you push back to your hz/hector repo so I can see diffs and send line comments, but if this isn't working for you, then email me a patch

[mgrogan]

Any idea when this problem is likely to be resolved? Occasionally I experience a transport error which causes the call to getKeyspace to fail. When this happens Hector does not recover and all subsequent calls to getKeyspace fail.

[rantav]

I guess I was just waiting for someone to need this bad enough to implement that...
If you're not into hacking that I'll try to find some time for it this week

[Mishail]

Well, http://github.com/Mishail/hector/commit/a8c9a0accd49d1e8f76ad15d8250f608f8a9d755

Though I'm not sure that all is done properly... The code is mostly borrowed from CassandraClusterImpl. And all unit-test are passed (except one, which I marked as @ignore and specified the reason)

[rantav]

fixed on master e5f29a4..1bf8ab5 master -> master