Ceph » rgw

This is going to be implemented using watch/notify. Each node keeps its own cache but notifies other nodes about changes. Cache should have an LRU, memory size should be configurable. The cache information for each object should contain:
- object data
- xattrs
- negative lookup

This is done, using watch/notify to sync between the different daemons. However, we should probably make a few changes so that it is more scalable.

This should include a (stress?) test that gives us some confidence it is working properly. Something like:

- start up two radosgw instances
- for N iterations
- write A or B to one of M random objects on one rgw instance
- maybe to xattr only, data only, and/or updates to both
- read same object from other instance and verify we see the correct value