I think for the particular use case of index sharding we can be a bit smarter than exposing the session to the sharding strategy implementor. We should be able to do it for him provided we get a bit configuration information.

Right. As Steve suggested, something like this should do the trick to keep the feature independent from ORM:

ShardValueCollector { public List getShardValues(); }

Either way, I still see this work as a prerequisite, so I am going to complete it and we can take it from there.

+1!