Currently, the short answer is no. The best way to do it right now would be to run a MapReduce job and then feed the results back into GFXD.

We are discussing ways to do this for future releases, so if you have a concrete use case we would love to know about it. Given that the in-memory, operational data set will be orders of magnitude smaller than what is on HDFS, knowing exactly what to read back into memory is the big question. Doing that as fast as possible is also important, but somewhat secondary.

Just adding that you can still operate over that data by using queryHDFS hint to your statements or even at the connection level (so all statements will be using the hint by default). Of course that it will have some performance implications when compared to data in-memory only.

William, Jens, Thank you for your responses !

Currently, I don't have a concrete use case, but not for long !

regards