One of the goals for our Relational Algebra COmpiler (Raco) is that it should be a “relational algebra middleware” for many big data system backends. At the moment, we support: Myria (Myria Team) and Grappa (Brandon Myers), plus anything that speaks SQL including SQL Server but also Hive or Shark. Expanding this set will make the middleware vision real.

I spent today hacking on the Raco → Flink pipeline. As of tonight, I can produce Flink Java programs from MyriaL programs for a fairly fleshed-out SPJA subset of the language. I’m hoping that this experiment will

1. help us make Raco a better middleware for big data systems. (stay tuned!)
2. lower the barrier to entry to using Flink (do you REALLY want to write Java code like this?)
3. make it significantly easier to compare big data systems head-to-head. (stay tuned!)

Want to try exporting code for your favorite system from Raco/MyriaL? Let us know!