What's the most scalable rules engine that's open source and implements the RETE algorithm? I'm looking for something that could be run on an RDMS and handle millions of rules, queried from several connections, but most of the open source systems I've found, like CLIPS and JESS, only operate on flat files and have no built-in server ability.
Jena has built reasoner support which the Jena project has provided documentation, which cites that it uses a RETE engine. You could then run this reasoner on top of data store built with SDB (though TDB is preferred it is not an RDBMS). Whether it could handle millions of rules seems a lot to ask of any reasoner but I guess it would depend on how disjoint the rules are.
As for the scalability of reasoners you can find an evaluation of Fact++, HermiT and jcel at the SEALS - Storage and Reasoning Systems Evaluation Campaign 2010 results
Drools has a community addition. Although I've given up trying to find it's actual licence on the web site...
How about looking at rules support in the HermiT reasoner? HermiT handles DL-safe SWRL rules and can be added directly to the input ontology. It's open source, but I'm not sure exactly how scalable it is.
And of course the Pellet reasoner also has SWRL rules support (and it available under dual licensing).
I'm suggesting these since your question does not make it clear on why exactly the rules engine needs to implement the RETE algorithm, and I would actually like to know whether others can comment on the scalability of using the rules support of these two reasoners.