This is one of those things that has always sounded really interesting but I haven’t had the opportunity to use. Does anybody have experience with it? I’m also trying to find some benchmarks but don’t see much.

As a real database language with ad-hoc-queries you're fine with SQL. Data-intensive graph analysis (the reason soufflé, souffle-lang.org, was developed) where you have a fixed set of deduction rules and change the database instance is an ideal playground for Datalog.

And soufflé is incredibly fast and for the usual benchmarks, like the transitive closure in comparison to recursive CTEs in SQL, blows everything out of the water. (my paper on that http://ceur-ws.org/Vol-2368/paper3.pdf, Figure 3)

I'm researching using Datalog as a general purpose programming language (specifically for use in microcontrollers, like Arduino) for my PhD and it can be very expressive.

What usable implementations/libraries of datalog would you recommend?

I would suggest soufflé (https://souffle-lang.github.io/) if you're batching data through for static analysis.

DDLog (https://github.com/vmware/differential-datalog) if you want a rule-engine within your interactive application, to encode decision procedures and business rules.

If you actually want this to drive the views of a (web-)application, I don't think we're really there yet. We have to wait at least for the next generation of developments. But you can give it a try with datascript https://github.com/tonsky/datascript