Hi, this relates to the "second-chance cache" (https://github.com/xtdb/xtdb/pull/1158) - the performance difference between this and the LRU cache will really depend on your data set and hardware, but I expect it will only be single-digit-% slower when using LRU. To test the difference yourself you should be able to try downgrading your JDK, or enabling the flag as described (see also https://github.com/xtdb/xtdb/issues/1462#issuecomment-802739964)

As in the equivalent of (keyword (name kw))? There's nothing built-in to our pull implementation to help with this. One option is to do the transformation within Datalog by wrapping your query in an outer query and calling out to Clojure functions. Is there definitely no chance of key collision in your data doing this though?

OK. I haven’t really put much thought into them when working in clojure/clojurescript. It just feels really unnatural from javascript which I’m dabbling with right now

but there won’t be collisions, no. It’s just a small pet project, so there’s no deep nesting going on

That sounds fair enough. Are you using XT embedded in your app? Or over HTTP?