It's great to see more tools in this area (querying data from various sources in-place) and the Lambda use case is a really cool idea!
I've recently done a bunch of benchmarking including ClickHouse Local, and the usage was straightforward, with everything working as it's supposed to.
Just to comment on the performance avenue though, one area I think ClickHouse could still possibly improve on - vs OctoSQL[0] at least - is that it seems like the JSON datasource is slower, especially if only a small part of the JSON objects is used. If only a single field of many is used, OctoSQL lazily parses only that field, and skips the others, which yields non-trivial performance gains on big JSON files with small queries.
Basically, for a query like `SELECT COUNT(*), AVG(overall) FROM books.json` with the Amazon Review Dataset (10GB), OctoSQL is twice as fast (3s vs 6s). That's a minor thing though (OctoSQL will slow down for more complicated queries, while for ClickHouse decoding the input is and remains the bottleneck, with the processing itself being ridiculously fast).
Godspeed with the future evolution of ClickHouse!