The best part I think is "pg_query is special. It doesn’t actually implement parsing SQL. It works by extracting C source code directly from Postgres and wraps it with a nice Rust interface. This allows PgDog to understand all queries that Postgres can."

One of the things that puts metaprogramming on wheels is being able to pull in grammars, schemas and such for external systems. Think how many cool tools we could have if we could do this in a more general way.

Having written several pg txn poolers, I like the implementation. I also love your simple branding and I think this could make a fantastic product for enterprise customers. However, my biggest concern as a buyer is longevity. PgCat (another popular pooler in rust) will likely lose funding shortly. Meanwhile, PgBouncer is battle tested, small in scope, and is starting to get consistent new contribution.

pgdog looks interesting, but I read the docs wondering how it handles foreign keys between tables, and the docs don't seem to cover it. It was the first question I had, and I'd assume would be an important question for most users. The project states "Data relationships are not sacrificed to go faster" and that it operates "transparently to the application", but it's not clear how it does that in regard to foreign keys.

Additionally, maybe this is just a pet peeve with Postgres documentation in general, but the docs use "database" to mean both a single database and a Postgres server interchangeably. On some pages this makes the content harder to understand. I think it's good practice to distinguish between databases and "clusters" (no, not those clusters, actually servers), and perhaps between clusters and groups of servers. While some of the naming is unfortunate for historical reasons, re-using the same names for different concepts just causes more confusion.

Wrapping up the actual Postgres code is a brilliant idea. Then you never have to worry about updating your code to keep up with their changing API.

And the fact that this works at all is another example of why Postgres is the default data store on the internet. It's battle tested and has a massive ecosystem of support tools, like this one.

>we need to ... understand SQL syntax using a parser, Rust ecosystem has a great library called pg_query

I'm impressed both that the Rust ecosystem had the right library available and that it was high enough quality code for the author's team to use it as a reference for building their own additional functionality.