Is it possible to improve this micro benchmark? Of course, by bundling all the inserts in a single transaction, for example, or by using another, non-async database driver, but it does not make sense as it's not how a real-world codebase accessing a database looks like. We favor simplicity over theorical numbers.

I don't understand. Using transactions and prepared statements you would get hundreds of thousands of inserts per second even using languages like python/ruby.

Nothing theoretical or not-real-world about it. It is how SQLite docs recommend you do it.

https://www.sqlite.org/faq.html#q19

https://rogerbinns.github.io/apsw/tips.html#sqlite-is-differ...