Comment by weiznich | Hacker News Reader

Comment by weiznich | Hacker News Reader

Parent Full thread

Please note that up until today an async rust database library does not give you any measurable performance advantage compared to a sync database library. In addition to that it won't even matter for most applications as they don't reach the required scale to even hit that bottleneck. As a matter of facts crates.io just run well on sync diesel up until maybe a month ago. They have now switched to diesel-async, for certain specific not-performance related reasons. The lead developer there told me specifically that from a performance point of view the service would have been fine with sync diesel for quite a while without problems, and that's with somewhat exponential growth of requests.

Other than that the async rust ecosystem is still in a place that makes it literally impossible to provide strong guarantees around handling transactions and make sure that they ended, which is a main reason why diesel-async is not considered stable from my side yet. This problem exists in all other async rust database libraries as well, as it's an language level problem. They just do not document this correctly.

WEIZNICH! I am your biggest fan. Thank you so much for the amazing crates you've made and for Diesel.

When I started my rust journey Diesel and Rocket were the first few crates I worked with and it really opened up my mind and I haven't turned back since.

As for async, my primary concern for that was to make sure you minimise blocking code calls in your futures. Even if there weren't many performance gains to be made from Diesel itself being async but having database calls marked as Futures could theoretically help the runtime schedule and manage other threads.

I'd like to thank you for you amazing work once more!

> As for async, my primary concern for that was to make sure you minimise blocking code calls in your futures. Even if there weren't many performance gains to be made from Diesel itself being async but having database calls marked as Futures could theoretically help the runtime schedule and manage other threads.

That doesn't require an async database library at all. It merely requires some abstraction like [`deadpool-diesel`](https://docs.rs/deadpool-diesel/latest/deadpool_diesel/index...) that ensures that the caller always uses `tokio::spawn_blocking` or similar to run database queries. And yes this scales rather well as the number of database connections in your application is rather low (usually a few ten) and therefore much smaller than the number of threads in the tokio blocking thread-pool.

Reply on news.ycombinator.com