Async And Pooling

[genusistimelord]

How would promote_to_send and such work across threads for a connection. If I were to say create a single connection and multiple threads tried to use this connection. Of course, I would place this behind a Mutex, but I was wondering if that is necessary or how would it function if it was a RWlock instead of a Mutex. Since I notice the Connection never needs to be mutable?
Would it be more advisable to Create a new connection and set it as Send each time like we do in the r2d2_odbc connector?
I am trying to understand this as I am going to create an Async library layer for ODBC-api for Axum. So, I am just making sure I have all its limitations correct.

Also, Markus, do you have a discord/telegram account?

[pacman82]

I would place this behind a Mutex, but I was wondering if that is necessary

Yes this is necessary. It actually depends on the capabilities of the driver and runtime configuration wether this is safe at all. Yet independent even of that it is not possible to assure the right diagnostics and up at the right thread without an encapsulating mutex.

Since I notice the Connection never needs to be mutable

mut is a misleading keyword. It actually means exclusive. There is plenty of mutation in the connection, but it must be able to be shared by several statements. The safeness of this in a multithreaded scenario depends on the odbc driver correctly protecting its internal state with a mutex. I read the ODBC reference as stating that drivers must do this, but have seen this fail in practice (in the past though). This is why I put the promotion behind an unsafe call. I've seen you put this behind a safe interface. Be concious about this decision. Your users may experience race conditions and segfaults all in safe code. However you can also plausible state that this is a driver bug in that case. It is a judgment call.

Would it be more advisable to Create a new connection and set it as Send each time like we do in the r2d2_odbc connector

More advisable than what? Probably not getting that question right. Personally I would create a new connection for each thread that needs one. Using the odbc-api innate connection pooling.

I am trying to understand this as I am going to create an Async library layer for ODBC-api for Axum

This is just FYI: The ODBC - C API his innate async capabilieties. Two different kinds actually. So at one point this crate may get an async interface, too. However there are more pressing issues right now. Using native system threads to emulate "true" async is the way to go now, then building on top of this crate.

Also, Markus, do you have a discord/telegram account?

Never used any of these services.

Cheers, Markus

[genusistimelord]

thank you for the quick reply Markus. yeah mostly I am building around what exists and when you do decide to put in async i can just update any code to match it or depreciate those library's at that time in place of using this directly then. I will go with creating a new connection each time versus making a few connections and passing them on. generally that is what my question was about XD. but yeah thank you.