Proper way to handle database connections ?

[kamek-pf]

Hey there,
So, I realize this might fall outside the scope of this crate, but Rust + Lambda isn't exactly a common thing yet, anyways :

From the AWS Lamba Python doc :

We recommend that pymysql.connect() is executed outside the [lambda] handler, as shown, for better performance.

My question is, is there any way to do this in Rust, and if not, what would you recommend ?

[iliana]

Hmm.

I'm pretty sure the current crowbar interface doesn't have a way of doing this at all, to directly answer your question. I suppose it's also very difficult to create a database connection in Python and hand it off to a Rust module, so perhaps that's not worth investigating either.

I'd like to see if there's a way to run code on import of a rust-cpython module. Then I'd need to make the return value of that code accessible to the handler.

[kamek-pf]

After further investigation, the lazy_static crate solves this problem nicely !

lazy_static! {
    static ref CONN: Mutex<Connection> = Mutex::new(Connection::connect("postgres:/your_connection_string", TlsMode::None).unwrap());
} 

lambda!(|_, _| {
    let conn = CONN.lock().unwrap();
    your_service(&conn)
});

Lambda seems to effectively re-use existing processes and keep connections open.
When opening connections directly in the lambda! closure, I'm seeing 20-30ms execution times, but with this, it drops to 2-3ms.

An equivalent (warm) function in Node is also in the 2-3ms range, so this is very good.
Cold starts are also way more efficient than with Node : I'm getting random cold starts between 200-600ms in Node, but nothing above 100ms in Rust yet.

Great job on this crate @ilianaw :)

[iliana]

Nice!

(Yeah, my understanding is that the Python interpreter stays running, with your code imported, until Lambda kills that process for inactivity.)