Correct way to handle async IO inside register_callback within Tokio runtime

[0xForerunner]

I would like to send http requests during the callback. Should I use spawn_blocking for the entire pipeline? Or is it okay to having blocking io within the callback?

[jtescher]

you can block for a reasonable amount of time within the callback, we can consider making the callback accept an async closure too.

[0xForerunner]

Okay cool good to know! I think something like register_callback_async would be pretty handy I'm sure!

[0xForerunner]

So after some investigation I have found that to not be true. This code sadly causes a deadlock.

    meter.register_callback(&anys, move |observer| {
        println!("In callback");
        block_on(metrics_callback(observer)).log();
    })?;

Not really sure what the solution is.

[0xForerunner]

Although now that I think of it this might be happening because I am accessing a static RwLock inside that function and it doesn't like block_on.... hmmm

[0xForerunner]

So it's not from the lazy static or the RwLock. it seems something from

client.query().await

is causing the deadlock. Not sure why this would be.

[GabrielBianconi]

For future reference:

Here's a hacky solution I found for this issue:

#1376 (comment)