how to import tracing_subscriber?

[evbo]

all the docs show this variable but do not show where it's imported from. Can it be clarified?

[evbo]

it appears to be a separate dependency:
https://github.com/awslabs/aws-lambda-rust-runtime/blob/main/examples/http-basic-lambda/Cargo.toml#L18

[greenwoodcm]

yeah it's referenced directly from the crate name. does that clarify?

[timClicks]

@evbo, when you say "all the docs", which is the document that you encountered this first? It may be worth adding some context there.

[evbo]

I actually have realized today that "_" characters search very poorly in github, but if you search "subscriber" or "tracing" you can see in each readme example where it uses it without any explanation of what it's for or where it comes from.

I'm actually not totally sure how errors are supposed to be logged? Aside from println!, nothing seems to end up in cloudwatch.

Maybe this isn't a bad thing though, as I should be handling these errors in a state machine and properly messaging them...

[evbo]

@timClicks do I need to use tracing to get errors logged to cloudwatch? Or can I just set the log_level env variable? The docs really leave you hanging:
https://github.com/awslabs/aws-lambda-rust-runtime?tab=readme-ov-file#tracing-and-logging

[calavera]

If you want anything written in CloudWatch, you have to print it somehow. If you use the tracing crate, the messages will be written depending on the log level, and following the log level rules that tracing uses. The default log level is INFO, so it you use tracing::error!, messages will be written to CloudWatch.

[evbo]

Thanks @calavera that's actually gold nuggets for the docs. Currently AWS docs look like you need to use instrument, but based on what you're saying at a minimum we just need to set LOG_LEVEL env var to our liking and then do something like:

tracing_subscriber::fmt()
        .json()
        .with_max_level(tracing::Level::ERROR)
        .with_current_span(false)
        .with_ansi(false)
        .without_time()
        .with_target(false)
        .init();

With that any error should go to cloudwatch, right?