Logging error chains to CloudWatch

[mlafeldt]

I noticed that the Rust runtime currently logs handler errors using {} as the format:

aws-lambda-rust-runtime/lambda-runtime/src/lib.rs

Line 178 in 3871cb4

error!("{}", err); // logs the error in CloudWatch

However, the widely used anyhow crate will only include the error chain with {:?} or {:#}:

https://github.com/dtolnay/anyhow/blob/71caf88b78cc3faf2b99ac3d7397a66a7da5a930/src/fmt.rs#L7-L40

Among other things, this means that only the last error will be logged to CloudWatch and that all underlying causes are lost when using with_context to augment errors.

My current solution/workaround is to add an additional log statement like this:

#[tokio::main]
async fn main() -> Result<(), Error> {
    env_logger::try_init()?;

    lambda_runtime::run(handler_fn(|_: Input, _: Context| async {
        let output = handler().await.map_err(|e| {
            error!("{:?}", e); // log error chain to CloudWatch
            e
        })?;
        Ok(output) as Result<Output>
    }))
    .await?;

    Ok(())
}

Of course, this duplicates error logging to some degree and would have to be done in every Lambda function. I wonder if there's a better way of telling the runtime how to log errors?

[mlafeldt]

Related to this, the error returned to the caller also misses any context:

aws-lambda-rust-runtime/lambda-runtime/src/lib.rs

Line 183 in 3871cb4

error_message: format!("{}", err), // returns the error to the caller via Lambda API

[calavera]

@mlafeldt do you want to open a PR and change this? I think we'd be open to that improvement.

[mlafeldt]

@calavera Absolutely. See #413