Allow sampling

[y2kappa]

I need to run this in production because I simply cannot profile locally. If my server has many hits this takes a big toll on my log quota so I'd like to sample when trace information is dumped.

I think the user should be able to provide a sampling rate, like 500, it has 1/500 chance of being recorded.

I think this should be done in the new constructor:

pub fn new(id: &str, processor: Option<fn(&str)>, stats: Option<TracingStats>) -> Trace {
        let trace = Trace {
            id: id.into(),
            processor: processor.unwrap_or(|x: &str| {
                println!("{}", x);
            }),
            stats: stats.unwrap_or(TracingStats::None),
        };
        Self::register(&trace.id);
        trace
}

Something like

if rand() == self.sampling_rate {
    Self::register(&trace.id);
}

[yonip23]

Why not calculating statistics on each function - min, max, and avg. ?
I don't like the idea of using random in such a deterministic crate...

[y2kappa]

Hey, thanks for the response. Let me clarify.

If for example you run this on a production task, you may not want to dump these stats on every invocation of your service/endpoint and as such you could do it every 100 requests, especially if your service gets many hits. However, once at the beginning of the request when register is called for the first time, and if the random condition passes, then you will be profiling the entire execution path.

This helps with logging for example, if you have a limited log quota. And if you want to calculate statistics on your request, even if it's sampled, you might get a good approximation due to law of large numbers. In my use case I want it every 100 requests and then aggregate them all using some log queries.

However, it's an opt-in feature and it would definitely not default to true.

I hope you're not confusing this with the issue of calculating stats for the collected data.

If you have a different idea about this, let me know, my use case may not be so widespread.