tracing-prometheus

This crate is experimental and not intended for use in production

This crate is an experiment for seamlessly capturing metrics data using the tracing family of macros. The goal is to enable something like the following:

let batch_size = 50_u64,
tracing::trace!(
    // Metric names and values encoded as `metrics.<type>.<name> = <value>`
    metrics.histogram.batch_size = batch_size,

    // Metric fields encoded as `<name>.<field>`
    metrics.histogram.batch_size.help = "Message count in a Kafka batch",

    // Key-value labels encoded as .label.<label_key> = "<label-value>"
    metrics.histogram.batch_size.label.topic = "my-topic",

    // Goal: Use boolean to decide whether to record exemplar
    metrics.histogram.batch_size.exemplar = batch_size > 100,

    // Emit standard traces/logs from the same call
    "Consumed batch from Kafka",
);

Here's a quick list of ideas/goals/non-goals that I have about the future of this crate:

• The metrics DSL/API should be prometheus-first, rather than trying to be fully general to any metrics system. Hence the name tracing-prometheus

  • Currently I'm using the official prometheus-client crate, since it's the only one I know of that natively supports exemplars, which is a goal of this crate.

• Experimental goal: It should be possible to use a single tracing call to instrument multiple metrics as well as emit a trace with arbitrary fields like normal.

• Experimental goal: It should be possible to automatically capture exemplars between metrics and traces. In this library, since metrics are captured with the same instrumentation as traces, it should be natural to attach tracing context to metrics samples.

• Goal: It should be possible to integrate with the opentelemetry ecosystem to automatically associate metrics recordings with distributed traces.

Related projects / Prior art

• tracing-opentelemetry has a MetricsLayer that allows similar metrics instrumentation