#metrics #telemetry #prometheus

metrics-exporter-prometheus

A metrics-compatible exporter for sending metrics to Prometheus

19 releases (7 breaking)

new 0.8.0 Jan 15, 2022
0.7.0 Dec 16, 2021
0.6.1 Sep 16, 2021
0.6.0 Jul 15, 2021
0.1.0-alpha.3 Jul 22, 2020

#67 in Debugging

Download history 6544/week @ 2021-09-29 6715/week @ 2021-10-06 7397/week @ 2021-10-13 8529/week @ 2021-10-20 8419/week @ 2021-10-27 7031/week @ 2021-11-03 6848/week @ 2021-11-10 9027/week @ 2021-11-17 7377/week @ 2021-11-24 7925/week @ 2021-12-01 8646/week @ 2021-12-08 8307/week @ 2021-12-15 4881/week @ 2021-12-22 6242/week @ 2021-12-29 8455/week @ 2022-01-05 7783/week @ 2022-01-12

28,668 downloads per month
Used in 8 crates (6 directly)

MIT license

300KB
5K SLoC

metrics-exporter-prometheus

conduct-badge downloads-badge release-badge docs-badge license-badge

metrics-exporter-prometheus is a metrics-compatible exporter that serves a Prometheus scrape endpoint.

code of conduct

NOTE: All conversations and contributions to this project shall adhere to the Code of Conduct.


lib.rs:

A [metrics]-compatible exporter for sending metrics to Prometheus.

Basics

metrics-exporter-prometheus is a [metrics]-compatible exporter for either exposing an HTTP endpoint that can be scraped by Prometheus, or that can push metrics to a Prometheus push gateway.

High-level features

  • scrape endpoint support
  • push gateway support
  • IP-based allowlist for scrape endpoint
  • ability to push histograms as either aggregated summaries or aggregated histograms, with configurable quantiles/buckets
  • ability to control bucket configuration on a per-metric basis
  • configurable global labels (applied to all metrics, overridden by metric's own labels if present)

Behavior

In general, interacting with the exporter should look and feel like interacting with any other implementation of a Prometheus scrape endpoint or push gateway implementation, but there are some small caveats around metric naming.

We strive to match both the Prometheus data model and follow the exposition format specification, but due to the decoupled nature of metrics, the exporter makes some specific trade-offs when ensuring compliance with the specification when it comes to metric names and label keys. Below is a matrix of scenarios where the exporter will modify a metric name or label key:

  • metric name starts with, or contains, an invalid character: replace character with underscore
  • label key starts with, or contains, an invalid character: replace character with underscore
  • label key starts with two underscores: add additional underscore (three underscores total)

This behavior may be confusing at first since metrics itself allows any valid UTF-8 string for a metric name or label, but there is no way to report to the user that a metric name or label key is invalid only when using the Prometheus exporter, so we must cope with these situations by replacing invalid characters at runtime.

Usage

Using the exporter is straightforward:

// First, create a builder.
//
// The builder can configure many aspects of the exporter, such as changing the
// listen address, adjusting how histograms will be reported, changing how long
// metrics can be idle before being removed, and more.
let builder = PrometheusBuilder::new();

// Normally, most users will want to "install" the exporter which sets it as the
// global recorder for all `metrics` calls, and installs either an HTTP listener
// when running as a scrape endpoint, or a simple asynchronous task which pushes
// to the configured push gateway on the given interval.
//
// If you're already inside a Tokio runtime, this will spawn a task for the
// exporter on that runtime, and otherwise, a new background thread will be
// spawned which a Tokio single-threaded runtime is launched on to, where we then
// finally launch the exporter:
builder.install().expect("failed to install recorder/exporter");

// Maybe you already have an HTTP endpoint that you want to expose a metrics
// endpoint on.. no problem!  You can build the recorder and install it, and get
// back a handle that can be used to generate the Prometheus scrape output on
// demand:
let handle = builder.install_recorder().expect("failed to install recorder");

// Maybe you have a more complicated setup and want to be handed back the recorder
// object and a future that can run the HTTP listener / push gateway so you can
// install/spawn them in a specific way.. also not a problem!
//
// As this is a more advanced method, it _must_ be called from within an existing
// Tokio runtime when the exporter is running in HTTP listener/scrape endpoint mode.
let (recorder, exporter) = builder.build().expect("failed to build recorder/exporter");

// Finally, maybe you literally only want to build the recorder and nothing else,
// and we've got you covered there, too:
let recorder = builder.build_recorder().expect("failed to build recorder");

Features

Two main feature flags control which modes that exporter can run in:

  • http-listener: allows running the exporter as a scrape endpoint (enabled by default)
  • push-gateway: allows running the exporter in push gateway mode (enabled by default)

Neither of these flags are required to create, or install, only a recorder. However, in order to create or build an exporter, at least one of these feature flags must be enabled. Builder methods that require certain feature flags will be documented as such.

Dependencies

~1.7–5MB
~92K SLoC