travis-ci build status CRAN status

openmetrics is an opinionated client for Prometheus and the related OpenMetrics project. It makes it possible to add predefined and custom metrics to any R web application and expose them on a /metrics endpoint, where they can be consumed by Prometheus services.

The package includes built-in support for Plumber and Shiny applications, but is highly extensible.


You can install openmetrics from CRAN with


or from GitHub with

# install.packages("remotes")

Use with Plumber

You can easily wrap an existing Plumber router object to add metrics for HTTP requests and their duration:

srv <- plumber::plumb("plumber.R")
srv <- register_plumber_metrics(srv)

This will automatically create a /metrics endpoint that exposes these metrics (and any others you have defined).


To add authentication to this endpoint, set the METRICS_HTTP_AUTHORIZATION environment variable to the expected Authorization header you want Prometheus to use. For example, to grant access to the username “aladdin” with password “opensesame”:

Sys.setenv(METRICS_HTTP_AUTHORIZATION = "Basic YWxhZGRpbjpvcGVuc2VzYW1l")

Meanwhile, in your Prometheus configuration:

# ...
- job_name: my-plumber-api
    username: aladdin
    password: opensesame
  # ...

Use with Shiny

You can also wrap an existing Shiny app object to add metrics on reactive flush duration and a running session count. In your app.R file, you can use something like the following:

app <- shiny::shinyApp(...)
app <- register_shiny_metrics(app)

Which will be picked up by shiny::runApp().

Again, this will automatically create a /metrics endpoint that exposes these metrics (and any others you have defined). It supports the same authentication method as well.

This feature should be considered experimental. It relies on certain unstable Shiny internals to add the /metrics endpoint, which is not usually possible.

Default Process Metrics

You can enable the default process metrics (which track CPU time, memory usage, and open files) with


Note that not all metrics are supported (or even meaningful) on all operating systems.

Custom Metrics

All required metrics and features of the informal Prometheus client specification are supported, so you can create app-specific counters, gauges, and histograms. If you want to make use of labels, you must pass a default value at the time of creation.

Here are some examples collected from around the house:

meows <- counter_metric("meows", "Heard around the house.", labels = "cat")
meows$inc(cat = "Shamus") # Count one meow from Shamus.
meows$inc(3, cat = "Unknown") # Count three meows of unknown origin.

thermostat <- gauge_metric("thermostat", "Thermostat display.")
thermostat$set(21.3) # Read from the display...
thermostat$dec(2) # ... and then turn it down 2 degrees.

temperature <- histogram_metric(
  "temperature", "Ambient room temperature measurements.",
  buckets = c(10, 15, 20, 22, 25), room = "kitchen"
# Simulate taking ambient temperature samples.
for (measure in rnorm(20, mean = 21.5)) {
  temperature$observe(measure, room = sample(c("kitchen", "bathroom"), 1))

All metrics (in fact all exported functions) take a registry parameter in case you want to avoid using the default global registry in some part of your application. You can construct new registries with registry().


Because metrics may be gathered with high frequency, some effort has been made to ensure that using them is fast – usually around 10 microseconds.

Rendering Metrics Manually

The render_metrics() function exposes the text-based format understood by Prometheus. Here’s what the metrics above look like:

# HELP meows Heard around the house.
# TYPE meows counter
meows_total{cat="Shamus"} 1
meows_created{cat="Shamus"} 1604597246.05814
meows_total{cat="Unknown"} 3
meows_created{cat="Unknown"} 1604597246.05893
# HELP thermostat Thermostat display.
# TYPE thermostat gauge
thermostat 19.3
# HELP temperature Ambient room temperature measurements.
# TYPE temperature histogram
temperature_bucket{room="bathroom",le="10.0"} 0
temperature_bucket{room="bathroom",le="15.0"} 0
temperature_bucket{room="bathroom",le="20.0"} 0
temperature_bucket{room="bathroom",le="22.0"} 9
temperature_bucket{room="bathroom",le="25.0"} 11
temperature_bucket{room="bathroom",le="+Inf"} 11
temperature_sum{room="bathroom"} 234.387663039796
temperature_count{room="bathroom"} 11
temperature_created{room="bathroom"} 1604597246.08967
temperature_bucket{room="kitchen",le="10.0"} 0
temperature_bucket{room="kitchen",le="15.0"} 0
temperature_bucket{room="kitchen",le="20.0"} 1
temperature_bucket{room="kitchen",le="22.0"} 4
temperature_bucket{room="kitchen",le="25.0"} 9
temperature_bucket{room="kitchen",le="+Inf"} 9
temperature_sum{room="kitchen"} 198.854388891071
temperature_count{room="kitchen"} 9
temperature_created{room="kitchen"} 1604597246.08957

You can use this to implement a /metrics endpoint in your application if it is not supported directly. For example, using a raw httpuv server:

  "", 8080,
  list(call = function(req) {
      status = 200L,
      headers = list("Content-Type" = "text/plain; version=0.0.4"),
      body = render_metrics()

Pushgateway Support

Some workloads may not want to run an HTTP server to expose metrics, especially in the case of short-lived batch jobs. For these cases metrics can also be manually “pushed” to a Prometheus Pushgateway instance, though there are drawbacks to this approach.

push_to_gateway() can be used to push metrics, and delete_from_gateway() can be used to clean them up when the workload is finished:

push_to_gateway("localhost:9091", job = "openmetrics-readme")
# Some time later...
delete_from_gateway("localhost:9091", job = "openmetrics-readme")



The package is made available under the terms of the MIT license. See LICENSE for details.