Skip to main content

Client Libraries

Updated Nov 20, 2022 ·

Overview​

Prometheus client libraries help instrument applications to expose metrics for collection by Prometheus. They are available in various programming languages.

  • Supports languages like Go, Java, and Python
  • Allows applications to expose custom metrics
  • Enables automatic scraping by Prometheus

Supported and Unofficial Libraries​

prometheus has the official client libraries for the following languages:

  • Go
  • Java/Scala
  • Python
  • Ruby
  • Rust

Unofficial third-party client libraries:

  • Bash
  • C
  • C++
  • C#
  • Common LISP
  • Dart
  • Elixir
  • Erlang
  • Haskell
  • Lua
  • Node.js
  • OCaml
  • Perl
  • PHP
  • R

You can also write your own client libraries:

  • Unsupported language
  • Avoide unnecessary dependencies

What to Instrument​

Instrumenting your system depends on the type of service and its requirements.

  • Online-serving systems

    • Immediate response expected
    • Examples: Database and web servers
    • Track:
      • Number of queries/requests
      • Number of errors
      • Latency
      • Number of in-progress requests

  • Offline processing services

    • No active waiting for response
    • Continuous running with multiple pipeline stages
    • Track:
      • Amount of queued work
      • Amount of work-in-progress
      • Rate of processing
      • Errors

  • Batch Jobs

    • Run on a schedule, similar to offline services
    • Requires a pushGateway
    • Track:
      • Time processing each stage of the job
      • Overall runtime
      • Last time job was completed

Naming Metrics​

When using client libraries or creating custom metrics, it's essential to follow a proper naming format. The standard format is:

library_name_unit_suffix

Here are a few naming tips:

  • Metric names should be all lowercase, with words separated by underscores ("_").

    Correct:

    http_requests_total

    Incorrect:

    HTTPRequestsTotal

  • The library should represent the application or library that the metric is used for.

    Correct:

    node_memory_usage

    Incorrect:

    webserver_memory_usage

  • The name should describe the metric’s purpose, whether one word or multiple.

    Correct (one word):

    request_latency

    Correct (multiple words):

    request_processing_duration

    Incorrect:

    node_restarts

  • The unit should always be included to clarify the metric’s measurement.

    Correct:

    response_time_seconds

    Incorrect:

    response_time

  • Use unprefixed, standard units like "seconds," "bytes," and "meters."

    Correct:

    data_received_bytes

    Incorrect:

    data_received_kilobytes

  • Suffixes like _total, _count, and _sum are best avoided unless referring to specific metric types like counters.

    Correct:

    request_duration_seconds_total

    Incorrect:

    request_duration_total