Measuring time intervals is trivial - you just have to be sure you are using monotonic time source. Basically interval is a difference between start time and end time. Erlang has standard erlang:monotonic_time function that returns so called native time units. Native time units are meaningless and have to be converted to seconds (or other units) using erlang:convert_time_unit. However as erlang:convert_time_unit documentation warns:

You may lose accuracy and precision when converting between  time units.
In order to minimize such loss, collect all data at native time unit and
do the conversion on the end result.

and because Prometheus mandates support for floats, set_duration/observe_duration functions always work with native time units and conversion is delayed until scraping/retrieving value. To implement this, metric needs to know desired time unit. Users can specify time unit explicitly via duration_unit or implicitly via metric name (preferred, since prometheus best practices guide insists on <name>_duration_<unit> metric name format).

Possible units:

  • :microseconds;
  • :milliseconds;
  • :seconds;
  • :minutes;
  • :hours;
  • :days.

Histogram also converts buckets bounds to native units if duration_unit is provided. It converts it back when scraping or retrieving value.

If values already converted to a "real" unit, conversion can be disabled by setting :duration_unit to false.


Example where duration unit derived from name:[name: :fun_duration_seconds,
               buckets: [0.5, 1.1], # in seconds
               help: ""])

Histogram.observe_duration(:fun_duration_seconds, do: Process.sleep(1000))

{[0, 1, 0], 1.001039204}

Example where duration unit set explicitly:[name: :fun_duration_histogram,
               buckets: [500, 1100], # in milliseconds
               duration_unit: :milliseconds,
               help: ""])

Histogram.observe_duration(:fun_duration_histogram, do: Process.sleep(1000))

{[0, 1, 0], 1000.714918}

Example where value is in seconds already:[name: :duration_seconds,
               buckets: [0.5, 1.1], # in seconds
               duration_unit: false,
               help: ""])

Histogram.dobserve(:duration_seconds, 1.2)

{[0, 0, 1], 1.2}