Sum metric is integer only

[narrowtux]

I had the need to export a duration as sum(), and didn't need distribution(), since I'm only interested in the duration as a total in relation to some other metric I export as counter().

However, when the event is executed, it results in a crash because it runs :ets.update_counter with a float as the Incr argument which results in a badarg.

This is the actual error:

13:31:35.068 [error] Handler {TelemetryMetricsPrometheus.Core.EventHandler, #PID<0.945.0>,
 [:platform, :parser, :parse_all, :seconds]} has failed and has been detached. Class=:error
Reason=:badarg
Stacktrace=[
  {:ets, :update_counter,
   [
     :prometheus_metrics,
     {[:platform, :parser, :parse_all, :seconds],
      %{parser_id: "be056e51-250e-486b-b71c-271a381e18de"}},
     2.3343e-5,
     {{[:platform, :parser, :parse_all, :seconds],
       %{parser_id: "be056e51-250e-486b-b71c-271a381e18de"}}, 0}
   ], [error_info: %{module: :erl_stdlib_errors}]},
  {TelemetryMetricsPrometheus.Core.Sum, :handle_event, 4,
   [file: 'lib/core/sum.ex', line: 62]},
  {:telemetry, :"-execute/3-fun-0-", 4,
   [
     file: '/Users/tux/projects/zenner/platform-base/platform/deps/telemetry/src/telemetry.erl',
     line: 135
   ]},
  {:lists, :foreach_1, 2, [file: 'lists.erl', line: 1442]},
  {:telemetry, :span, 3,
   [
     file: '/Users/tux/projects/zenner/platform-base/platform/deps/telemetry/src/telemetry.erl',
     line: 274
   ]},
  {Platform.Parsing.Stage, :"-parse/3-fun-16-", 4,
   [file: 'lib/platform/parsing/parse_stage.ex', line: 369]},
  {Enum, :"-map/2-fun-0-", 3, [file: 'lib/enum.ex', line: 1662]},
  {:maps, :fold_1, 3, [file: 'maps.erl', line: 411]},
  {Enum, :map, 2, [file: 'lib/enum.ex', line: 2480]},
  {Platform.Parsing.Stage, :handle_events, 3,
   [file: 'lib/platform/parsing/parse_stage.ex', line: 57]},
  {GenStage, :consumer_dispatch, 6, [file: 'lib/gen_stage.ex', line: 2471]},
  {GenStage, :take_pc_events, 3, [file: 'lib/gen_stage.ex', line: 2660]},
  {:gen_server, :try_dispatch, 4, [file: 'gen_server.erl', line: 1120]},
  {:gen_server, :handle_msg, 6, [file: 'gen_server.erl', line: 1197]},
  {:proc_lib, :init_p_do_apply, 3, [file: 'proc_lib.erl', line: 240]}
]

As I said, I've worked around this issue by using distribution() with buckets like this: [0], but this exports 3 extra metrics per label (of which I expect about 1000 in production), which I'm not really interested in.

[josevalim]

I think this is an issue with Prometheus metrics, no? This lib does not consume the metrics, only declares them.

[narrowtux]

You are correct. I will report it there.