Sampling

Sampling is a process that restricts the amount of traces that are generated by a system. The Erlang SDK offers several head samplers.

Default behavior

By default, all spans are sampled, and thus, 100% of traces are sampled. If you do not need to manage data volume, you do not need to configure a sampler.

ParentBasedSampler

When sampling, the ParentBasedSampler is most often used for head sampling. It uses the sampling decision of the Span’s parent, or the fact that there is no parent, to know which secondary sampler to use.

The sampler can be configured with the environment variables OTEL_TRACES_SAMPLER and OTEL_TRACES_SAMPLER_ARG or using the Application configuration allows you to configure each of the 5 potential states of a Span’s parent:

  • root - No parent
  • remote_parent_sampled - Parent is from a remote Span that is sampled
  • remote_parent_not_sampled - Parent is from a remote Span that is not sampled
  • local_parent_sampled - Parent is from a local Span that is sampled
  • local_parent_not_sampled - Parent is from a local Span that is not sampled

TraceIdRatioBasedSampler

Within the ParentBasedSampler the most common is the TraceIdRatioBasedSampler. It deterministically samples a percentage of traces that you pass in as a parameter.

Environment Variables

You can configure the TraceIdRatioBasedSampler with environment variables:

export OTEL_TRACES_SAMPLER="parentbased_traceidratio"
export OTEL_TRACES_SAMPLER_ARG="0.1"

This tells the SDK to sample spans such that only 10% of Traces get created.

Application configuration

Example in the Application configuration with a root sampler for sampling 10% of Traces and using the parent decision in the other cases:

%% config/sys.config.src
{opentelemetry, {sampler, {parent_based, #{root => {trace_id_ratio_based, 0.10},
                                          remote_parent_sampled => always_on,
                                          remote_parent_not_sampled => always_off,
                                          local_parent_sampled => always_on,
                                          local_parent_not_sampled => always_off}}}}
# config/runtime.exs
config :opentelemetry, sampler: {:parent_based, %{root: {:trace_id_ratio_based, 0.10},
                                                  remote_parent_sampled: :always_on,
                                                  remote_parent_not_sampled: :always_off,
                                                  local_parent_sampled: :always_on,
                                                  local_parent_not_sampled: :always_off}}

AlwaysOn and AlwaysOff Sampler

The other two built-in samplers are the AlwaysOnSampler and the AlwaysOffSampler.

Environment Variables

You can configure the ParentBasedSampler to use either the AlwaysOnSampler or AlwaysOffSampler with the environment variable OTEL_TRACES_SAMPLER:

export OTEL_TRACES_SAMPLER="parentbased_always_on"

And for the AlwaysOffSampler:

export OTEL_TRACES_SAMPLER="parentbased_always_off"

Application configuration

Here’s an example in the Application configuration with a root sampler that always samples and using the parent decision in the other cases:

%% config/sys.config.src
{opentelemetry, {sampler, {parent_based, #{root => always_on,
                                          remote_parent_sampled => always_on,
                                          remote_parent_not_sampled => always_off,
                                          local_parent_sampled => always_on,
                                          local_parent_not_sampled => always_off}}}}
# config/runtime.exs
config :opentelemetry, sampler: {:parent_based, %{root: :always_on,
                                                  remote_parent_sampled: :always_on,
                                                  remote_parent_not_sampled: :always_off,
                                                  local_parent_sampled: :always_on,
                                                  local_parent_not_sampled: :always_off}}

Custom Sampler

Custom samplers can be created by implementing the otel_sampler behaviour. This example sampler:

-module(attribute_sampler).

-behavior(otel_sampler).

-export([description/1,
         setup/1,
         should_sample/7]).

-include("otel_sampler.hrl").

setup(Attributes) when is_map(Attributes) ->
    Attributes;
setup(_) ->
    #{}.

description(_) ->
    <<"AttributeSampler">>.

should_sample(_Ctx, _TraceId, _Links, _SpanName, _SpanKind, Attributes, ConfigAttributes) ->
    AttributesSet = sets:from_list(maps:to_list(Attributes)),
    ConfigSet = sets:from_list(maps:to_list(ConfigAttributes)),
    case sets:is_disjoint(AttributesSet, ConfigSet) of
        true -> {?RECORD_AND_SAMPLE, [], []};
        _ -> {?DROP, [], []}
end.
defmodule AttributesSampler do
  def setup(attributes) when is_map(attributes) do
    attributes
  end

  def setup(_) do
    %{}
  end

  def description(_) do
    "ExampleSampler"
  end

  def should_sample(_ctx, _trace_id, _links, _span_name, _span_kind, attributes, config_attributes) do
    no_match =
      Enum.into(attributes, %MapSet{})
      |> MapSet.disjoint?(Enum.into(config_attributes, %MapSet{}))

    if no_match, do: {:record_and_sample, [], []}, else: {:drop, [], []}
  end
end

Will sample Spans that do not have any attributes that match the attributes passed as the sampler’s configuration.

Example configuration to not sample any Span with an attribute specifying the URL requested is /healthcheck:

{opentelemetry, {sampler, {attributes_sampler, #{'http.target' => <<"/healthcheck">>}}}}
config :opentelemetry, sampler: {AttributesSampler, %{"http.target": "/healthcheck"}}