defmodule Demo do
  def merge_values(_key, value1, value2),
    do: value1 + value2

  def reducer({key, data}, aggregate) do
    merge_data = fn all_data -> Map.merge(all_data, data, &merge_values/3) end
    Map.update(aggregate, key, data, merge_data)
  end

  def run(list),
    do: List.foldl(list, %{}, &reducer/2)
end

list = [
  {:group_1, %{timeout: 1, failure: 0, hits: 0}},
  {:group_1, %{timeout: 0, failure: 0, hits: 0}},
  {:group_2, %{timeout: 1, failure: 1, hits: 0}}
]

IO.inspect(Demo.run(list))

$ elixir demo.exs
%{
  group_1: %{failure: 0, hits: 0, timeout: 1},
  group_2: %{failure: 1, hits: 0, timeout: 1}
}

One thing to consider might be using an ets table instead of a server with state. In general, if you need high performance storage, ets tables are a very good solution. You could still have a process periodically reading from the table and doing some statistics.

There is Erlang’s queue data type.