Some final feedbacks:

Tried using the “catch all” from sentry, but it catch way too much thing, (if a render fail, I have multiple Logger.error that occurs (one from phoenix’s GenServer, the one from raise, etc…).

The sentry became polluted very fast on any Logger.error.

My current solution: write my own little Logger backend :

defmodule MyApp.SentryBackend do
  @behaviour :gen_event

  def init(__MODULE__), do: {:ok, nil}

  def handle_call({:configure, _options}, state), do: {:ok, :ok, state}

  # only Logger errors with crash_reason
  def handle_event({level, _gl, {Logger, _msg, _ts, meta}}, state) when level in [:emergency, :alert, :critical, :error] do
    if meta[:crash_reason] do
      {reason, stacktrace} = meta[:crash_reason]
      Sentry.capture_message(inspect(reason), stacktrace: stacktrace)
    end

    {:ok, state}
  end

  def handle_event(_, state), do: {:ok, state}

  def terminate(_reason, _state), do: :ok
end

I allows my to further filter if I need to.

cheers !