我决定使用一个额外的中间件来解决这个问题，我将它添加到应用程序的中间件堆栈中，正好在ActionDispatch::RemoteIp之前，它为Rails应用程序提取remote_ip。
中间件只在检查应用程序实际上在代理之后运行之后才操作X-Forwarded-For头，我可以通过特定头的存在来检查。我只交换了最后两个条目，以使所有IP地址仍然可用，以便进行调试，并且现在检查默认行为太多。

# in config/initializers/proxy_remote_ip_fix.rb
class ProxyRemoteIpFix
  # There is a special header only set when running behind the proxy. Use this 
  # header to decide when to enable the `X-Forwared-For` rewrite.
  PROXY_IDENTIFING_HEADER = 'HTTP_X_PROXY_WAS_HERE'
  X_FORWARDED_FOR_HEADER  = 'HTTP_X_FORWARDED_FOR'

  def initialize(app)
    @app = app
  end

  def call(env)
    manipulate_x_forwarded_for_header(env) if behind_proxy?(env)

    @app.call(env)
  end

  private

  # We can assume that the last entry in the list is the IP of the proxy
  # and the second last entry has a high chance to be the client's
  # real IP address.
  #
  # To trick Ruby on Rails to pick the second last IP instead of the last when
  # calling `remote_ip` we simply swap their positions in the list. We do not
  # remove the proxy instance's IP completely to still keep its IP in the header
  # for debugging reasons.
  def manipulate_x_forwarded_for_header(env)
    ips = ips_from_forwarded_for_header(env)
    return unless ips.size > 1

    # swap the last two entries
    ips[-1], ips[-2] = ips[-2], ips[-1]

    env[X_FORWARDED_FOR_HEADER] = ips.join(', ')
  end

  def ips_from_forwarded_for_header(env)
    return [] unless env.key?(X_FORWARDED_FOR_HEADER)

    env[X_FORWARDED_FOR_HEADER].split(',').map(&:strip)
  end

  def behind_proxy?(env)
    env.key?(PROXY_IDENTIFING_HEADER)
  end
end

# ProxyRemoteIpFix must be added before the ActionDispatch::RemoteIp,
# because its behavior changes the way how the ActionDispatch::RemoteIp works.
Rails.application.config.middleware.insert_before(
  ActionDispatch::RemoteIp, ProxyRemoteIpFix
)