feat(http): back off from hosts that answer 429 or 503

Object fetches and WebFinger stop sending requests to a host until
Retry-After or X-RateLimit-Reset (default 5 minutes, max 1 hour).
Ported from Akkoma's Pleroma.HTTP.Backoff.

Co-authored-by: FloatingGhost <hannah@coffee-and-dreams.uk>
Co-authored-by: Oneric <oneric@oneric.stub>

changelog.d/http-backoff.add

@@ -0,0 +1 @@
+Back off from hosts that answer 429 or 503 to object fetches and WebFinger, until Retry-After or X-RateLimit-Reset (default 5 minutes, max 1 hour)

lib/mix/pleroma.ex

@@ -17,7 +17,7 @@ defmodule Mix.Pleroma do
     :oban,
     :logger_backends
   ]
-  @cachex_children ["object", "user", "scrubber", "web_resp"]
+  @cachex_children ["object", "user", "scrubber", "web_resp", "http_backoff"]
   @doc "Common functions to be reused in mix tasks"
   def start_pleroma do
     Pleroma.Config.Holder.save_default()

lib/pleroma/application.ex

@@ -198,6 +198,7 @@ defmodule Pleroma.Application do
       ),
       build_cachex("rel_me", limit: 2500),
       build_cachex("host_meta", default_ttl: :timer.minutes(120), limit: 5_000),
+      build_cachex("http_backoff", limit: 10_000),
       build_cachex("translations", default_ttl: :timer.hours(24), limit: 5_000)
     ]
   end

lib/pleroma/http/backoff.ex

@@ -0,0 +1,101 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2022 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+defmodule Pleroma.HTTP.Backoff do
+  @moduledoc """
+  GET wrapper that stops sending requests to a host after it answers
+  429 or 503, until the time given by `Retry-After` or `X-RateLimit-Reset`.
+
+  Based on Akkoma's `Pleroma.HTTP.Backoff`.
+  """
+
+  alias Pleroma.HTTP
+
+  require Logger
+
+  @cachex Pleroma.Config.get([:cachex, :provider], Cachex)
+  @cache :http_backoff_cache
+
+  @default_seconds 5 * 60
+  @max_seconds 60 * 60
+
+  @spec get(String.t(), list(), keyword()) :: {:ok, Tesla.Env.t()} | {:error, any()}
+  def get(url, headers \\ [], options \\ []) when is_binary(url) do
+    host = URI.parse(url).host
+
+    case @cachex.get(@cache, host) do
+      {:ok, true} ->
+        {:error, :ratelimit}
+
+      _ ->
+        url
+        |> HTTP.get(headers, options)
+        |> check_backoff(host)
+    end
+  end
+
+  defp check_backoff({:ok, %{status: status, headers: headers}}, host)
+       when status in [429, 503] do
+    seconds = backoff_seconds(headers, DateTime.utc_now())
+    Logger.warning("#{host} answered #{status}, backing off for #{seconds} seconds")
+    @cachex.put(@cache, host, true, ttl: :timer.seconds(seconds))
+    {:error, :ratelimit}
+  end
+
+  defp check_backoff(response, _host), do: response
+
+  @doc """
+  Seconds to back off for, from the first valid `Retry-After` or
+  `X-RateLimit-Reset` header. Missing, invalid or past values give the
+  default of #{@default_seconds} seconds; values are capped at #{@max_seconds}.
+  """
+  @spec backoff_seconds(list(), DateTime.t()) :: pos_integer()
+  def backoff_seconds(headers, now) do
+    [
+      parse_retry_after(header(headers, "retry-after"), now),
+      parse_timestamp(header(headers, "x-ratelimit-reset"), now)
+    ]
+    |> Enum.find(@default_seconds, &is_integer/1)
+    |> min(@max_seconds)
+  end
+
+  defp header(headers, name) do
+    Enum.find_value(headers, fn
+      {key, value} when is_binary(key) and is_binary(value) ->
+        if String.downcase(key) == name, do: String.trim(value)
+
+      _ ->
+        nil
+    end)
+  end
+
+  # Retry-After is either delta-seconds or an HTTP-date (RFC 9110, 10.2.3)
+  defp parse_retry_after(nil, _now), do: nil
+
+  defp parse_retry_after(value, now) do
+    case Integer.parse(value) do
+      {seconds, ""} when seconds > 0 -> seconds
+      {_, ""} -> nil
+      _ -> parse_timestamp(value, now)
+    end
+  end
+
+  defp parse_timestamp(nil, _now), do: nil
+
+  defp parse_timestamp(value, now) do
+    with {:ok, %DateTime{} = datetime} <- parse_datetime(value),
+         seconds when seconds > 0 <- DateTime.diff(datetime, now) do
+      seconds
+    else
+      _ -> nil
+    end
+  end
+
+  defp parse_datetime(value) do
+    with {:error, _} <- Timex.parse(value, "{RFC1123}"),
+         {:ok, datetime, _offset} <- DateTime.from_iso8601(value) do
+      {:ok, datetime}
+    end
+  end
+end

lib/pleroma/object/fetcher.ex

@@ -216,7 +216,7 @@ defmodule Pleroma.Object.Fetcher do
       |> maybe_date_fetch(date)
       |> sign_fetch(id, date)
 
-    case HTTP.get(id, headers) do
+    case HTTP.Backoff.get(id, headers) do
       {:ok, %{body: body, status: code, headers: headers, url: final_url}}
       when code in 200..299 ->
         remote_host = if final_url, do: URI.parse(final_url).host, else: nil

lib/pleroma/web/web_finger.ex

@@ -158,7 +158,11 @@ defmodule Pleroma.Web.WebFinger do
   @cachex Pleroma.Config.get([:cachex, :provider], Cachex)
   def find_lrdd_template(domain) do
     @cachex.fetch!(:host_meta_cache, domain, fn _ ->
-      {:commit, fetch_lrdd_template(domain)}
+      case fetch_lrdd_template(domain) do
+        # retry once the backoff expires instead of caching the miss
+        {:error, :ratelimit} = error -> {:ignore, error}
+        result -> {:commit, result}
+      end
     end)
   rescue
     e -> {:error, "Cachex error: #{inspect(e)}"}
@@ -168,9 +172,13 @@ defmodule Pleroma.Web.WebFinger do
     # WebFinger is restricted to HTTPS - https://tools.ietf.org/html/rfc7033#section-9.1
     meta_url = "https://#{domain}/.well-known/host-meta"
 
-    with {:ok, %{status: status, body: body}} when status in 200..299 <- HTTP.get(meta_url) do
+    with {:ok, %{status: status, body: body}} when status in 200..299 <-
+           HTTP.Backoff.get(meta_url) do
       get_template_from_xml(body)
     else
+      {:error, :ratelimit} = error ->
+        error
+
       error ->
         Logger.warning("Can't find LRDD template in #{inspect(meta_url)}: #{inspect(error)}")
         {:error, :lrdd_not_found}
@@ -210,7 +218,7 @@ defmodule Pleroma.Web.WebFinger do
 
     with address when is_binary(address) <- get_address_from_domain(domain, account),
          {:ok, %{status: status, body: body, headers: headers}} when status in 200..299 <-
-           HTTP.get(
+           HTTP.Backoff.get(
              address,
              [{"accept", "application/xrd+xml,application/jrd+json"}]
            ) do

test/pleroma/http/backoff_test.exs

@@ -0,0 +1,154 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2022 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+# Based on Akkoma's test/pleroma/http/backoff_test.exs
+defmodule Pleroma.HTTP.BackoffTest do
+  use Pleroma.DataCase, async: false
+
+  alias Pleroma.HTTP.Backoff
+
+  @cache :http_backoff_cache
+  @url "https://ratelimited.example.com/users/alice"
+  @host "ratelimited.example.com"
+
+  defp mock_response(status, headers \\ []) do
+    test_pid = self()
+
+    Tesla.Mock.mock(fn %{method: :get} ->
+      send(test_pid, :request)
+      %Tesla.Env{status: status, headers: headers, body: ""}
+    end)
+  end
+
+  defp ttl_seconds do
+    {:ok, ttl} = Cachex.ttl(@cache, @host)
+    div(ttl, 1000)
+  end
+
+  defp assert_ttl_about(expected) do
+    ttl = ttl_seconds()
+    assert ttl in (expected - 15)..expected//1, "expected a TTL near #{expected}, got #{ttl}"
+  end
+
+  describe "get/3" do
+    test "returns the response when the host is not backed off" do
+      mock_response(200)
+
+      assert {:ok, %Tesla.Env{status: 200}} = Backoff.get(@url)
+      assert {:ok, nil} = Cachex.get(@cache, @host)
+    end
+
+    test "does not send a request while the host is backed off" do
+      Cachex.put(@cache, @host, true)
+      mock_response(200)
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      refute_received :request
+    end
+
+    for status <- [429, 503] do
+      test "backs off from the host after a #{status} response" do
+        mock_response(unquote(status))
+
+        assert {:error, :ratelimit} = Backoff.get(@url)
+        assert_received :request
+        assert {:ok, true} = Cachex.get(@cache, @host)
+
+        assert {:error, :ratelimit} = Backoff.get(@url <> "/other")
+        refute_received :request
+      end
+    end
+
+    test "uses a default backoff without headers" do
+      mock_response(429)
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(300)
+    end
+
+    test "only backs off from the host that answered" do
+      mock_response(429)
+      assert {:error, :ratelimit} = Backoff.get(@url)
+
+      mock_response(200)
+      assert {:ok, _} = Backoff.get("https://other.example.com/users/bob")
+    end
+  end
+
+  describe "backoff headers" do
+    test "parses Retry-After in seconds" do
+      mock_response(429, [{"retry-after", "600"}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(600)
+    end
+
+    test "parses Retry-After as an HTTP-date" do
+      date =
+        DateTime.utc_now()
+        |> DateTime.add(600)
+        |> Timex.format!("{WDshort}, {0D} {Mshort} {YYYY} {h24}:{m}:{s} GMT")
+
+      mock_response(503, [{"retry-after", date}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(600)
+    end
+
+    test "parses Retry-After as an ISO 8601 timestamp" do
+      date = DateTime.utc_now() |> DateTime.add(600) |> DateTime.to_iso8601()
+      mock_response(429, [{"retry-after", date}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(600)
+    end
+
+    test "parses X-RateLimit-Reset as an ISO 8601 timestamp" do
+      date = DateTime.utc_now() |> DateTime.add(600) |> DateTime.to_iso8601()
+      mock_response(429, [{"x-ratelimit-reset", date}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(600)
+    end
+
+    test "matches header names case-insensitively" do
+      mock_response(429, [{"Retry-After", "600"}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(600)
+    end
+
+    test "prefers Retry-After over X-RateLimit-Reset" do
+      date = DateTime.utc_now() |> DateTime.add(1200) |> DateTime.to_iso8601()
+      mock_response(429, [{"x-ratelimit-reset", date}, {"retry-after", "600"}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(600)
+    end
+
+    test "falls back to the next header when one is invalid" do
+      date = DateTime.utc_now() |> DateTime.add(600) |> DateTime.to_iso8601()
+      mock_response(429, [{"retry-after", "soon"}, {"x-ratelimit-reset", date}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(600)
+    end
+
+    for value <- ["soon", "-5", "0", "12.5", "", "Wed, 21 Oct 2015 07:28:00 GMT"] do
+      test "ignores the invalid or past Retry-After value #{inspect(value)}" do
+        mock_response(429, [{"retry-after", unquote(value)}])
+
+        assert {:error, :ratelimit} = Backoff.get(@url)
+        assert_ttl_about(300)
+      end
+    end
+
+    test "caps absurd values at one hour" do
+      mock_response(429, [{"retry-after", "99999999999"}])
+
+      assert {:error, :ratelimit} = Backoff.get(@url)
+      assert_ttl_about(3600)
+    end
+  end
+end

test/pleroma/object/fetcher_test.exs

@@ -94,6 +94,27 @@ defmodule Pleroma.Object.FetcherTest do
     end
   end
 
+  describe "rate limiting" do
+    test "a backed-off host gets no further requests" do
+      test_pid = self()
+
+      mock(fn %{method: :get, url: url} ->
+        send(test_pid, {:request, url})
+        %Tesla.Env{status: 429, headers: [{"retry-after", "600"}]}
+      end)
+
+      first = "https://ratelimited.example.com/objects/1"
+      second = "https://ratelimited.example.com/objects/2"
+
+      assert {:fetch, {:error, :ratelimit}} = Fetcher.fetch_object_from_id(first)
+      assert_received {:request, ^first}
+
+      assert {:error, :ratelimit} = Fetcher.fetch_and_contain_remote_object_from_id(second)
+      assert {:fetch, {:error, :ratelimit}} = Fetcher.fetch_object_from_id(second)
+      refute_received {:request, _}
+    end
+  end
+
   describe "max thread distance restriction" do
     @ap_id "http://mastodon.example.org/@admin/99541947525187367"
     setup do: clear_config([:instance, :federation_incoming_replies_max_depth])

test/pleroma/web/web_finger_test.exs

@@ -305,4 +305,29 @@ defmodule Pleroma.Web.WebFingerTest do
       {:ok, _data} = WebFinger.finger("a@sub.mastodon.example")
     end
   end
+
+  describe "rate limiting" do
+    setup do
+      Mox.stub_with(Pleroma.CachexMock, Pleroma.CachexProxy)
+      test_pid = self()
+
+      Tesla.Mock.mock_global(fn %{url: url} ->
+        send(test_pid, {:request, url})
+        %Tesla.Env{status: 429, headers: [{"retry-after", "600"}]}
+      end)
+    end
+
+    test "backs off from a host that rate limits host-meta" do
+      assert {:error, :ratelimit} = WebFinger.finger("alice@ratelimited.example")
+      assert_received {:request, "https://ratelimited.example/.well-known/host-meta"}
+      refute_received {:request, _}
+
+      # a rate-limited host-meta lookup must not be cached as "no LRDD template"
+      assert {:ok, nil} = Cachex.get(:host_meta_cache, "ratelimited.example")
+
+      assert {:error, :ratelimit} = WebFinger.finger("bob@ratelimited.example")
+      assert {:error, :ratelimit} = WebFinger.find_lrdd_template("ratelimited.example")
+      refute_received {:request, _}
+    end
+  end
 end

test/pleroma/workers/remote_fetcher_worker_test.exs

@@ -13,6 +13,7 @@ defmodule Pleroma.Workers.RemoteFetcherWorkerTest do
   @unauthorized_object "https://unauthorized.example.com/"
   @depth_object "https://depth.example.com/"
   @content_type_object "https://bad_content_type.example.com/"
+  @ratelimited_object "https://ratelimited.example.com/"
 
   describe "RemoteFetcherWorker" do
     setup do
@@ -37,6 +38,11 @@ defmodule Pleroma.Workers.RemoteFetcherWorkerTest do
             status: 200
           }
 
+        %{method: :get, url: @ratelimited_object} ->
+          %Tesla.Env{
+            status: 429
+          }
+
         %{method: :get, url: @content_type_object} ->
           %Tesla.Env{
             status: 200,
@@ -73,6 +79,14 @@ defmodule Pleroma.Workers.RemoteFetcherWorkerTest do
                })
     end
 
+    test "retries jobs for a rate-limited host" do
+      assert {:error, :ratelimit} =
+               perform_job(RemoteFetcherWorker, %{
+                 "op" => "fetch_remote",
+                 "id" => @ratelimited_object
+               })
+    end
+
     test "does not retry jobs for when object returns wrong content type" do
       assert {:cancel, _} =
                perform_job(RemoteFetcherWorker, %{