Merge branch 'feature/object-hashtags-rework' into 'develop'

Hashtags extraction from objects. Background migration infrastructure.

Closes #1840 and #2455

See merge request pleroma/pleroma!3213

CHANGELOG.md

@@ -50,6 +50,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 - Pleroma API: Reroute `/api/pleroma/*` to `/api/v1/pleroma/*`
 
 </details>
+- Improved hashtag timeline performance (requires a background migration). 
 
 ### Added
 

config/config.exs

@@ -654,6 +654,10 @@
 
 config :pleroma, :database, rum_enabled: false
 
+config :pleroma, :features, improved_hashtag_timeline: :auto
+
+config :pleroma, :populate_hashtags_table, fault_rate_allowance: 0.01
+
 config :pleroma, :env, Mix.env()
 
 config :http_signatures,

config/description.exs

@@ -459,6 +459,42 @@
       }
     ]
   },
+  %{
+    group: :pleroma,
+    key: :features,
+    type: :group,
+    description: "Customizable features",
+    children: [
+      %{
+        key: :improved_hashtag_timeline,
+        type: {:dropdown, :atom},
+        description:
+          "Setting to force toggle / force disable improved hashtags timeline. `:enabled` forces hashtags to be fetched from `hashtags` table for hashtags timeline. `:disabled` forces object-embedded hashtags to be used (slower). Keep it `:auto` for automatic behaviour (it is auto-set to `:enabled` [unless overridden] when HashtagsTableMigrator completes).",
+        suggestions: [:auto, :enabled, :disabled]
+      }
+    ]
+  },
+  %{
+    group: :pleroma,
+    key: :populate_hashtags_table,
+    type: :group,
+    description: "`populate_hashtags_table` background migration settings",
+    children: [
+      %{
+        key: :fault_rate_allowance,
+        type: :float,
+        description:
+          "Max accepted rate of objects that failed in the migration. Any value from 0.0 which tolerates no errors to 1.0 which will enable the feature even if hashtags transfer failed for all records.",
+        suggestions: [0.01]
+      },
+      %{
+        key: :sleep_interval_ms,
+        type: :integer,
+        description:
+          "Sleep interval between each chunk of processed records in order to decrease the load on the system (defaults to 0 and should be keep default on most instances)."
+      }
+    ]
+  },
   %{
     group: :pleroma,
     key: :instance,

docs/configuration/cheatsheet.md

@@ -65,6 +65,13 @@ To add configuration to your config file, you can copy it from the base config.
 * `show_reactions`: Let favourites and emoji reactions be viewed through the API (default: `true`).
 * `password_reset_token_validity`: The time after which reset tokens aren't accepted anymore, in seconds (default: one day).
 
+## :database
+* `improved_hashtag_timeline`: Setting to force toggle / force disable improved hashtags timeline. `:enabled` forces hashtags to be fetched from `hashtags` table for hashtags timeline. `:disabled` forces object-embedded hashtags to be used (slower). Keep it `:auto` for automatic behaviour (it is auto-set to `:enabled` [unless overridden] when HashtagsTableMigrator completes).
+
+## Background migrations
+* `populate_hashtags_table/sleep_interval_ms`: Sleep interval between each chunk of processed records in order to decrease the load on the system (defaults to 0 and should be keep default on most instances).
+* `populate_hashtags_table/fault_rate_allowance`: Max rate of failed objects to actually processed objects in order to enable the feature (any value from 0.0 which tolerates no errors to 1.0 which will enable the feature even if hashtags transfer failed for all records).
+
 ## Welcome
 * `direct_message`: - welcome message sent as a direct message.
   * `enabled`: Enables the send a direct message to a newly registered user. Defaults to `false`.

lib/mix/tasks/pleroma/database.ex

@@ -8,10 +8,13 @@ defmodule Mix.Tasks.Pleroma.Database do
   alias Pleroma.Object
   alias Pleroma.Repo
   alias Pleroma.User
+
   require Logger
   require Pleroma.Constants
+
   import Ecto.Query
   import Mix.Pleroma
+
   use Mix.Task
 
   @shortdoc "A collection of database related tasks"
@@ -214,4 +217,32 @@ def run(["set_text_search_config", tsconfig]) do
       shell_info('Done.')
     end
   end
+
+  # Rolls back a specific migration (leaving subsequent migrations applied).
+  # WARNING: imposes a risk of unrecoverable data loss — proceed at your own responsibility.
+  # Based on https://stackoverflow.com/a/53825840
+  def run(["rollback", version]) do
+    prompt = "SEVERE WARNING: this operation may result in unrecoverable data loss. Continue?"
+
+    if shell_prompt(prompt, "n") in ~w(Yn Y y) do
+      {_, result, _} =
+        Ecto.Migrator.with_repo(Pleroma.Repo, fn repo ->
+          version = String.to_integer(version)
+          re = ~r/^#{version}_.*\.exs/
+          path = Ecto.Migrator.migrations_path(repo)
+
+          with {_, "" <> file} <- {:find, Enum.find(File.ls!(path), &String.match?(&1, re))},
+               {_, [{mod, _} | _]} <- {:compile, Code.compile_file(Path.join(path, file))},
+               {_, :ok} <- {:rollback, Ecto.Migrator.down(repo, version, mod)} do
+            {:ok, "Reversed migration: #{file}"}
+          else
+            {:find, _} -> {:error, "No migration found with version prefix: #{version}"}
+            {:compile, e} -> {:error, "Problem compiling migration module: #{inspect(e)}"}
+            {:rollback, e} -> {:error, "Problem reversing migration: #{inspect(e)}"}
+          end
+        end)
+
+      shell_info(inspect(result))
+    end
+  end
 end

lib/pleroma/activity.ex

@@ -113,6 +113,7 @@ def with_preloaded_bookmark(query, %User{} = user) do
     from([a] in query,
       left_join: b in Bookmark,
       on: b.user_id == ^user.id and b.activity_id == a.id,
+      as: :bookmark,
       preload: [bookmark: b]
     )
   end
@@ -123,6 +124,7 @@ def with_preloaded_report_notes(query) do
     from([a] in query,
       left_join: r in ReportNote,
       on: a.id == r.activity_id,
+      as: :report_note,
       preload: [report_notes: r]
     )
   end

lib/pleroma/activity/ir/topics.ex

@@ -48,14 +48,12 @@ defp item_creation_tags(tags, _, _) do
     tags
   end
 
-  defp hashtags_to_topics(%{data: %{"tag" => tags}}) do
+  defp hashtags_to_topics(object) do
-    tags
+    object
-    |> Enum.filter(&is_bitstring(&1))
+    |> Object.hashtags()
-    |> Enum.map(fn tag -> "hashtag:" <> tag end)
+    |> Enum.map(fn hashtag -> "hashtag:" <> hashtag end)
   end
 
-  defp hashtags_to_topics(_), do: []
-
   defp remote_topics(%{local: true}), do: []
 
   defp remote_topics(%{actor: actor}) when is_binary(actor),

lib/pleroma/application.ex

@@ -103,9 +103,7 @@ def start(_type, _args) do
         task_children(@mix_env) ++
         dont_run_in_test(@mix_env) ++
         chat_child(chat_enabled?()) ++
-        [
+        [Pleroma.Gopher.Server]
-          Pleroma.Gopher.Server
-        ]
 
     # See http://elixir-lang.org/docs/stable/elixir/Supervisor.html
     # for other strategies and supported options
@@ -230,6 +228,12 @@ defp dont_run_in_test(_) do
          keys: :duplicate,
          partitions: System.schedulers_online()
        ]}
+    ] ++ background_migrators()
+  end
+
+  defp background_migrators do
+    [
+      Pleroma.Migrators.HashtagsTableMigrator
     ]
   end
 

lib/pleroma/config.ex

@@ -99,4 +99,8 @@ def restrict_unauthenticated_access?(resource, kind) do
   def oauth_consumer_strategies, do: get([:auth, :oauth_consumer_strategies], [])
 
   def oauth_consumer_enabled?, do: oauth_consumer_strategies() != []
+
+  def feature_enabled?(feature_name) do
+    get([:features, feature_name]) not in [nil, false, :disabled, :auto]
+  end
 end

lib/pleroma/data_migration.ex

@@ -0,0 +1,45 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2021 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+defmodule Pleroma.DataMigration do
+  use Ecto.Schema
+
+  alias Pleroma.DataMigration
+  alias Pleroma.DataMigration.State
+  alias Pleroma.Repo
+
+  import Ecto.Changeset
+  import Ecto.Query
+
+  schema "data_migrations" do
+    field(:name, :string)
+    field(:state, State, default: :pending)
+    field(:feature_lock, :boolean, default: false)
+    field(:params, :map, default: %{})
+    field(:data, :map, default: %{})
+
+    timestamps()
+  end
+
+  def changeset(data_migration, params \\ %{}) do
+    data_migration
+    |> cast(params, [:name, :state, :feature_lock, :params, :data])
+    |> validate_required([:name])
+    |> unique_constraint(:name)
+  end
+
+  def update_one_by_id(id, params \\ %{}) do
+    with {1, _} <-
+           from(dm in DataMigration, where: dm.id == ^id)
+           |> Repo.update_all(set: params) do
+      :ok
+    end
+  end
+
+  def get_by_name(name) do
+    Repo.get_by(DataMigration, name: name)
+  end
+
+  def populate_hashtags_table, do: get_by_name("populate_hashtags_table")
+end

lib/pleroma/delivery.ex

@@ -9,7 +9,6 @@ defmodule Pleroma.Delivery do
   alias Pleroma.Object
   alias Pleroma.Repo
   alias Pleroma.User
-  alias Pleroma.User
 
   import Ecto.Changeset
   import Ecto.Query

lib/pleroma/ecto_enums.ex

@@ -17,3 +17,11 @@
   follow_accept: 2,
   follow_reject: 3
 )
+
+defenum(Pleroma.DataMigration.State,
+  pending: 1,
+  running: 2,
+  complete: 3,
+  failed: 4,
+  manual: 5
+)

lib/pleroma/hashtag.ex

@@ -0,0 +1,106 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2020 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+defmodule Pleroma.Hashtag do
+  use Ecto.Schema
+
+  import Ecto.Changeset
+  import Ecto.Query
+
+  alias Ecto.Multi
+  alias Pleroma.Hashtag
+  alias Pleroma.Object
+  alias Pleroma.Repo
+
+  schema "hashtags" do
+    field(:name, :string)
+
+    many_to_many(:objects, Object, join_through: "hashtags_objects", on_replace: :delete)
+
+    timestamps()
+  end
+
+  def normalize_name(name) do
+    name
+    |> String.downcase()
+    |> String.trim()
+  end
+
+  def get_or_create_by_name(name) do
+    changeset = changeset(%Hashtag{}, %{name: name})
+
+    Repo.insert(
+      changeset,
+      on_conflict: [set: [name: get_field(changeset, :name)]],
+      conflict_target: :name,
+      returning: true
+    )
+  end
+
+  def get_or_create_by_names(names) when is_list(names) do
+    names = Enum.map(names, &normalize_name/1)
+    timestamp = NaiveDateTime.truncate(NaiveDateTime.utc_now(), :second)
+
+    structs =
+      Enum.map(names, fn name ->
+        %Hashtag{}
+        |> changeset(%{name: name})
+        |> Map.get(:changes)
+        |> Map.merge(%{inserted_at: timestamp, updated_at: timestamp})
+      end)
+
+    try do
+      with {:ok, %{query_op: hashtags}} <-
+             Multi.new()
+             |> Multi.insert_all(:insert_all_op, Hashtag, structs,
+               on_conflict: :nothing,
+               conflict_target: :name
+             )
+             |> Multi.run(:query_op, fn _repo, _changes ->
+               {:ok, Repo.all(from(ht in Hashtag, where: ht.name in ^names))}
+             end)
+             |> Repo.transaction() do
+        {:ok, hashtags}
+      else
+        {:error, _name, value, _changes_so_far} -> {:error, value}
+      end
+    rescue
+      e -> {:error, e}
+    end
+  end
+
+  def changeset(%Hashtag{} = struct, params) do
+    struct
+    |> cast(params, [:name])
+    |> update_change(:name, &normalize_name/1)
+    |> validate_required([:name])
+    |> unique_constraint(:name)
+  end
+
+  def unlink(%Object{id: object_id}) do
+    with {_, hashtag_ids} <-
+           from(hto in "hashtags_objects",
+             where: hto.object_id == ^object_id,
+             select: hto.hashtag_id
+           )
+           |> Repo.delete_all(),
+         {:ok, unreferenced_count} <- delete_unreferenced(hashtag_ids) do
+      {:ok, length(hashtag_ids), unreferenced_count}
+    end
+  end
+
+  @delete_unreferenced_query """
+  DELETE FROM hashtags WHERE id IN
+    (SELECT hashtags.id FROM hashtags
+      LEFT OUTER JOIN hashtags_objects
+        ON hashtags_objects.hashtag_id = hashtags.id
+      WHERE hashtags_objects.hashtag_id IS NULL AND hashtags.id = ANY($1));
+  """
+
+  def delete_unreferenced(ids) do
+    with {:ok, %{num_rows: deleted_count}} <- Repo.query(@delete_unreferenced_query, [ids]) do
+      {:ok, deleted_count}
+    end
+  end
+end

lib/pleroma/migrators/hashtags_table_migrator.ex

@@ -0,0 +1,208 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2021 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+defmodule Pleroma.Migrators.HashtagsTableMigrator do
+  defmodule State do
+    use Pleroma.Migrators.Support.BaseMigratorState
+
+    @impl Pleroma.Migrators.Support.BaseMigratorState
+    defdelegate data_migration(), to: Pleroma.DataMigration, as: :populate_hashtags_table
+  end
+
+  use Pleroma.Migrators.Support.BaseMigrator
+
+  alias Pleroma.Hashtag
+  alias Pleroma.Migrators.Support.BaseMigrator
+  alias Pleroma.Object
+
+  @impl BaseMigrator
+  def feature_config_path, do: [:features, :improved_hashtag_timeline]
+
+  @impl BaseMigrator
+  def fault_rate_allowance, do: Config.get([:populate_hashtags_table, :fault_rate_allowance], 0)
+
+  @impl BaseMigrator
+  def perform do
+    data_migration_id = data_migration_id()
+    max_processed_id = get_stat(:max_processed_id, 0)
+
+    Logger.info("Transferring embedded hashtags to `hashtags` (from oid: #{max_processed_id})...")
+
+    query()
+    |> where([object], object.id > ^max_processed_id)
+    |> Repo.chunk_stream(100, :batches, timeout: :infinity)
+    |> Stream.each(fn objects ->
+      object_ids = Enum.map(objects, & &1.id)
+
+      results = Enum.map(objects, &transfer_object_hashtags(&1))
+
+      failed_ids =
+        results
+        |> Enum.filter(&(elem(&1, 0) == :error))
+        |> Enum.map(&elem(&1, 1))
+
+      # Count of objects with hashtags: `{:noop, id}` is returned for objects having other AS2 tags
+      chunk_affected_count =
+        results
+        |> Enum.filter(&(elem(&1, 0) == :ok))
+        |> length()
+
+      for failed_id <- failed_ids do
+        _ =
+          Repo.query(
+            "INSERT INTO data_migration_failed_ids(data_migration_id, record_id) " <>
+              "VALUES ($1, $2) ON CONFLICT DO NOTHING;",
+            [data_migration_id, failed_id]
+          )
+      end
+
+      _ =
+        Repo.query(
+          "DELETE FROM data_migration_failed_ids " <>
+            "WHERE data_migration_id = $1 AND record_id = ANY($2)",
+          [data_migration_id, object_ids -- failed_ids]
+        )
+
+      max_object_id = Enum.at(object_ids, -1)
+
+      put_stat(:max_processed_id, max_object_id)
+      increment_stat(:iteration_processed_count, length(object_ids))
+      increment_stat(:processed_count, length(object_ids))
+      increment_stat(:failed_count, length(failed_ids))
+      increment_stat(:affected_count, chunk_affected_count)
+      put_stat(:records_per_second, records_per_second())
+      persist_state()
+
+      # A quick and dirty approach to controlling the load this background migration imposes
+      sleep_interval = Config.get([:populate_hashtags_table, :sleep_interval_ms], 0)
+      Process.sleep(sleep_interval)
+    end)
+    |> Stream.run()
+  end
+
+  @impl BaseMigrator
+  def query do
+    # Note: most objects have Mention-type AS2 tags and no hashtags (but we can't filter them out)
+    # Note: not checking activity type, expecting remove_non_create_objects_hashtags/_ to clean up
+    from(
+      object in Object,
+      where:
+        fragment("(?)->'tag' IS NOT NULL AND (?)->'tag' != '[]'::jsonb", object.data, object.data),
+      select: %{
+        id: object.id,
+        tag: fragment("(?)->'tag'", object.data)
+      }
+    )
+    |> join(:left, [o], hashtags_objects in fragment("SELECT object_id FROM hashtags_objects"),
+      on: hashtags_objects.object_id == o.id
+    )
+    |> where([_o, hashtags_objects], is_nil(hashtags_objects.object_id))
+  end
+
+  @spec transfer_object_hashtags(Map.t()) :: {:noop | :ok | :error, integer()}
+  defp transfer_object_hashtags(object) do
+    embedded_tags = if Map.has_key?(object, :tag), do: object.tag, else: object.data["tag"]
+    hashtags = Object.object_data_hashtags(%{"tag" => embedded_tags})
+
+    if Enum.any?(hashtags) do
+      transfer_object_hashtags(object, hashtags)
+    else
+      {:noop, object.id}
+    end
+  end
+
+  defp transfer_object_hashtags(object, hashtags) do
+    Repo.transaction(fn ->
+      with {:ok, hashtag_records} <- Hashtag.get_or_create_by_names(hashtags) do
+        maps = Enum.map(hashtag_records, &%{hashtag_id: &1.id, object_id: object.id})
+        base_error = "ERROR when inserting hashtags_objects for object with id #{object.id}"
+
+        try do
+          with {rows_count, _} when is_integer(rows_count) <-
+                 Repo.insert_all("hashtags_objects", maps, on_conflict: :nothing) do
+            object.id
+          else
+            e ->
+              Logger.error("#{base_error}: #{inspect(e)}")
+              Repo.rollback(object.id)
+          end
+        rescue
+          e ->
+            Logger.error("#{base_error}: #{inspect(e)}")
+            Repo.rollback(object.id)
+        end
+      else
+        e ->
+          error = "ERROR: could not create hashtags for object #{object.id}: #{inspect(e)}"
+          Logger.error(error)
+          Repo.rollback(object.id)
+      end
+    end)
+  end
+
+  @impl BaseMigrator
+  def retry_failed do
+    data_migration_id = data_migration_id()
+
+    failed_objects_query()
+    |> Repo.chunk_stream(100, :one)
+    |> Stream.each(fn object ->
+      with {res, _} when res != :error <- transfer_object_hashtags(object) do
+        _ =
+          Repo.query(
+            "DELETE FROM data_migration_failed_ids " <>
+              "WHERE data_migration_id = $1 AND record_id = $2",
+            [data_migration_id, object.id]
+          )
+      end
+    end)
+    |> Stream.run()
+
+    put_stat(:failed_count, failures_count())
+    persist_state()
+
+    force_continue()
+  end
+
+  defp failed_objects_query do
+    from(o in Object)
+    |> join(:inner, [o], dmf in fragment("SELECT * FROM data_migration_failed_ids"),
+      on: dmf.record_id == o.id
+    )
+    |> where([_o, dmf], dmf.data_migration_id == ^data_migration_id())
+    |> order_by([o], asc: o.id)
+  end
+
+  @doc """
+  Service func to delete `hashtags_objects` for legacy objects not associated with Create activity.
+  Also deletes unreferenced `hashtags` records (might occur after deletion of `hashtags_objects`).
+  """
+  def delete_non_create_activities_hashtags do
+    hashtags_objects_cleanup_query = """
+    DELETE FROM hashtags_objects WHERE object_id IN
+      (SELECT DISTINCT objects.id FROM objects
+        JOIN hashtags_objects ON hashtags_objects.object_id = objects.id LEFT JOIN activities
+          ON COALESCE(activities.data->'object'->>'id', activities.data->>'object') =
+            (objects.data->>'id')
+          AND activities.data->>'type' = 'Create'
+        WHERE activities.id IS NULL);
+    """
+
+    hashtags_cleanup_query = """
+    DELETE FROM hashtags WHERE id IN
+      (SELECT hashtags.id FROM hashtags
+        LEFT OUTER JOIN hashtags_objects
+          ON hashtags_objects.hashtag_id = hashtags.id
+        WHERE hashtags_objects.hashtag_id IS NULL);
+    """
+
+    {:ok, %{num_rows: hashtags_objects_count}} =
+      Repo.query(hashtags_objects_cleanup_query, [], timeout: :infinity)
+
+    {:ok, %{num_rows: hashtags_count}} =
+      Repo.query(hashtags_cleanup_query, [], timeout: :infinity)
+
+    {:ok, hashtags_objects_count, hashtags_count}
+  end
+end

lib/pleroma/migrators/support/base_migrator.ex

@@ -0,0 +1,210 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2021 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+defmodule Pleroma.Migrators.Support.BaseMigrator do
+  @moduledoc """
+  Base background migrator functionality.
+  """
+
+  @callback perform() :: any()
+  @callback retry_failed() :: any()
+  @callback feature_config_path() :: list(atom())
+  @callback query() :: Ecto.Query.t()
+  @callback fault_rate_allowance() :: integer() | float()
+
+  defmacro __using__(_opts) do
+    quote do
+      use GenServer
+
+      require Logger
+
+      import Ecto.Query
+
+      alias __MODULE__.State
+      alias Pleroma.Config
+      alias Pleroma.Repo
+
+      @behaviour Pleroma.Migrators.Support.BaseMigrator
+
+      defdelegate data_migration(), to: State
+      defdelegate data_migration_id(), to: State
+      defdelegate state(), to: State
+      defdelegate persist_state(), to: State, as: :persist_to_db
+      defdelegate get_stat(key, value \\ nil), to: State, as: :get_data_key
+      defdelegate put_stat(key, value), to: State, as: :put_data_key
+      defdelegate increment_stat(key, increment), to: State, as: :increment_data_key
+
+      @reg_name {:global, __MODULE__}
+
+      def whereis, do: GenServer.whereis(@reg_name)
+
+      def start_link(_) do
+        case whereis() do
+          nil ->
+            GenServer.start_link(__MODULE__, nil, name: @reg_name)
+
+          pid ->
+            {:ok, pid}
+        end
+      end
+
+      @impl true
+      def init(_) do
+        {:ok, nil, {:continue, :init_state}}
+      end
+
+      @impl true
+      def handle_continue(:init_state, _state) do
+        {:ok, _} = State.start_link(nil)
+
+        data_migration = data_migration()
+        manual_migrations = Config.get([:instance, :manual_data_migrations], [])
+
+        cond do
+          Config.get(:env) == :test ->
+            update_status(:noop)
+
+          is_nil(data_migration) ->
+            message = "Data migration does not exist."
+            update_status(:failed, message)
+            Logger.error("#{__MODULE__}: #{message}")
+
+          data_migration.state == :manual or data_migration.name in manual_migrations ->
+            message = "Data migration is in manual execution or manual fix mode."
+            update_status(:manual, message)
+            Logger.warn("#{__MODULE__}: #{message}")
+
+          data_migration.state == :complete ->
+            on_complete(data_migration)
+
+          true ->
+            send(self(), :perform)
+        end
+
+        {:noreply, nil}
+      end
+
+      @impl true
+      def handle_info(:perform, state) do
+        State.reinit()
+
+        update_status(:running)
+        put_stat(:iteration_processed_count, 0)
+        put_stat(:started_at, NaiveDateTime.utc_now())
+
+        perform()
+
+        fault_rate = fault_rate()
+        put_stat(:fault_rate, fault_rate)
+        fault_rate_allowance = fault_rate_allowance()
+
+        cond do
+          fault_rate == 0 ->
+            set_complete()
+
+          is_float(fault_rate) and fault_rate <= fault_rate_allowance ->
+            message = """
+            Done with fault rate of #{fault_rate} which doesn't exceed #{fault_rate_allowance}.
+            Putting data migration to manual fix mode. Try running `#{__MODULE__}.retry_failed/0`.
+            """
+
+            Logger.warn("#{__MODULE__}: #{message}")
+            update_status(:manual, message)
+            on_complete(data_migration())
+
+          true ->
+            message = "Too many failures. Try running `#{__MODULE__}.retry_failed/0`."
+            Logger.error("#{__MODULE__}: #{message}")
+            update_status(:failed, message)
+        end
+
+        persist_state()
+        {:noreply, state}
+      end
+
+      defp on_complete(data_migration) do
+        if data_migration.feature_lock || feature_state() == :disabled do
+          Logger.warn(
+            "#{__MODULE__}: migration complete but feature is locked; consider enabling."
+          )
+
+          :noop
+        else
+          Config.put(feature_config_path(), :enabled)
+          :ok
+        end
+      end
+
+      @doc "Approximate count for current iteration (including processed records count)"
+      def count(force \\ false, timeout \\ :infinity) do
+        stored_count = get_stat(:count)
+
+        if stored_count && !force do
+          stored_count
+        else
+          processed_count = get_stat(:processed_count, 0)
+          max_processed_id = get_stat(:max_processed_id, 0)
+          query = where(query(), [entity], entity.id > ^max_processed_id)
+
+          count = Repo.aggregate(query, :count, :id, timeout: timeout) + processed_count
+          put_stat(:count, count)
+          persist_state()
+
+          count
+        end
+      end
+
+      def failures_count do
+        with {:ok, %{rows: [[count]]}} <-
+               Repo.query(
+                 "SELECT COUNT(record_id) FROM data_migration_failed_ids WHERE data_migration_id = $1;",
+                 [data_migration_id()]
+               ) do
+          count
+        end
+      end
+
+      def feature_state, do: Config.get(feature_config_path())
+
+      def force_continue do
+        send(whereis(), :perform)
+      end
+
+      def force_restart do
+        :ok = State.reset()
+        force_continue()
+      end
+
+      def set_complete do
+        update_status(:complete)
+        persist_state()
+        on_complete(data_migration())
+      end
+
+      defp update_status(status, message \\ nil) do
+        put_stat(:state, status)
+        put_stat(:message, message)
+      end
+
+      defp fault_rate do
+        with failures_count when is_integer(failures_count) <- failures_count() do
+          failures_count / Enum.max([get_stat(:affected_count, 0), 1])
+        else
+          _ -> :error
+        end
+      end
+
+      defp records_per_second do
+        get_stat(:iteration_processed_count, 0) / Enum.max([running_time(), 1])
+      end
+
+      defp running_time do
+        NaiveDateTime.diff(
+          NaiveDateTime.utc_now(),
+          get_stat(:started_at, NaiveDateTime.utc_now())
+        )
+      end
+    end
+  end
+end

lib/pleroma/migrators/support/base_migrator_state.ex

@@ -0,0 +1,117 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2021 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+defmodule Pleroma.Migrators.Support.BaseMigratorState do
+  @moduledoc """
+  Base background migrator state functionality.
+  """
+
+  @callback data_migration() :: Pleroma.DataMigration.t()
+
+  defmacro __using__(_opts) do
+    quote do
+      use Agent
+
+      alias Pleroma.DataMigration
+
+      @behaviour Pleroma.Migrators.Support.BaseMigratorState
+      @reg_name {:global, __MODULE__}
+
+      def start_link(_) do
+        Agent.start_link(fn -> load_state_from_db() end, name: @reg_name)
+      end
+
+      def data_migration, do: raise("data_migration/0 is not implemented")
+      defoverridable data_migration: 0
+
+      defp load_state_from_db do
+        data_migration = data_migration()
+
+        data =
+          if data_migration do
+            Map.new(data_migration.data, fn {k, v} -> {String.to_atom(k), v} end)
+          else
+            %{}
+          end
+
+        %{
+          data_migration_id: data_migration && data_migration.id,
+          data: data
+        }
+      end
+
+      def persist_to_db do
+        %{data_migration_id: data_migration_id, data: data} = state()
+
+        if data_migration_id do
+          DataMigration.update_one_by_id(data_migration_id, data: data)
+        else
+          {:error, :nil_data_migration_id}
+        end
+      end
+
+      def reset do
+        %{data_migration_id: data_migration_id} = state()
+
+        with false <- is_nil(data_migration_id),
+             :ok <-
+               DataMigration.update_one_by_id(data_migration_id,
+                 state: :pending,
+                 data: %{}
+               ) do
+          reinit()
+        else
+          true -> {:error, :nil_data_migration_id}
+          e -> e
+        end
+      end
+
+      def reinit do
+        Agent.update(@reg_name, fn _state -> load_state_from_db() end)
+      end
+
+      def state do
+        Agent.get(@reg_name, & &1)
+      end
+
+      def get_data_key(key, default \\ nil) do
+        get_in(state(), [:data, key]) || default
+      end
+
+      def put_data_key(key, value) do
+        _ = persist_non_data_change(key, value)
+
+        Agent.update(@reg_name, fn state ->
+          put_in(state, [:data, key], value)
+        end)
+      end
+
+      def increment_data_key(key, increment \\ 1) do
+        Agent.update(@reg_name, fn state ->
+          initial_value = get_in(state, [:data, key]) || 0
+          updated_value = initial_value + increment
+          put_in(state, [:data, key], updated_value)
+        end)
+      end
+
+      defp persist_non_data_change(:state, value) do
+        with true <- get_data_key(:state) != value,
+             true <- value in Pleroma.DataMigration.State.__valid_values__(),
+             %{data_migration_id: data_migration_id} when not is_nil(data_migration_id) <-
+               state() do
+          DataMigration.update_one_by_id(data_migration_id, state: value)
+        else
+          false -> :ok
+          _ -> {:error, :nil_data_migration_id}
+        end
+      end
+
+      defp persist_non_data_change(_, _) do
+        nil
+      end
+
+      def data_migration_id, do: Map.get(state(), :data_migration_id)
+    end
+  end
+end

lib/pleroma/object.ex

@@ -10,6 +10,7 @@ defmodule Pleroma.Object do
 
   alias Pleroma.Activity
   alias Pleroma.Config
+  alias Pleroma.Hashtag
   alias Pleroma.Object
   alias Pleroma.Object.Fetcher
   alias Pleroma.ObjectTombstone
@@ -28,6 +29,8 @@ defmodule Pleroma.Object do
   schema "objects" do
     field(:data, :map)
 
+    many_to_many(:hashtags, Hashtag, join_through: "hashtags_objects", on_replace: :delete)
+
     timestamps()
   end
 
@@ -49,7 +52,8 @@ def with_joined_activity(query, activity_type \\ "Create", join_type \\ :inner)
   end
 
   def create(data) do
-    Object.change(%Object{}, %{data: data})
+    %Object{}
+    |> Object.change(%{data: data})
     |> Repo.insert()
   end
 
@@ -58,8 +62,41 @@ def change(struct, params \\ %{}) do
     |> cast(params, [:data])
     |> validate_required([:data])
     |> unique_constraint(:ap_id, name: :objects_unique_apid_index)
+    # Expecting `maybe_handle_hashtags_change/1` to run last:
+    |> maybe_handle_hashtags_change(struct)
   end
 
+  # Note: not checking activity type (assuming non-legacy objects are associated with Create act.)
+  defp maybe_handle_hashtags_change(changeset, struct) do
+    with %Ecto.Changeset{valid?: true} <- changeset,
+         data_hashtags_change = get_change(changeset, :data),
+         {_, true} <- {:changed, hashtags_changed?(struct, data_hashtags_change)},
+         {:ok, hashtag_records} <-
+           data_hashtags_change
+           |> object_data_hashtags()
+           |> Hashtag.get_or_create_by_names() do
+      put_assoc(changeset, :hashtags, hashtag_records)
+    else
+      %{valid?: false} ->
+        changeset
+
+      {:changed, false} ->
+        changeset
+
+      {:error, _} ->
+        validate_change(changeset, :data, fn _, _ ->
+          [data: "error referencing hashtags"]
+        end)
+    end
+  end
+
+  defp hashtags_changed?(%Object{} = struct, %{"tag" => _} = data) do
+    Enum.sort(embedded_hashtags(struct)) !=
+      Enum.sort(object_data_hashtags(data))
+  end
+
+  defp hashtags_changed?(_, _), do: false
+
   def get_by_id(nil), do: nil
   def get_by_id(id), do: Repo.get(Object, id)
 
@@ -187,9 +224,13 @@ def make_tombstone(%Object{data: %{"id" => id, "type" => type}}, deleted \\ Date
   def swap_object_with_tombstone(object) do
     tombstone = make_tombstone(object)
 
+    with {:ok, object} <-
            object
            |> Object.change(%{data: tombstone})
-    |> Repo.update()
+           |> Repo.update() do
+      Hashtag.unlink(object)
+      {:ok, object}
+    end
   end
 
   def delete(%Object{data: %{"id" => id}} = object) do
@@ -349,4 +390,39 @@ def replies(object, opts \\ []) do
 
   def self_replies(object, opts \\ []),
     do: replies(object, Keyword.put(opts, :self_only, true))
+
+  def tags(%Object{data: %{"tag" => tags}}) when is_list(tags), do: tags
+
+  def tags(_), do: []
+
+  def hashtags(%Object{} = object) do
+    # Note: always using embedded hashtags regardless whether they are migrated to hashtags table
+    #   (embedded hashtags stay in sync anyways, and we avoid extra joins and preload hassle)
+    embedded_hashtags(object)
+  end
+
+  def embedded_hashtags(%Object{data: data}) do
+    object_data_hashtags(data)
+  end
+
+  def embedded_hashtags(_), do: []
+
+  def object_data_hashtags(%{"tag" => tags}) when is_list(tags) do
+    tags
+    |> Enum.filter(fn
+      %{"type" => "Hashtag"} = data -> Map.has_key?(data, "name")
+      plain_text when is_bitstring(plain_text) -> true
+      _ -> false
+    end)
+    |> Enum.map(fn
+      %{"name" => "#" <> hashtag} -> String.downcase(hashtag)
+      %{"name" => hashtag} -> String.downcase(hashtag)
+      hashtag when is_bitstring(hashtag) -> String.downcase(hashtag)
+    end)
+    |> Enum.uniq()
+    # Note: "" elements (plain text) might occur in `data.tag` for incoming objects
+    |> Enum.filter(&(&1 not in [nil, ""]))
+  end
+
+  def object_data_hashtags(_), do: []
 end

lib/pleroma/pagination.ex

@@ -93,6 +93,7 @@ defp cast_params(params) do
       max_id: :string,
       offset: :integer,
       limit: :integer,
+      skip_extra_order: :boolean,
       skip_order: :boolean
     }
 
@@ -114,6 +115,8 @@ defp restrict(query, :max_id, %{max_id: max_id}, table_binding) do
 
   defp restrict(query, :order, %{skip_order: true}, _), do: query
 
+  defp restrict(%{order_bys: [_ | _]} = query, :order, %{skip_extra_order: true}, _), do: query
+
   defp restrict(query, :order, %{min_id: _}, table_binding) do
     order_by(
       query,

lib/pleroma/repo.ex

@@ -8,6 +8,8 @@ defmodule Pleroma.Repo do
     adapter: Ecto.Adapters.Postgres,
     migration_timestamps: [type: :naive_datetime_usec]
 
+  use Ecto.Explain
+
   import Ecto.Query
   require Logger
 
@@ -63,8 +65,8 @@ def get_assoc(resource, association) do
   iex> Pleroma.Repo.chunk_stream(Pleroma.Activity.Queries.by_actor(ap_id), 500, :batches)
   """
   @spec chunk_stream(Ecto.Query.t(), integer(), atom()) :: Enumerable.t()
-  def chunk_stream(query, chunk_size, returns_as \\ :one) do
+  def chunk_stream(query, chunk_size, returns_as \\ :one, query_options \\ []) do
-    # We don't actually need start and end funcitons of resource streaming,
+    # We don't actually need start and end functions of resource streaming,
     # but it seems to be the only way to not fetch records one-by-one and
     # have individual records be the elements of the stream, instead of
     # lists of records
@@ -76,7 +78,7 @@ def chunk_stream(query, chunk_size, returns_as \\ :one) do
           |> order_by(asc: :id)
           |> where([r], r.id > ^last_id)
           |> limit(^chunk_size)
-          |> all()
+          |> all(query_options)
           |> case do
             [] ->
               {:halt, last_id}

lib/pleroma/web/activity_pub/activity_pub.ex

@@ -10,6 +10,7 @@ defmodule Pleroma.Web.ActivityPub.ActivityPub do
   alias Pleroma.Conversation
   alias Pleroma.Conversation.Participation
   alias Pleroma.Filter
+  alias Pleroma.Hashtag
   alias Pleroma.Maps
   alias Pleroma.Notification
   alias Pleroma.Object
@@ -465,6 +466,23 @@ def fetch_latest_direct_activity_id_for_context(context, opts \\ %{}) do
     |> Repo.one()
   end
 
+  defp fetch_paginated_optimized(query, opts, pagination) do
+    # Note: tag-filtering funcs may apply "ORDER BY objects.id DESC",
+    #   and extra sorting on "activities.id DESC NULLS LAST" would worse the query plan
+    opts = Map.put(opts, :skip_extra_order, true)
+
+    Pagination.fetch_paginated(query, opts, pagination)
+  end
+
+  def fetch_activities(recipients, opts \\ %{}, pagination \\ :keyset) do
+    list_memberships = Pleroma.List.memberships(opts[:user])
+
+    fetch_activities_query(recipients ++ list_memberships, opts)
+    |> fetch_paginated_optimized(opts, pagination)
+    |> Enum.reverse()
+    |> maybe_update_cc(list_memberships, opts[:user])
+  end
+
   @spec fetch_public_or_unlisted_activities(map(), Pagination.type()) :: [Activity.t()]
   def fetch_public_or_unlisted_activities(opts \\ %{}, pagination \\ :keyset) do
     opts = Map.delete(opts, :user)
@@ -472,7 +490,7 @@ def fetch_public_or_unlisted_activities(opts \\ %{}, pagination \\ :keyset) do
     [Constants.as_public()]
     |> fetch_activities_query(opts)
     |> restrict_unlisted(opts)
-    |> Pagination.fetch_paginated(opts, pagination)
+    |> fetch_paginated_optimized(opts, pagination)
   end
 
   @spec fetch_public_activities(map(), Pagination.type()) :: [Activity.t()]
@@ -693,52 +711,144 @@ defp restrict_since(query, %{since_id: since_id}) do
 
   defp restrict_since(query, _), do: query
 
-  defp restrict_tag_reject(_query, %{tag_reject: _tag_reject, skip_preload: true}) do
+  defp restrict_embedded_tag_all(_query, %{tag_all: _tag_all, skip_preload: true}) do
-    raise "Can't use the child object without preloading!"
+    raise_on_missing_preload()
   end
 
-  defp restrict_tag_reject(query, %{tag_reject: [_ | _] = tag_reject}) do
+  defp restrict_embedded_tag_all(query, %{tag_all: [_ | _] = tag_all}) do
-    from(
-      [_activity, object] in query,
-      where: fragment("not (?)->'tag' \\?| (?)", object.data, ^tag_reject)
-    )
-  end
-
-  defp restrict_tag_reject(query, _), do: query
-
-  defp restrict_tag_all(_query, %{tag_all: _tag_all, skip_preload: true}) do
-    raise "Can't use the child object without preloading!"
-  end
-
-  defp restrict_tag_all(query, %{tag_all: [_ | _] = tag_all}) do
     from(
       [_activity, object] in query,
       where: fragment("(?)->'tag' \\?& (?)", object.data, ^tag_all)
     )
   end
 
-  defp restrict_tag_all(query, _), do: query
+  defp restrict_embedded_tag_all(query, %{tag_all: tag}) when is_binary(tag) do
+    restrict_embedded_tag_any(query, %{tag: tag})
+  end
 
-  defp restrict_tag(_query, %{tag: _tag, skip_preload: true}) do
+  defp restrict_embedded_tag_all(query, _), do: query
+
+  defp restrict_embedded_tag_any(_query, %{tag: _tag, skip_preload: true}) do
+    raise_on_missing_preload()
+  end
+
+  defp restrict_embedded_tag_any(query, %{tag: [_ | _] = tag_any}) do
+    from(
+      [_activity, object] in query,
+      where: fragment("(?)->'tag' \\?| (?)", object.data, ^tag_any)
+    )
+  end
+
+  defp restrict_embedded_tag_any(query, %{tag: tag}) when is_binary(tag) do
+    restrict_embedded_tag_any(query, %{tag: [tag]})
+  end
+
+  defp restrict_embedded_tag_any(query, _), do: query
+
+  defp restrict_embedded_tag_reject_any(_query, %{tag_reject: _tag_reject, skip_preload: true}) do
+    raise_on_missing_preload()
+  end
+
+  defp restrict_embedded_tag_reject_any(query, %{tag_reject: [_ | _] = tag_reject}) do
+    from(
+      [_activity, object] in query,
+      where: fragment("not (?)->'tag' \\?| (?)", object.data, ^tag_reject)
+    )
+  end
+
+  defp restrict_embedded_tag_reject_any(query, %{tag_reject: tag_reject})
+       when is_binary(tag_reject) do
+    restrict_embedded_tag_reject_any(query, %{tag_reject: [tag_reject]})
+  end
+
+  defp restrict_embedded_tag_reject_any(query, _), do: query
+
+  defp object_ids_query_for_tags(tags) do
+    from(hto in "hashtags_objects")
+    |> join(:inner, [hto], ht in Pleroma.Hashtag, on: hto.hashtag_id == ht.id)
+    |> where([hto, ht], ht.name in ^tags)
+    |> select([hto], hto.object_id)
+    |> distinct([hto], true)
+  end
+
+  defp restrict_hashtag_all(_query, %{tag_all: _tag, skip_preload: true}) do
+    raise_on_missing_preload()
+  end
+
+  defp restrict_hashtag_all(query, %{tag_all: [single_tag]}) do
+    restrict_hashtag_any(query, %{tag: single_tag})
+  end
+
+  defp restrict_hashtag_all(query, %{tag_all: [_ | _] = tags}) do
+    from(
+      [_activity, object] in query,
+      where:
+        fragment(
+          """
+          (SELECT array_agg(hashtags.name) FROM hashtags JOIN hashtags_objects
+            ON hashtags_objects.hashtag_id = hashtags.id WHERE hashtags.name = ANY(?)
+              AND hashtags_objects.object_id = ?) @> ?
+          """,
+          ^tags,
+          object.id,
+          ^tags
+        )
+    )
+  end
+
+  defp restrict_hashtag_all(query, %{tag_all: tag}) when is_binary(tag) do
+    restrict_hashtag_all(query, %{tag_all: [tag]})
+  end
+
+  defp restrict_hashtag_all(query, _), do: query
+
+  defp restrict_hashtag_any(_query, %{tag: _tag, skip_preload: true}) do
+    raise_on_missing_preload()
+  end
+
+  defp restrict_hashtag_any(query, %{tag: [_ | _] = tags}) do
+    hashtag_ids =
+      from(ht in Hashtag, where: ht.name in ^tags, select: ht.id)
+      |> Repo.all()
+
+    # Note: NO extra ordering should be done on "activities.id desc nulls last" for optimal plan
+    from(
+      [_activity, object] in query,
+      join: hto in "hashtags_objects",
+      on: hto.object_id == object.id,
+      where: hto.hashtag_id in ^hashtag_ids,
+      distinct: [desc: object.id],
+      order_by: [desc: object.id]
+    )
+  end
+
+  defp restrict_hashtag_any(query, %{tag: tag}) when is_binary(tag) do
+    restrict_hashtag_any(query, %{tag: [tag]})
+  end
+
+  defp restrict_hashtag_any(query, _), do: query
+
+  defp restrict_hashtag_reject_any(_query, %{tag_reject: _tag_reject, skip_preload: true}) do
+    raise_on_missing_preload()
+  end
+
+  defp restrict_hashtag_reject_any(query, %{tag_reject: [_ | _] = tags_reject}) do
+    from(
+      [_activity, object] in query,
+      where: object.id not in subquery(object_ids_query_for_tags(tags_reject))
+    )
+  end
+
+  defp restrict_hashtag_reject_any(query, %{tag_reject: tag_reject}) when is_binary(tag_reject) do
+    restrict_hashtag_reject_any(query, %{tag_reject: [tag_reject]})
+  end
+
+  defp restrict_hashtag_reject_any(query, _), do: query
+
+  defp raise_on_missing_preload do
     raise "Can't use the child object without preloading!"
   end
 
-  defp restrict_tag(query, %{tag: tag}) when is_list(tag) do
-    from(
-      [_activity, object] in query,
-      where: fragment("(?)->'tag' \\?| (?)", object.data, ^tag)
-    )
-  end
-
-  defp restrict_tag(query, %{tag: tag}) when is_binary(tag) do
-    from(
-      [_activity, object] in query,
-      where: fragment("(?)->'tag' \\? (?)", object.data, ^tag)
-    )
-  end
-
-  defp restrict_tag(query, _), do: query
-
   defp restrict_recipients(query, [], _user), do: query
 
   defp restrict_recipients(query, recipients, nil) do
@@ -1098,6 +1208,26 @@ defp maybe_order(query, %{order: :asc}) do
 
   defp maybe_order(query, _), do: query
 
+  defp normalize_fetch_activities_query_opts(opts) do
+    Enum.reduce([:tag, :tag_all, :tag_reject], opts, fn key, opts ->
+      case opts[key] do
+        value when is_bitstring(value) ->
+          Map.put(opts, key, Hashtag.normalize_name(value))
+
+        value when is_list(value) ->
+          normalized_value =
+            value
+            |> Enum.map(&Hashtag.normalize_name/1)
+            |> Enum.uniq()
+
+          Map.put(opts, key, normalized_value)
+
+        _ ->
+          opts
+      end
+    end)
+  end
+
   defp fetch_activities_query_ap_ids_ops(opts) do
     source_user = opts[:muting_user]
     ap_id_relationships = if source_user, do: [:mute, :reblog_mute], else: []
@@ -1121,6 +1251,8 @@ defp fetch_activities_query_ap_ids_ops(opts) do
   end
 
   def fetch_activities_query(recipients, opts \\ %{}) do
+    opts = normalize_fetch_activities_query_opts(opts)
+
     {restrict_blocked_opts, restrict_muted_opts, restrict_muted_reblogs_opts} =
       fetch_activities_query_ap_ids_ops(opts)
 
@@ -1128,6 +1260,7 @@ def fetch_activities_query(recipients, opts \\ %{}) do
       skip_thread_containment: Config.get([:instance, :skip_thread_containment])
     }
 
+    query =
       Activity
       |> maybe_preload_objects(opts)
       |> maybe_preload_bookmarks(opts)
@@ -1136,9 +1269,6 @@ def fetch_activities_query(recipients, opts \\ %{}) do
       |> maybe_order(opts)
       |> restrict_recipients(recipients, opts[:user])
       |> restrict_replies(opts)
-    |> restrict_tag(opts)
-    |> restrict_tag_reject(opts)
-    |> restrict_tag_all(opts)
       |> restrict_since(opts)
       |> restrict_local(opts)
       |> restrict_remote(opts)
@@ -1163,15 +1293,18 @@ def fetch_activities_query(recipients, opts \\ %{}) do
       |> exclude_chat_messages(opts)
       |> exclude_invisible_actors(opts)
       |> exclude_visibility(opts)
+
+    if Config.feature_enabled?(:improved_hashtag_timeline) do
+      query
+      |> restrict_hashtag_any(opts)
+      |> restrict_hashtag_all(opts)
+      |> restrict_hashtag_reject_any(opts)
+    else
+      query
+      |> restrict_embedded_tag_any(opts)
+      |> restrict_embedded_tag_all(opts)
+      |> restrict_embedded_tag_reject_any(opts)
     end
-
-  def fetch_activities(recipients, opts \\ %{}, pagination \\ :keyset) do
-    list_memberships = Pleroma.List.memberships(opts[:user])
-
-    fetch_activities_query(recipients ++ list_memberships, opts)
-    |> Pagination.fetch_paginated(opts, pagination)
-    |> Enum.reverse()
-    |> maybe_update_cc(list_memberships, opts[:user])
   end
 
   @doc """

lib/pleroma/web/activity_pub/mrf/simple_policy.ex

@@ -74,9 +74,11 @@ defp check_media_nsfw(
 
     object =
       if MRF.subdomain_match?(media_nsfw, actor_host) do
-        tags = (child_object["tag"] || []) ++ ["nsfw"]
+        child_object =
-        child_object = Map.put(child_object, "tag", tags)
+          child_object
-        child_object = Map.put(child_object, "sensitive", true)
+          |> Map.put("tag", (child_object["tag"] || []) ++ ["nsfw"])
+          |> Map.put("sensitive", true)
+
         Map.put(object, "object", child_object)
       else
         object

lib/pleroma/web/activity_pub/transmogrifier.ex

@@ -32,18 +32,18 @@ defmodule Pleroma.Web.ActivityPub.Transmogrifier do
   """
   def fix_object(object, options \\ []) do
     object
-    |> strip_internal_fields
+    |> strip_internal_fields()
-    |> fix_actor
+    |> fix_actor()
-    |> fix_url
+    |> fix_url()
-    |> fix_attachments
+    |> fix_attachments()
-    |> fix_context
+    |> fix_context()
     |> fix_in_reply_to(options)
-    |> fix_emoji
+    |> fix_emoji()
-    |> fix_tag
+    |> fix_tag()
-    |> set_sensitive
+    |> set_sensitive()
-    |> fix_content_map
+    |> fix_content_map()
-    |> fix_addressing
+    |> fix_addressing()
-    |> fix_summary
+    |> fix_summary()
     |> fix_type(options)
   end
 
@@ -315,10 +315,9 @@ def fix_tag(%{"tag" => tag} = object) when is_list(tag) do
     tags =
       tag
       |> Enum.filter(fn data -> data["type"] == "Hashtag" and data["name"] end)
-      |> Enum.map(fn %{"name" => name} ->
+      |> Enum.map(fn
-        name
+        %{"name" => "#" <> hashtag} -> String.downcase(hashtag)
-        |> String.slice(1..-1)
+        %{"name" => hashtag} -> String.downcase(hashtag)
-        |> String.downcase()
       end)
 
     Map.put(object, "tag", tag ++ tags)

lib/pleroma/web/feed/feed_view.ex

@@ -32,6 +32,7 @@ def prepare_activity(activity, opts \\ []) do
 
     %{
       activity: activity,
+      object: object,
       data: Map.get(object, :data),
       actor: actor
     }

lib/pleroma/web/mastodon_api/controllers/timeline_controller.ex

@@ -133,31 +133,22 @@ defp fail_on_bad_auth(conn) do
   end
 
   defp hashtag_fetching(params, user, local_only) do
-    tags =
+    # Note: not sanitizing tag options at this stage (may be mix-cased, have duplicates etc.)
+    tags_any =
       [params[:tag], params[:any]]
       |> List.flatten()
-      |> Enum.uniq()
+      |> Enum.filter(& &1)
-      |> Enum.reject(&is_nil/1)
-      |> Enum.map(&String.downcase/1)
 
-    tag_all =
+    tag_all = Map.get(params, :all, [])
-      params
+    tag_reject = Map.get(params, :none, [])
-      |> Map.get(:all, [])
-      |> Enum.map(&String.downcase/1)
 
-    tag_reject =
-      params
-      |> Map.get(:none, [])
-      |> Enum.map(&String.downcase/1)
-
-    _activities =
     params
     |> Map.put(:type, "Create")
     |> Map.put(:local_only, local_only)
     |> Map.put(:blocking_user, user)
     |> Map.put(:muting_user, user)
     |> Map.put(:user, user)
-      |> Map.put(:tag, tags)
+    |> Map.put(:tag, tags_any)
     |> Map.put(:tag_all, tag_all)
     |> Map.put(:tag_reject, tag_reject)
     |> ActivityPub.fetch_public_activities()

lib/pleroma/web/mastodon_api/views/status_view.ex

@@ -198,8 +198,10 @@ def render("show.json", %{activity: %{data: %{"object" => _object}} = activity}
     like_count = object.data["like_count"] || 0
     announcement_count = object.data["announcement_count"] || 0
 
-    tags = object.data["tag"] || []
+    hashtags = Object.hashtags(object)
-    sensitive = object.data["sensitive"] || Enum.member?(tags, "nsfw")
+    sensitive = object.data["sensitive"] || Enum.member?(hashtags, "nsfw")
+
+    tags = Object.tags(object)
 
     tag_mentions =
       tags

lib/pleroma/web/templates/feed/feed/_activity.atom.eex

@@ -22,7 +22,7 @@
     <link type="text/html" href='<%= @data["external_url"] %>' rel="alternate"/>
   <% end %>
 
-  <%= for tag <- @data["tag"] || [] do %>
+  <%= for tag <- Pleroma.Object.hashtags(@object) do %>
     <category term="<%= tag %>"></category>
   <% end %>
 

lib/pleroma/web/templates/feed/feed/_activity.rss.eex

@@ -22,7 +22,7 @@
 
   <link rel="ostatus:conversation"><%= activity_context(@activity) %></link>
 
-  <%= for tag <- @data["tag"] || [] do %>
+  <%= for tag <- Pleroma.Object.hashtags(@object) do %>
     <category term="<%= tag %>"></category>
   <% end %>
 

lib/pleroma/web/templates/feed/feed/_tag_activity.atom.eex

@@ -41,7 +41,7 @@
       <% end %>
     <% end %>
 
-    <%= for tag <- @data["tag"] || [] do %>
+    <%= for tag <- Pleroma.Object.hashtags(@object) do %>
       <category term="<%= tag %>"></category>
     <% end %>
 

mix.exs

@@ -121,6 +121,7 @@ defp deps do
       {:phoenix_pubsub, "~> 2.0"},
       {:phoenix_ecto, "~> 4.0"},
       {:ecto_enum, "~> 1.4"},
+      {:ecto_explain, "~> 0.1.2"},
       {:ecto_sql, "~> 3.4.4"},
       {:postgrex, ">= 0.15.5"},
       {:oban, "~> 2.3.4"},

mix.lock

@@ -31,6 +31,7 @@
   "earmark_parser": {:hex, :earmark_parser, "1.4.10", "6603d7a603b9c18d3d20db69921527f82ef09990885ed7525003c7fe7dc86c56", [:mix], [], "hexpm", "8e2d5370b732385db2c9b22215c3f59c84ac7dda7ed7e544d7c459496ae519c0"},
   "ecto": {:hex, :ecto, "3.4.6", "08f7afad3257d6eb8613309af31037e16c36808dfda5a3cd0cb4e9738db030e4", [:mix], [{:decimal, "~> 1.6 or ~> 2.0", [hex: :decimal, repo: "hexpm", optional: false]}, {:jason, "~> 1.0", [hex: :jason, repo: "hexpm", optional: true]}, {:telemetry, "~> 0.4", [hex: :telemetry, repo: "hexpm", optional: false]}], "hexpm", "6f13a9e2a62e75c2dcfc7207bfc65645ab387af8360db4c89fee8b5a4bf3f70b"},
   "ecto_enum": {:hex, :ecto_enum, "1.4.0", "d14b00e04b974afc69c251632d1e49594d899067ee2b376277efd8233027aec8", [:mix], [{:ecto, ">= 3.0.0", [hex: :ecto, repo: "hexpm", optional: false]}, {:ecto_sql, "> 3.0.0", [hex: :ecto_sql, repo: "hexpm", optional: false]}, {:mariaex, ">= 0.0.0", [hex: :mariaex, repo: "hexpm", optional: true]}, {:postgrex, ">= 0.0.0", [hex: :postgrex, repo: "hexpm", optional: true]}], "hexpm", "8fb55c087181c2b15eee406519dc22578fa60dd82c088be376d0010172764ee4"},
+  "ecto_explain": {:hex, :ecto_explain, "0.1.2", "a9d504cbd4adc809911f796d5ef7ebb17a576a6d32286c3d464c015bd39d5541", [:mix], [], "hexpm", "1d0e7798ae30ecf4ce34e912e5354a0c1c832b7ebceba39298270b9a9f316330"},
   "ecto_sql": {:hex, :ecto_sql, "3.4.5", "30161f81b167d561a9a2df4329c10ae05ff36eca7ccc84628f2c8b9fa1e43323", [:mix], [{:db_connection, "~> 2.2", [hex: :db_connection, repo: "hexpm", optional: false]}, {:ecto, "~> 3.4.3", [hex: :ecto, repo: "hexpm", optional: false]}, {:myxql, "~> 0.3.0 or ~> 0.4.0", [hex: :myxql, repo: "hexpm", optional: true]}, {:postgrex, "~> 0.15.0", [hex: :postgrex, repo: "hexpm", optional: true]}, {:tds, "~> 2.1.0", [hex: :tds, repo: "hexpm", optional: true]}, {:telemetry, "~> 0.4.0", [hex: :telemetry, repo: "hexpm", optional: false]}], "hexpm", "31990c6a3579b36a3c0841d34a94c275e727de8b84f58509da5f1b2032c98ac2"},
   "eimp": {:hex, :eimp, "1.0.14", "fc297f0c7e2700457a95a60c7010a5f1dcb768a083b6d53f49cd94ab95a28f22", [:rebar3], [{:p1_utils, "1.0.18", [hex: :p1_utils, repo: "hexpm", optional: false]}], "hexpm", "501133f3112079b92d9e22da8b88bf4f0e13d4d67ae9c15c42c30bd25ceb83b6"},
   "elixir_make": {:hex, :elixir_make, "0.6.2", "7dffacd77dec4c37b39af867cedaabb0b59f6a871f89722c25b28fcd4bd70530", [:mix], [], "hexpm", "03e49eadda22526a7e5279d53321d1cced6552f344ba4e03e619063de75348d9"},

priv/repo/migrations/20190711042021_create_safe_jsonb_set.exs

@@ -9,7 +9,7 @@ def change do
     begin
       result := jsonb_set(target, path, coalesce(new_value, 'null'::jsonb), create_missing);
       if result is NULL then
-        raise 'jsonb_set tried to wipe the object, please report this incindent to Pleroma bug tracker. https://git.pleroma.social/pleroma/pleroma/issues/new';
+        raise 'jsonb_set tried to wipe the object, please report this incident to Pleroma bug tracker. https://git.pleroma.social/pleroma/pleroma/issues/new';
         return target;
       else
         return result;

priv/repo/migrations/20201221202251_create_hashtags.exs

@@ -0,0 +1,13 @@
+defmodule Pleroma.Repo.Migrations.CreateHashtags do
+  use Ecto.Migration
+
+  def change do
+    create_if_not_exists table(:hashtags) do
+      add(:name, :citext, null: false)
+
+      timestamps()
+    end
+
+    create_if_not_exists(unique_index(:hashtags, [:name]))
+  end
+end

priv/repo/migrations/20201221202252_remove_data_from_hashtags.exs

@@ -0,0 +1,15 @@
+defmodule Pleroma.Repo.Migrations.RemoveDataFromHashtags do
+  use Ecto.Migration
+
+  def up do
+    alter table(:hashtags) do
+      remove_if_exists(:data, :map)
+    end
+  end
+
+  def down do
+    alter table(:hashtags) do
+      add_if_not_exists(:data, :map, default: %{})
+    end
+  end
+end

priv/repo/migrations/20201221203824_create_hashtags_objects.exs

@@ -0,0 +1,13 @@
+defmodule Pleroma.Repo.Migrations.CreateHashtagsObjects do
+  use Ecto.Migration
+
+  def change do
+    create_if_not_exists table(:hashtags_objects, primary_key: false) do
+      add(:hashtag_id, references(:hashtags), null: false, primary_key: true)
+      add(:object_id, references(:objects), null: false, primary_key: true)
+    end
+
+    # Note: PK index: "hashtags_objects_pkey" PRIMARY KEY, btree (hashtag_id, object_id)
+    create_if_not_exists(index(:hashtags_objects, [:object_id]))
+  end
+end

priv/repo/migrations/20210105195018_create_data_migrations.exs

@@ -0,0 +1,17 @@
+defmodule Pleroma.Repo.Migrations.CreateDataMigrations do
+  use Ecto.Migration
+
+  def change do
+    create_if_not_exists table(:data_migrations) do
+      add(:name, :string, null: false)
+      add(:state, :integer, default: 1)
+      add(:feature_lock, :boolean, default: false)
+      add(:params, :map, default: %{})
+      add(:data, :map, default: %{})
+
+      timestamps()
+    end
+
+    create_if_not_exists(unique_index(:data_migrations, [:name]))
+  end
+end

priv/repo/migrations/20210106183301_data_migration_create_populate_hashtags_table.exs

@@ -0,0 +1,16 @@
+defmodule Pleroma.Repo.Migrations.DataMigrationCreatePopulateHashtagsTable do
+  use Ecto.Migration
+
+  def up do
+    dt = NaiveDateTime.utc_now()
+
+    execute(
+      "INSERT INTO data_migrations(name, inserted_at, updated_at) " <>
+        "VALUES ('populate_hashtags_table', '#{dt}', '#{dt}') ON CONFLICT DO NOTHING;"
+    )
+  end
+
+  def down do
+    execute("DELETE FROM data_migrations WHERE name = 'populate_hashtags_table';")
+  end
+end

priv/repo/migrations/20210111172254_create_data_migration_failed_ids.exs

@@ -0,0 +1,14 @@
+defmodule Pleroma.Repo.Migrations.CreateDataMigrationFailedIds do
+  use Ecto.Migration
+
+  def change do
+    create_if_not_exists table(:data_migration_failed_ids, primary_key: false) do
+      add(:data_migration_id, references(:data_migrations), null: false, primary_key: true)
+      add(:record_id, :bigint, null: false, primary_key: true)
+    end
+
+    create_if_not_exists(
+      unique_index(:data_migration_failed_ids, [:data_migration_id, :record_id])
+    )
+  end
+end

priv/repo/migrations/20210222183840_remove_hashtags_objects_duplicate_index.exs

@@ -0,0 +1,11 @@
+defmodule Pleroma.Repo.Migrations.RemoveHashtagsObjectsDuplicateIndex do
+  use Ecto.Migration
+
+  @moduledoc "Removes `hashtags_objects_hashtag_id_object_id_index` index (duplicate of PK index)."
+
+  def up do
+    drop_if_exists(unique_index(:hashtags_objects, [:hashtag_id, :object_id]))
+  end
+
+  def down, do: nil
+end

priv/repo/migrations/20210222184616_change_hashtags_name_to_text.exs

@@ -0,0 +1,15 @@
+defmodule Pleroma.Repo.Migrations.ChangeHashtagsNameToText do
+  use Ecto.Migration
+
+  def up do
+    alter table(:hashtags) do
+      modify(:name, :text)
+    end
+  end
+
+  def down do
+    alter table(:hashtags) do
+      modify(:name, :citext)
+    end
+  end
+end

test/pleroma/activity/ir/topics_test.exs

@@ -11,6 +11,8 @@ defmodule Pleroma.Activity.Ir.TopicsTest do
 
   require Pleroma.Constants
 
+  import Mock
+
   describe "poll answer" do
     test "produce no topics" do
       activity = %Activity{object: %Object{data: %{"type" => "Answer"}}}
@@ -77,15 +79,14 @@ test "with no attachments doesn't produce public:media topics", %{activity: acti
       refute Enum.member?(topics, "public:local:media")
     end
 
-    test "converts tags to hash tags", %{activity: %{object: %{data: data} = object} = activity} do
+    test "converts tags to hash tags", %{activity: activity} do
-      tagged_data = Map.put(data, "tag", ["foo", "bar"])
+      with_mock(Object, [:passthrough], hashtags: fn _ -> ["foo", "bar"] end) do
-      activity = %{activity | object: %{object | data: tagged_data}}
-
         topics = Topics.get_activity_topics(activity)
 
         assert Enum.member?(topics, "hashtag:foo")
         assert Enum.member?(topics, "hashtag:bar")
       end
+    end
 
     test "only converts strings to hash tags", %{
       activity: %{object: %{data: data} = object} = activity

test/pleroma/hashtag_test.exs

@@ -0,0 +1,17 @@
+# Pleroma: A lightweight social networking server
+# Copyright © 2017-2021 Pleroma Authors <https://pleroma.social/>
+# SPDX-License-Identifier: AGPL-3.0-only
+
+defmodule Pleroma.HashtagTest do
+  use Pleroma.DataCase
+
+  alias Pleroma.Hashtag
+
+  describe "changeset validations" do
+    test "ensure non-blank :name" do
+      changeset = Hashtag.changeset(%Hashtag{}, %{name: ""})
+
+      assert {:name, {"can't be blank", [validation: :required]}} in changeset.errors
+    end
+  end
+end

test/pleroma/object_test.exs

@@ -5,10 +5,13 @@
 defmodule Pleroma.ObjectTest do
   use Pleroma.DataCase
   use Oban.Testing, repo: Pleroma.Repo
+
   import ExUnit.CaptureLog
   import Pleroma.Factory
   import Tesla.Mock
+
   alias Pleroma.Activity
+  alias Pleroma.Hashtag
   alias Pleroma.Object
   alias Pleroma.Repo
   alias Pleroma.Tests.ObanHelpers
@@ -417,4 +420,28 @@ test "preserves internal fields on refetch", %{mock_modified: mock_modified} do
       assert updated_object.data["like_count"] == 1
     end
   end
+
+  describe ":hashtags association" do
+    test "Hashtag records are created with Object record and updated on its change" do
+      user = insert(:user)
+
+      {:ok, %{object: object}} =
+        CommonAPI.post(user, %{status: "some text #hashtag1 #hashtag2 ..."})
+
+      assert [%Hashtag{name: "hashtag1"}, %Hashtag{name: "hashtag2"}] =
+               Enum.sort_by(object.hashtags, & &1.name)
+
+      {:ok, object} = Object.update_data(object, %{"tag" => []})
+
+      assert [] = object.hashtags
+
+      object = Object.get_by_id(object.id) |> Repo.preload(:hashtags)
+      assert [] = object.hashtags
+
+      {:ok, object} = Object.update_data(object, %{"tag" => ["abc", "def"]})
+
+      assert [%Hashtag{name: "abc"}, %Hashtag{name: "def"}] =
+               Enum.sort_by(object.hashtags, & &1.name)
+    end
+  end
 end

test/pleroma/web/activity_pub/activity_pub_test.exs

@@ -213,18 +213,24 @@ test "works for guppe actors" do
   test "it fetches the appropriate tag-restricted posts" do
     user = insert(:user)
 
-    {:ok, status_one} = CommonAPI.post(user, %{status: ". #test"})
+    {:ok, status_one} = CommonAPI.post(user, %{status: ". #TEST"})
     {:ok, status_two} = CommonAPI.post(user, %{status: ". #essais"})
-    {:ok, status_three} = CommonAPI.post(user, %{status: ". #test #reject"})
+    {:ok, status_three} = CommonAPI.post(user, %{status: ". #test #Reject"})
+
+    {:ok, status_four} = CommonAPI.post(user, %{status: ". #Any1 #any2"})
+    {:ok, status_five} = CommonAPI.post(user, %{status: ". #Any2 #any1"})
+
+    for hashtag_timeline_strategy <- [:enabled, :disabled] do
+      clear_config([:features, :improved_hashtag_timeline], hashtag_timeline_strategy)
 
       fetch_one = ActivityPub.fetch_activities([], %{type: "Create", tag: "test"})
 
-    fetch_two = ActivityPub.fetch_activities([], %{type: "Create", tag: ["test", "essais"]})
+      fetch_two = ActivityPub.fetch_activities([], %{type: "Create", tag: ["TEST", "essais"]})
 
       fetch_three =
         ActivityPub.fetch_activities([], %{
           type: "Create",
-        tag: ["test", "essais"],
+          tag: ["test", "Essais"],
           tag_reject: ["reject"]
         })
 
@@ -232,13 +238,39 @@ test "it fetches the appropriate tag-restricted posts" do
         ActivityPub.fetch_activities([], %{
           type: "Create",
           tag: ["test"],
-        tag_all: ["test", "reject"]
+          tag_all: ["test", "REJECT"]
         })
 
+      # Testing that deduplication (if needed) is done on DB (not Ecto) level; :limit is important
+      fetch_five =
+        ActivityPub.fetch_activities([], %{
+          type: "Create",
+          tag: ["ANY1", "any2"],
+          limit: 2
+        })
+
+      fetch_six =
+        ActivityPub.fetch_activities([], %{
+          type: "Create",
+          tag: ["any1", "Any2"],
+          tag_all: [],
+          tag_reject: []
+        })
+
+      # Regression test: passing empty lists as filter options shouldn't affect the results
+      assert fetch_five == fetch_six
+
+      [fetch_one, fetch_two, fetch_three, fetch_four, fetch_five] =
+        Enum.map([fetch_one, fetch_two, fetch_three, fetch_four, fetch_five], fn statuses ->
+          Enum.map(statuses, fn s -> Repo.preload(s, object: :hashtags) end)
+        end)
+
       assert fetch_one == [status_one, status_three]
       assert fetch_two == [status_one, status_two, status_three]
       assert fetch_three == [status_one, status_two]
       assert fetch_four == [status_three]
+      assert fetch_five == [status_four, status_five]
+    end
   end
 
   describe "insertion" do

test/pleroma/web/activity_pub/transmogrifier/note_handling_test.exs

@@ -39,7 +39,8 @@ test "it works for incoming notices with tag not being an array (kroeg)" do
       {:ok, %Activity{data: data, local: false}} = Transmogrifier.handle_incoming(data)
       object = Object.normalize(data["object"], fetch: false)
 
-      assert "test" in object.data["tag"]
+      assert "test" in Object.tags(object)
+      assert Object.hashtags(object) == ["test"]
     end
 
     test "it cleans up incoming notices which are not really DMs" do
@@ -220,7 +221,8 @@ test "it works for incoming notices with hashtags" do
       {:ok, %Activity{data: data, local: false}} = Transmogrifier.handle_incoming(data)
       object = Object.normalize(data["object"], fetch: false)
 
-      assert Enum.at(object.data["tag"], 2) == "moo"
+      assert Enum.at(Object.tags(object), 2) == "moo"
+      assert Object.hashtags(object) == ["moo"]
     end
 
     test "it works for incoming notices with contentMap" do

test/pleroma/web/common_api_test.exs

@@ -493,7 +493,7 @@ test "it de-duplicates tags" do
 
     object = Object.normalize(activity, fetch: false)
 
-    assert object.data["tag"] == ["2hu"]
+    assert Object.tags(object) == ["2hu"]
   end
 
   test "it adds emoji in the object" do

test/pleroma/web/mastodon_api/views/status_view_test.exs

@@ -262,8 +262,8 @@ test "a note activity" do
       mentions: [],
       tags: [
         %{
-          name: "#{object_data["tag"]}",
+          name: "#{hd(object_data["tag"])}",
-          url: "http://localhost:4001/tag/#{object_data["tag"]}"
+          url: "http://localhost:4001/tag/#{hd(object_data["tag"])}"
         }
       ],
       application: nil,