Add feature to automatically delete old toots (#16529)

* Add account statuses cleanup policy model

* Record last inspected toot to delete to speed up successive calls to statuses_to_delete

* Add service to cleanup a given account's statuses within a budget

* Add worker to go through account policies and delete old toots

* Fix last inspected status id logic

All existing statuses older or equal to last inspected status id must be
kept by the current policy. This is an invariant that must be kept so that
resuming deletion from the last inspected status remains sound.

* Add tests

* Refactor scheduler and add tests

* Add user interface

* Add support for discriminating based on boosts/favs

* Add UI support for min_reblogs and min_favs, rework UI

* Address first round of review comments

* Replace Snowflake#id_at_start with with_random parameter

* Add tests

* Add tests for StatusesCleanupController

* Rework settings page

* Adjust load-avoiding mechanisms

* Please CodeClimate
This commit is contained in:
Claire
2021-08-09 23:11:50 +02:00
committed by GitHub
parent 432e3d1eaf
commit 4ac78e2a06
21 changed files with 1287 additions and 3 deletions

View File

@ -0,0 +1,35 @@
# frozen_string_literal: true
class StatusesCleanupController < ApplicationController
layout 'admin'
before_action :authenticate_user!
before_action :set_policy
before_action :set_body_classes
def show; end
def update
if @policy.update(resource_params)
redirect_to statuses_cleanup_path, notice: I18n.t('generic.changes_saved_msg')
else
render action: :show
end
rescue ActionController::ParameterMissing
# Do nothing
end
private
def set_policy
@policy = current_account.statuses_cleanup_policy || current_account.build_statuses_cleanup_policy(enabled: false)
end
def resource_params
params.require(:account_statuses_cleanup_policy).permit(:enabled, :min_status_age, :keep_direct, :keep_pinned, :keep_polls, :keep_media, :keep_self_fav, :keep_self_bookmark, :min_favs, :min_reblogs)
end
def set_body_classes
@body_classes = 'admin'
end
end

View File

@ -0,0 +1,171 @@
# frozen_string_literal: true
# == Schema Information
#
# Table name: account_statuses_cleanup_policies
#
# id :bigint not null, primary key
# account_id :bigint not null
# enabled :boolean default(TRUE), not null
# min_status_age :integer default(1209600), not null
# keep_direct :boolean default(TRUE), not null
# keep_pinned :boolean default(TRUE), not null
# keep_polls :boolean default(FALSE), not null
# keep_media :boolean default(FALSE), not null
# keep_self_fav :boolean default(TRUE), not null
# keep_self_bookmark :boolean default(TRUE), not null
# min_favs :integer
# min_reblogs :integer
# created_at :datetime not null
# updated_at :datetime not null
#
class AccountStatusesCleanupPolicy < ApplicationRecord
include Redisable
ALLOWED_MIN_STATUS_AGE = [
2.weeks.seconds,
1.month.seconds,
2.months.seconds,
3.months.seconds,
6.months.seconds,
1.year.seconds,
2.years.seconds,
].freeze
EXCEPTION_BOOLS = %w(keep_direct keep_pinned keep_polls keep_media keep_self_fav keep_self_bookmark).freeze
EXCEPTION_THRESHOLDS = %w(min_favs min_reblogs).freeze
# Depending on the cleanup policy, the query to discover the next
# statuses to delete my get expensive if the account has a lot of old
# statuses otherwise excluded from deletion by the other exceptions.
#
# Therefore, `EARLY_SEARCH_CUTOFF` is meant to be the maximum number of
# old statuses to be considered for deletion prior to checking exceptions.
#
# This is used in `compute_cutoff_id` to provide a `max_id` to
# `statuses_to_delete`.
EARLY_SEARCH_CUTOFF = 5_000
belongs_to :account
validates :min_status_age, inclusion: { in: ALLOWED_MIN_STATUS_AGE }
validates :min_favs, numericality: { greater_than_or_equal_to: 1, allow_nil: true }
validates :min_reblogs, numericality: { greater_than_or_equal_to: 1, allow_nil: true }
validate :validate_local_account
before_save :update_last_inspected
def statuses_to_delete(limit = 50, max_id = nil, min_id = nil)
scope = account.statuses
scope.merge!(old_enough_scope(max_id))
scope = scope.where(Status.arel_table[:id].gteq(min_id)) if min_id.present?
scope.merge!(without_popular_scope) unless min_favs.nil? && min_reblogs.nil?
scope.merge!(without_direct_scope) if keep_direct?
scope.merge!(without_pinned_scope) if keep_pinned?
scope.merge!(without_poll_scope) if keep_polls?
scope.merge!(without_media_scope) if keep_media?
scope.merge!(without_self_fav_scope) if keep_self_fav?
scope.merge!(without_self_bookmark_scope) if keep_self_bookmark?
scope.reorder(id: :asc).limit(limit)
end
# This computes a toot id such that:
# - the toot would be old enough to be candidate for deletion
# - there are at most EARLY_SEARCH_CUTOFF toots between the last inspected toot and this one
#
# The idea is to limit expensive SQL queries when an account has lots of toots excluded from
# deletion, while not starting anew on each run.
def compute_cutoff_id
min_id = last_inspected || 0
max_id = Mastodon::Snowflake.id_at(min_status_age.seconds.ago, with_random: false)
subquery = account.statuses.where(Status.arel_table[:id].gteq(min_id)).where(Status.arel_table[:id].lteq(max_id))
subquery = subquery.select(:id).reorder(id: :asc).limit(EARLY_SEARCH_CUTOFF)
# We're textually interpolating a subquery here as ActiveRecord seem to not provide
# a way to apply the limit to the subquery
Status.connection.execute("SELECT MAX(id) FROM (#{subquery.to_sql}) t").values.first.first
end
# The most important thing about `last_inspected` is that any toot older than it is guaranteed
# not to be kept by the policy regardless of its age.
def record_last_inspected(last_id)
redis.set("account_cleanup:#{account.id}", last_id, ex: 1.week.seconds)
end
def last_inspected
redis.get("account_cleanup:#{account.id}")&.to_i
end
def invalidate_last_inspected(status, action)
last_value = last_inspected
return if last_value.nil? || status.id > last_value || status.account_id != account_id
case action
when :unbookmark
return unless keep_self_bookmark?
when :unfav
return unless keep_self_fav?
when :unpin
return unless keep_pinned?
end
record_last_inspected(status.id)
end
private
def update_last_inspected
if EXCEPTION_BOOLS.map { |name| attribute_change_to_be_saved(name) }.compact.include?([true, false])
# Policy has been widened in such a way that any previously-inspected status
# may need to be deleted, so we'll have to start again.
redis.del("account_cleanup:#{account.id}")
end
if EXCEPTION_THRESHOLDS.map { |name| attribute_change_to_be_saved(name) }.compact.any? { |old, new| old.present? && (new.nil? || new > old) }
redis.del("account_cleanup:#{account.id}")
end
end
def validate_local_account
errors.add(:account, :invalid) unless account&.local?
end
def without_direct_scope
Status.where.not(visibility: :direct)
end
def old_enough_scope(max_id = nil)
# Filtering on `id` rather than `min_status_age` ago will treat
# non-snowflake statuses as older than they really are, but Mastodon
# has switched to snowflake IDs significantly over 2 years ago anyway.
max_id = [max_id, Mastodon::Snowflake.id_at(min_status_age.seconds.ago, with_random: false)].compact.min
Status.where(Status.arel_table[:id].lteq(max_id))
end
def without_self_fav_scope
Status.where('NOT EXISTS (SELECT * FROM favourites fav WHERE fav.account_id = statuses.account_id AND fav.status_id = statuses.id)')
end
def without_self_bookmark_scope
Status.where('NOT EXISTS (SELECT * FROM bookmarks bookmark WHERE bookmark.account_id = statuses.account_id AND bookmark.status_id = statuses.id)')
end
def without_pinned_scope
Status.where('NOT EXISTS (SELECT * FROM status_pins pin WHERE pin.account_id = statuses.account_id AND pin.status_id = statuses.id)')
end
def without_media_scope
Status.where('NOT EXISTS (SELECT * FROM media_attachments media WHERE media.status_id = statuses.id)')
end
def without_poll_scope
Status.where(poll_id: nil)
end
def without_popular_scope
scope = Status.left_joins(:status_stat)
scope = scope.where('COALESCE(status_stats.reblogs_count, 0) <= ?', min_reblogs) unless min_reblogs.nil?
scope = scope.where('COALESCE(status_stats.favourites_count, 0) <= ?', min_favs) unless min_favs.nil?
scope
end
end

View File

@ -23,4 +23,12 @@ class Bookmark < ApplicationRecord
before_validation do
self.status = status.reblog if status&.reblog?
end
after_destroy :invalidate_cleanup_info
def invalidate_cleanup_info
return unless status&.account_id == account_id && account.local?
account.statuses_cleanup_policy&.invalidate_last_inspected(status, :unbookmark)
end
end

View File

@ -66,5 +66,8 @@ module AccountAssociations
# Follow recommendations
has_one :follow_recommendation_suppression, inverse_of: :account, dependent: :destroy
# Account statuses cleanup policy
has_one :statuses_cleanup_policy, class_name: 'AccountStatusesCleanupPolicy', inverse_of: :account, dependent: :destroy
end
end

View File

@ -28,6 +28,7 @@ class Favourite < ApplicationRecord
after_create :increment_cache_counters
after_destroy :decrement_cache_counters
after_destroy :invalidate_cleanup_info
private
@ -39,4 +40,10 @@ class Favourite < ApplicationRecord
return if association(:status).loaded? && status.marked_for_destruction?
status&.decrement_count!(:favourites_count)
end
def invalidate_cleanup_info
return unless status&.account_id == account_id && account.local?
account.statuses_cleanup_policy&.invalidate_last_inspected(status, :unfav)
end
end

View File

@ -15,4 +15,12 @@ class StatusPin < ApplicationRecord
belongs_to :status
validates_with StatusPinValidator
after_destroy :invalidate_cleanup_info
def invalidate_cleanup_info
return unless status&.account_id == account_id && account.local?
account.statuses_cleanup_policy&.invalidate_last_inspected(status, :unpin)
end
end

View File

@ -0,0 +1,27 @@
# frozen_string_literal: true
class AccountStatusesCleanupService < BaseService
# @param [AccountStatusesCleanupPolicy] account_policy
# @param [Integer] budget
# @return [Integer]
def call(account_policy, budget = 50)
return 0 unless account_policy.enabled?
cutoff_id = account_policy.compute_cutoff_id
return 0 if cutoff_id.blank?
num_deleted = 0
last_deleted = nil
account_policy.statuses_to_delete(budget, cutoff_id, account_policy.last_inspected).reorder(nil).find_each(order: :asc) do |status|
status.discard
RemovalWorker.perform_async(status.id, redraft: false)
num_deleted += 1
last_deleted = status.id
end
account_policy.record_last_inspected(last_deleted.presence || cutoff_id)
num_deleted
end
end

View File

@ -0,0 +1,45 @@
- content_for :page_title do
= t('settings.statuses_cleanup')
- content_for :heading_actions do
= button_tag t('generic.save_changes'), class: 'button', form: 'edit_policy'
= simple_form_for @policy, url: statuses_cleanup_path, method: :put, html: { id: 'edit_policy' } do |f|
.fields-row
.fields-row__column.fields-row__column-6.fields-group
= f.input :enabled, as: :boolean, wrapper: :with_label, label: t('statuses_cleanup.enabled'), hint: t('statuses_cleanup.enabled_hint')
.fields-row__column.fields-row__column-6.fields-group
= f.input :min_status_age, wrapper: :with_label, label: t('statuses_cleanup.min_age_label'), collection: AccountStatusesCleanupPolicy::ALLOWED_MIN_STATUS_AGE.map(&:to_i), label_method: lambda { |i| t("statuses_cleanup.min_age.#{i}") }, include_blank: false, hint: false
.flash-message= t('statuses_cleanup.explanation')
%h4= t('statuses_cleanup.exceptions')
.fields-row
.fields-row__column.fields-row__column-6.fields-group
= f.input :keep_pinned, wrapper: :with_label, label: t('statuses_cleanup.keep_pinned'), hint: t('statuses_cleanup.keep_pinned_hint')
.fields-row__column.fields-row__column-6.fields-group
= f.input :keep_direct, wrapper: :with_label, label: t('statuses_cleanup.keep_direct'), hint: t('statuses_cleanup.keep_direct_hint')
.fields-row
.fields-row__column.fields-row__column-6.fields-group
= f.input :keep_self_fav, wrapper: :with_label, label: t('statuses_cleanup.keep_self_fav'), hint: t('statuses_cleanup.keep_self_fav_hint')
.fields-row__column.fields-row__column-6.fields-group
= f.input :keep_self_bookmark, wrapper: :with_label, label: t('statuses_cleanup.keep_self_bookmark'), hint: t('statuses_cleanup.keep_self_bookmark_hint')
.fields-row
.fields-row__column.fields-row__column-6.fields-group
= f.input :keep_polls, wrapper: :with_label, label: t('statuses_cleanup.keep_polls'), hint: t('statuses_cleanup.keep_polls_hint')
.fields-row__column.fields-row__column-6.fields-group
= f.input :keep_media, wrapper: :with_label, label: t('statuses_cleanup.keep_media'), hint: t('statuses_cleanup.keep_media_hint')
%h4= t('statuses_cleanup.interaction_exceptions')
.fields-row
.fields-row__column.fields-row__column-6.fields-group
= f.input :min_favs, wrapper: :with_label, label: t('statuses_cleanup.min_favs'), hint: t('statuses_cleanup.min_favs_hint'), input_html: { min: 1, placeholder: t('statuses_cleanup.ignore_favs') }
.fields-row__column.fields-row__column-6.fields-group
= f.input :min_reblogs, wrapper: :with_label, label: t('statuses_cleanup.min_reblogs'), hint: t('statuses_cleanup.min_reblogs_hint'), input_html: { min: 1, placeholder: t('statuses_cleanup.ignore_reblogs') }
.flash-message= t('statuses_cleanup.interaction_exceptions_explanation')

View File

@ -0,0 +1,96 @@
# frozen_string_literal: true
class Scheduler::AccountsStatusesCleanupScheduler
include Sidekiq::Worker
# This limit is mostly to be nice to the fediverse at large and not
# generate too much traffic.
# This also helps limiting the running time of the scheduler itself.
MAX_BUDGET = 50
# This is an attempt to spread the load across instances, as various
# accounts are likely to have various followers.
PER_ACCOUNT_BUDGET = 5
# This is an attempt to limit the workload generated by status removal
# jobs to something the particular instance can handle.
PER_THREAD_BUDGET = 5
# Those avoid loading an instance that is already under load
MAX_DEFAULT_SIZE = 2
MAX_DEFAULT_LATENCY = 5
MAX_PUSH_SIZE = 5
MAX_PUSH_LATENCY = 10
# 'pull' queue has lower priority jobs, and it's unlikely that pushing
# deletes would cause much issues with this queue if it didn't cause issues
# with default and push. Yet, do not enqueue deletes if the instance is
# lagging behind too much.
MAX_PULL_SIZE = 500
MAX_PULL_LATENCY = 300
# This is less of an issue in general, but deleting old statuses is likely
# to cause delivery errors, and thus increase the number of jobs to be retried.
# This doesn't directly translate to load, but connection errors and a high
# number of dead instances may lead to this spiraling out of control if
# unchecked.
MAX_RETRY_SIZE = 50_000
sidekiq_options retry: 0, lock: :until_executed
def perform
return if under_load?
budget = compute_budget
first_policy_id = last_processed_id
loop do
num_processed_accounts = 0
scope = AccountStatusesCleanupPolicy.where(enabled: true)
scope.where(Account.arel_table[:id].gt(first_policy_id)) if first_policy_id.present?
scope.find_each(order: :asc) do |policy|
num_deleted = AccountStatusesCleanupService.new.call(policy, [budget, PER_ACCOUNT_BUDGET].min)
num_processed_accounts += 1 unless num_deleted.zero?
budget -= num_deleted
if budget.zero?
save_last_processed_id(policy.id)
break
end
end
# The idea here is to loop through all policies at least once until the budget is exhausted
# and start back after the last processed account otherwise
break if budget.zero? || (num_processed_accounts.zero? && first_policy_id.nil?)
first_policy_id = nil
end
end
def compute_budget
threads = Sidekiq::ProcessSet.new.filter { |x| x['queues'].include?('push') }.map { |x| x['concurrency'] }.sum
[PER_THREAD_BUDGET * threads, MAX_BUDGET].min
end
def under_load?
return true if Sidekiq::Stats.new.retry_size > MAX_RETRY_SIZE
queue_under_load?('default', MAX_DEFAULT_SIZE, MAX_DEFAULT_LATENCY) || queue_under_load?('push', MAX_PUSH_SIZE, MAX_PUSH_LATENCY) || queue_under_load?('pull', MAX_PULL_SIZE, MAX_PULL_LATENCY)
end
private
def queue_under_load?(name, max_size, max_latency)
queue = Sidekiq::Queue.new(name)
queue.size > max_size || queue.latency > max_latency
end
def last_processed_id
Redis.current.get('account_statuses_cleanup_scheduler:last_account_id')
end
def save_last_processed_id(id)
if id.nil?
Redis.current.del('account_statuses_cleanup_scheduler:last_account_id')
else
Redis.current.set('account_statuses_cleanup_scheduler:last_account_id', id, ex: 1.hour.seconds)
end
end
end