1
0
mirror of https://github.com/mastodon/mastodon synced 2024-12-15 15:18:14 +09:00
mastodon/app/lib/vacuum/statuses_vacuum.rb

48 lines
1.2 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
class Vacuum::StatusesVacuum
include Redisable
def initialize(retention_period)
@retention_period = retention_period
end
def perform
refactor(vacuum statuses): reduce amount of db queries and load for each query - improve performance (#21487) * refactor(statuses_vacuum): remove dead code - unused Method is not called inside class and private. Clean up dead code. * refactor(statuses_vacuum): make retention_period present test explicit This private method only hides functionality. It is best practice to be as explicit as possible. * refactor(statuses_vacuum): improve query performance - fix statuses_scope having sub-select for Account.remote scope by `joins(:account).merge(Account.remote)` - fix statuses_scope unnecessary use of `Status.arel_table[:id].lt` because it is inexplicit, bad practice and even slower than normal `.where('statuses.id < ?'` - fix statuses_scope remove select(:id, :visibility) for having reusable active record query batches (no re queries) - fix vacuum_statuses! to use in_batches instead of find_in_batches, because in_batches delivers a full blown active record query result, in stead of an array - no requeries necessary - send(:unlink_from_conversations) not to perform another db query, but reuse the in_batches result instead. - remove now obsolete remove_from_account_conversations method - remove_from_search_index uses array of ids, instead of mapping the ids from an array - this should be more efficient - use the in_batches scope to call delete_all, instead of running another db query for this - because it is again more efficient - add TODO comment for calling models private method with send * refactor(status): simplify unlink_from_conversations - add `has_many through:` relation mentioned_accounts - use model scope local instead of method call `Status#local?` - more readable add account to inbox_owners when account.local? * refactor(status): searchable_by way less sub selects These queries all included a sub-select. Doing the same with a joins should be more efficient. Since this method does 5 such queries, this should be significant, since it technically halves the query count. This is how it was: ```ruby [3] pry(main)> Status.first.mentions.where(account: Account.local, silent: false).explain Status Load (1.6ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (1.5ms) SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ``` This is how it is with this change: ```ruby [4] pry(main)> Status.first.mentions.joins(:account).merge(Account.local).active.explain Status Load (1.7ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (0.7ms) SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ```
2022-11-28 04:41:18 +09:00
vacuum_statuses! if @retention_period.present?
end
private
def vacuum_statuses!
refactor(vacuum statuses): reduce amount of db queries and load for each query - improve performance (#21487) * refactor(statuses_vacuum): remove dead code - unused Method is not called inside class and private. Clean up dead code. * refactor(statuses_vacuum): make retention_period present test explicit This private method only hides functionality. It is best practice to be as explicit as possible. * refactor(statuses_vacuum): improve query performance - fix statuses_scope having sub-select for Account.remote scope by `joins(:account).merge(Account.remote)` - fix statuses_scope unnecessary use of `Status.arel_table[:id].lt` because it is inexplicit, bad practice and even slower than normal `.where('statuses.id < ?'` - fix statuses_scope remove select(:id, :visibility) for having reusable active record query batches (no re queries) - fix vacuum_statuses! to use in_batches instead of find_in_batches, because in_batches delivers a full blown active record query result, in stead of an array - no requeries necessary - send(:unlink_from_conversations) not to perform another db query, but reuse the in_batches result instead. - remove now obsolete remove_from_account_conversations method - remove_from_search_index uses array of ids, instead of mapping the ids from an array - this should be more efficient - use the in_batches scope to call delete_all, instead of running another db query for this - because it is again more efficient - add TODO comment for calling models private method with send * refactor(status): simplify unlink_from_conversations - add `has_many through:` relation mentioned_accounts - use model scope local instead of method call `Status#local?` - more readable add account to inbox_owners when account.local? * refactor(status): searchable_by way less sub selects These queries all included a sub-select. Doing the same with a joins should be more efficient. Since this method does 5 such queries, this should be significant, since it technically halves the query count. This is how it was: ```ruby [3] pry(main)> Status.first.mentions.where(account: Account.local, silent: false).explain Status Load (1.6ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (1.5ms) SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ``` This is how it is with this change: ```ruby [4] pry(main)> Status.first.mentions.joins(:account).merge(Account.local).active.explain Status Load (1.7ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (0.7ms) SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ```
2022-11-28 04:41:18 +09:00
statuses_scope.in_batches do |statuses|
# Side-effects not covered by foreign keys, such
# as the search index, must be handled first.
refactor(vacuum statuses): reduce amount of db queries and load for each query - improve performance (#21487) * refactor(statuses_vacuum): remove dead code - unused Method is not called inside class and private. Clean up dead code. * refactor(statuses_vacuum): make retention_period present test explicit This private method only hides functionality. It is best practice to be as explicit as possible. * refactor(statuses_vacuum): improve query performance - fix statuses_scope having sub-select for Account.remote scope by `joins(:account).merge(Account.remote)` - fix statuses_scope unnecessary use of `Status.arel_table[:id].lt` because it is inexplicit, bad practice and even slower than normal `.where('statuses.id < ?'` - fix statuses_scope remove select(:id, :visibility) for having reusable active record query batches (no re queries) - fix vacuum_statuses! to use in_batches instead of find_in_batches, because in_batches delivers a full blown active record query result, in stead of an array - no requeries necessary - send(:unlink_from_conversations) not to perform another db query, but reuse the in_batches result instead. - remove now obsolete remove_from_account_conversations method - remove_from_search_index uses array of ids, instead of mapping the ids from an array - this should be more efficient - use the in_batches scope to call delete_all, instead of running another db query for this - because it is again more efficient - add TODO comment for calling models private method with send * refactor(status): simplify unlink_from_conversations - add `has_many through:` relation mentioned_accounts - use model scope local instead of method call `Status#local?` - more readable add account to inbox_owners when account.local? * refactor(status): searchable_by way less sub selects These queries all included a sub-select. Doing the same with a joins should be more efficient. Since this method does 5 such queries, this should be significant, since it technically halves the query count. This is how it was: ```ruby [3] pry(main)> Status.first.mentions.where(account: Account.local, silent: false).explain Status Load (1.6ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (1.5ms) SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ``` This is how it is with this change: ```ruby [4] pry(main)> Status.first.mentions.joins(:account).merge(Account.local).active.explain Status Load (1.7ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (0.7ms) SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ```
2022-11-28 04:41:18 +09:00
statuses.direct_visibility
.includes(mentions: :account)
.find_each(&:unlink_from_conversations!)
if Chewy.enabled?
remove_from_index(statuses.ids, 'chewy:queue:StatusesIndex')
remove_from_index(statuses.ids, 'chewy:queue:PublicStatusesIndex')
end
refactor(vacuum statuses): reduce amount of db queries and load for each query - improve performance (#21487) * refactor(statuses_vacuum): remove dead code - unused Method is not called inside class and private. Clean up dead code. * refactor(statuses_vacuum): make retention_period present test explicit This private method only hides functionality. It is best practice to be as explicit as possible. * refactor(statuses_vacuum): improve query performance - fix statuses_scope having sub-select for Account.remote scope by `joins(:account).merge(Account.remote)` - fix statuses_scope unnecessary use of `Status.arel_table[:id].lt` because it is inexplicit, bad practice and even slower than normal `.where('statuses.id < ?'` - fix statuses_scope remove select(:id, :visibility) for having reusable active record query batches (no re queries) - fix vacuum_statuses! to use in_batches instead of find_in_batches, because in_batches delivers a full blown active record query result, in stead of an array - no requeries necessary - send(:unlink_from_conversations) not to perform another db query, but reuse the in_batches result instead. - remove now obsolete remove_from_account_conversations method - remove_from_search_index uses array of ids, instead of mapping the ids from an array - this should be more efficient - use the in_batches scope to call delete_all, instead of running another db query for this - because it is again more efficient - add TODO comment for calling models private method with send * refactor(status): simplify unlink_from_conversations - add `has_many through:` relation mentioned_accounts - use model scope local instead of method call `Status#local?` - more readable add account to inbox_owners when account.local? * refactor(status): searchable_by way less sub selects These queries all included a sub-select. Doing the same with a joins should be more efficient. Since this method does 5 such queries, this should be significant, since it technically halves the query count. This is how it was: ```ruby [3] pry(main)> Status.first.mentions.where(account: Account.local, silent: false).explain Status Load (1.6ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (1.5ms) SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ``` This is how it is with this change: ```ruby [4] pry(main)> Status.first.mentions.joins(:account).merge(Account.local).active.explain Status Load (1.7ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (0.7ms) SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ```
2022-11-28 04:41:18 +09:00
# Foreign keys take care of most associated records for us.
# Media attachments will be orphaned.
statuses.delete_all
end
end
def statuses_scope
refactor(vacuum statuses): reduce amount of db queries and load for each query - improve performance (#21487) * refactor(statuses_vacuum): remove dead code - unused Method is not called inside class and private. Clean up dead code. * refactor(statuses_vacuum): make retention_period present test explicit This private method only hides functionality. It is best practice to be as explicit as possible. * refactor(statuses_vacuum): improve query performance - fix statuses_scope having sub-select for Account.remote scope by `joins(:account).merge(Account.remote)` - fix statuses_scope unnecessary use of `Status.arel_table[:id].lt` because it is inexplicit, bad practice and even slower than normal `.where('statuses.id < ?'` - fix statuses_scope remove select(:id, :visibility) for having reusable active record query batches (no re queries) - fix vacuum_statuses! to use in_batches instead of find_in_batches, because in_batches delivers a full blown active record query result, in stead of an array - no requeries necessary - send(:unlink_from_conversations) not to perform another db query, but reuse the in_batches result instead. - remove now obsolete remove_from_account_conversations method - remove_from_search_index uses array of ids, instead of mapping the ids from an array - this should be more efficient - use the in_batches scope to call delete_all, instead of running another db query for this - because it is again more efficient - add TODO comment for calling models private method with send * refactor(status): simplify unlink_from_conversations - add `has_many through:` relation mentioned_accounts - use model scope local instead of method call `Status#local?` - more readable add account to inbox_owners when account.local? * refactor(status): searchable_by way less sub selects These queries all included a sub-select. Doing the same with a joins should be more efficient. Since this method does 5 such queries, this should be significant, since it technically halves the query count. This is how it was: ```ruby [3] pry(main)> Status.first.mentions.where(account: Account.local, silent: false).explain Status Load (1.6ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (1.5ms) SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" WHERE "mentions"."status_id" = $1 AND "mentions"."account_id" IN (SELECT "accounts"."id" FROM "accounts" WHERE "accounts"."domain" IS NULL) AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ``` This is how it is with this change: ```ruby [4] pry(main)> Status.first.mentions.joins(:account).merge(Account.local).active.explain Status Load (1.7ms) SELECT "statuses".* FROM "statuses" WHERE "statuses"."deleted_at" IS NULL ORDER BY "statuses"."id" DESC LIMIT $1 [["LIMIT", 1]] Mention Load (0.7ms) SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] => EXPLAIN for: SELECT "mentions".* FROM "mentions" INNER JOIN "accounts" ON "accounts"."id" = "mentions"."account_id" WHERE "mentions"."status_id" = $1 AND "accounts"."domain" IS NULL AND "mentions"."silent" = $2 [["status_id", 109382923142288414], ["silent", false]] QUERY PLAN ------------------------------------------------------------------------------------------------------------------ Nested Loop (cost=0.15..23.08 rows=1 width=41) -> Seq Scan on accounts (cost=0.00..10.90 rows=1 width=8) Filter: (domain IS NULL) -> Index Scan using index_mentions_on_account_id_and_status_id on mentions (cost=0.15..8.17 rows=1 width=41) Index Cond: ((account_id = accounts.id) AND (status_id = '109382923142288414'::bigint)) Filter: (NOT silent) (6 rows) ```
2022-11-28 04:41:18 +09:00
Status.unscoped.kept
.joins(:account).merge(Account.remote)
.where('statuses.id < ?', retention_period_as_id)
end
def retention_period_as_id
Mastodon::Snowflake.id_at(@retention_period.ago, with_random: false)
end
def remove_from_index(status_ids, index)
with_redis { |redis| redis.sadd(index, status_ids) }
end
end