gearheads/mastodon
Archived
2
0
Fork 0
Code Releases 4 Activity

refactor(vacuum statuses): reduce amount of db queries and load for each query - improve performance (#21487)

Browse source 
* 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)
```
This commit is contained in:
Kaspar V 2022-11-27 20:41:18 +01:00 • committed by GitHub
parent 625216d8e1
commit 47f0d7021e
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 25 additions and 31 deletions
Download patch file Download diff file Expand all files Collapse all files

41
app/lib/vacuum/statuses_vacuum.rb
View file

@ -8,47 +8,40 @@ class Vacuum::StatusesVacuum
end
def perform
vacuum_statuses! if retention_period?
vacuum_statuses! if @retention_period.present?
end
private
def vacuum_statuses!
statuses_scope.find_in_batches do |statuses|
statuses_scope.in_batches do |statuses|
# Side-effects not covered by foreign keys, such
# as the search index, must be handled first.
statuses.direct_visibility
.includes(mentions: :account)
.find_each do |status|
# TODO: replace temporary solution - call of private model method
status.send(:unlink_from_conversations)
end
remove_from_search_index(statuses.ids) if Chewy.enabled?
remove_from_account_conversations(statuses)
remove_from_search_index(statuses)
# Foreign keys take care of most associated records
# for us. Media attachments will be orphaned.
Status.where(id: statuses.map(&:id)).delete_all
# Foreign keys take care of most associated records for us.
# Media attachments will be orphaned.
statuses.delete_all
end
end
def statuses_scope
Status.unscoped.kept.where(account: Account.remote).where(Status.arel_table[:id].lt(retention_period_as_id)).select(:id, :visibility)
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 analyze_statuses!
ActiveRecord::Base.connection.execute('ANALYZE statuses')
end
def remove_from_account_conversations(statuses)
Status.where(id: statuses.select(&:direct_visibility?).map(&:id)).includes(:account, mentions: :account).each(&:unlink_from_conversations)
end
def remove_from_search_index(statuses)
with_redis { |redis| redis.sadd('chewy:queue:StatusesIndex', statuses.map(&:id)) } if Chewy.enabled?
end
def retention_period?
@retention_period.present?
def remove_from_search_index(status_ids)
with_redis { |redis| redis.sadd('chewy:queue:StatusesIndex', status_ids) }
end
end