Merge branch 'tc-geo-prune-events-correctly-ce' into 'master'

Add Gitlab::Database::Subquery.self_join to delete_all with limit

See merge request gitlab-org/gitlab-ce!21839

doc/development/README.md

@@ -94,6 +94,7 @@ description: 'Learn how to contribute to GitLab.'
 - [Verifying database capabilities](verifying_database_capabilities.md)
 - [Database Debugging and Troubleshooting](database_debugging.md)
 - [Query Count Limits](query_count_limits.md)
+- [Database helper modules](database_helpers.md)
 
 ## Testing guides
 

doc/development/database_helpers.md

+# Database helpers
+
+There are a number of useful helper modules defined in `/lib/gitlab/database/`.
+
+## Subquery
+
+In some cases it is not possible to perform an operation on a query.
+For example:
+
+```ruby
+Geo::EventLog.where('id < 100').limit(10).delete_all
+```
+
+Will give this error:
+
+> ActiveRecord::ActiveRecordError: delete_all doesn't support limit
+
+One solution would be to wrap it in another `where`:
+
+```ruby
+Geo::EventLog.where(id: Geo::EventLog.where('id < 100').limit(10)).delete_all
+```
+
+This works with PostgreSQL, but with MySQL it gives this error:
+
+> ActiveRecord::StatementInvalid: Mysql2::Error: This version of MySQL
+> doesn't yet support 'LIMIT & IN/ALL/ANY/SOME subquery'
+
+Also, that query doesn't have very good performance. Using a
+`INNER JOIN` with itself is better.
+
+So instead of this query:
+
+```sql
+SELECT geo_event_log.*
+FROM geo_event_log
+WHERE geo_event_log.id IN
+    (SELECT geo_event_log.id
+     FROM geo_event_log
+     WHERE (id < 100)
+     LIMIT 10)
+```
+
+It's better to write:
+
+```sql
+SELECT geo_event_log.*
+FROM geo_event_log
+INNER JOIN
+  (SELECT geo_event_log.*
+   FROM geo_event_log
+   WHERE (id < 100)
+   LIMIT 10) t2 ON geo_event_log.id = t2.id
+```
+
+And this is where `Gitlab::Database::Subquery.self_join` can help
+you. So you can rewrite the above statement as:
+
+```ruby
+Gitlab::Database::Subquery.self_join(Geo::EventLog.where('id < 100').limit(10)).delete_all
+```
+
+And this also works with MySQL, so you don't need to worry about that.

lib/gitlab/database/subquery.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module Database
+    module Subquery
+      class << self
+        def self_join(relation)
+          t = relation.arel_table
+          t2 = relation.arel.as('t2')
+
+          relation.unscoped.joins(t.join(t2).on(t[:id].eq(t2[:id])).join_sources.first)
+        end
+      end
+    end
+  end
+end

spec/lib/gitlab/database/subquery_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+RSpec.describe Gitlab::Database::Subquery do
+  describe '.self_join' do
+    set(:project) { create(:project) }
+
+    it 'allows you to delete_all rows with WHERE and LIMIT' do
+      events = create_list(:event, 8, project: project)
+
+      expect do
+        described_class.self_join(Event.where('id < ?', events[5]).recent.limit(2)).delete_all
+      end.to change { Event.count }.by(-2)
+    end
+  end
+end