Fix statement timeouts in RemoveRestrictedTodos migration

On GitLab.com, the RemoveRestrictedTodos background migration
encountered about 700+ failures a day due to statement timeouts.

PostgreSQL might perform badly with a LIMIT 1 because the planner is
guessing that scanning the index in ID order will come across the
desired row in less time it will take the planner than using another
index. The order_hint does not affect the search results. For example,
`ORDER BY id ASC, updated_at ASC` means the same thing as `ORDER BY id
ASC`.

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/52649

app/models/concerns/each_batch.rb

@@ -39,7 +39,15 @@ module EachBatch
     #
     # of - The number of rows to retrieve per batch.
     # column - The column to use for ordering the batches.
-    def each_batch(of: 1000, column: primary_key)
+    # order_hint - An optional column to append to the `ORDER BY id`
+    #   clause to help the query planner. PostgreSQL might perform badly
+    #   with a LIMIT 1 because the planner is guessing that scanning the
+    #   index in ID order will come across the desired row in less time
+    #   it will take the planner than using another index. The
+    #   order_hint does not affect the search results. For example,
+    #   `ORDER BY id ASC, updated_at ASC` means the same thing as `ORDER
+    #   BY id ASC`.
+    def each_batch(of: 1000, column: primary_key, order_hint: nil)
       unless column
         raise ArgumentError,
           'the column: argument must be set to a column name to use for ordering rows'
@@ -48,7 +56,9 @@ module EachBatch
       start = except(:select)
         .select(column)
         .reorder(column => :asc)
-        .take
+
+      start = start.order(order_hint) if order_hint
+      start = start.take
 
       return unless start
 
@@ -60,6 +70,9 @@ module EachBatch
           .select(column)
           .where(arel_table[column].gteq(start_id))
           .reorder(column => :asc)
+
+        stop = stop.order(order_hint) if order_hint
+        stop = stop
           .offset(of)
           .limit(1)
           .take

changelogs/unreleased/sh-fix-issue-52649.yml

+---
+title: Fix statement timeouts in RemoveRestrictedTodos migration
+merge_request: 22795
+author:
+type: other

lib/gitlab/background_migration/remove_restricted_todos.rb

@@ -67,7 +67,7 @@ module Gitlab
           .where('access_level >= ?', 20)
 
         confidential_issues = Issue.select(:id, :author_id).where(confidential: true, project_id: project_id)
-        confidential_issues.each_batch(of: 100) do |batch|
+        confidential_issues.each_batch(of: 100, order_hint: :confidential) do |batch|
           batch.each do |issue|
             assigned_users = IssueAssignee.select(:user_id).where(issue_id: issue.id)
 

spec/models/concerns/each_batch_spec.rb

@@ -14,14 +14,15 @@ describe EachBatch do
       5.times { create(:user, updated_at: 1.day.ago) }
     end
 
+    shared_examples 'each_batch handling' do |kwargs|
       it 'yields an ActiveRecord::Relation when a block is given' do
-      model.each_batch do |relation|
+        model.each_batch(kwargs) do |relation|
           expect(relation).to be_a_kind_of(ActiveRecord::Relation)
         end
       end
 
       it 'yields a batch index as the second argument' do
-      model.each_batch do |_, index|
+        model.each_batch(kwargs) do |_, index|
           expect(index).to eq(1)
         end
       end
@@ -29,7 +30,7 @@ describe EachBatch do
       it 'accepts a custom batch size' do
         amount = 0
 
-      model.each_batch(of: 1) { amount += 1 }
+        model.each_batch(kwargs.merge({ of: 1 })) { amount += 1 }
 
         expect(amount).to eq(5)
       end
@@ -50,4 +51,8 @@ describe EachBatch do
         expect(model.where(updated_at: time).count).to eq(5)
       end
     end
+
+    it_behaves_like 'each_batch handling', {}
+    it_behaves_like 'each_batch handling', { order_hint: :updated_at }
+  end
 end