Merge branch...

Merge branch '38358-update-migration-helpers-to-use-check-constraints-instead-of-change-column-null' into 'master'

Update migration helpers to use check constraints instead of change_column_null

See merge request gitlab-org/gitlab!30943

doc/development/what_requires_downtime.md

@@ -171,8 +171,39 @@ Adding or removing a NOT NULL clause (or another constraint) can typically be
 done without requiring downtime. However, this does require that any application
 changes are deployed _first_. Thus, changing the constraints of a column should
 happen in a post-deployment migration.
-NOTE: Avoid using `change_column` as it produces inefficient query because it re-defines
-the whole column type. For example, to add a NOT NULL constraint, prefer `change_column_null`
+
+NOTE: Avoid using `change_column` as it produces an inefficient query because it re-defines
+the whole column type.
+
+To add a NOT NULL constraint, use the `add_not_null_constraint` migration helper:
+
+```ruby
+# A post-deployment migration in db/post_migrate
+class AddNotNull < ActiveRecord::Migration[4.2]
+  include Gitlab::Database::MigrationHelpers
+
+  disable_ddl_transaction!
+
+  def up
+    add_not_null_constraint :users, :username
+  end
+
+  def down
+    remove_not_null_constraint :users, :username
+  end
+end
+```
+
+If the column to be updated requires cleaning first (e.g. there are `NULL` values), you should:
+
+1. Add the `NOT NULL` constraint with `validate: false`
+
+   `add_not_null_constraint :users, :username, validate: false`
+
+1. Clean up the data with a data migration
+1. Validate the `NOT NULL` constraint with a followup migration
+
+   `validate_not_null_constraint :users, :username`
 
 ## Changing Column Types
 

lib/gitlab/database/migration_helpers.rb

@@ -265,6 +265,12 @@ module Gitlab
       # or `RESET ALL` is executed
       def disable_statement_timeout
         if block_given?
+          if statement_timeout_disabled?
+            # Don't do anything if the statement_timeout is already disabled
+            # Allows for nested calls of disable_statement_timeout without
+            # resetting the timeout too early (before the outer call ends)
+            yield
+          else
             begin
               execute('SET statement_timeout TO 0')
 
@@ -272,6 +278,7 @@ module Gitlab
             ensure
               execute('RESET ALL')
             end
+          end
         else
           unless transaction_open?
             raise <<~ERROR
@@ -495,7 +502,7 @@ module Gitlab
               update_column_in_batches(table, column, default_after_type_cast, &block)
             end
 
-            change_column_null(table, column, false) unless allow_null
+            add_not_null_constraint(table, column) unless allow_null
           # We want to rescue _all_ exceptions here, even those that don't inherit
           # from StandardError.
           rescue Exception => error # rubocop: disable all
@@ -1334,12 +1341,73 @@ into similar problems in the future (e.g. when new tables are created).
         check_constraint_exists?(table, text_limit_name(table, column, name: constraint_name))
       end
 
+      # Migration Helpers for managing not null constraints
+      def add_not_null_constraint(table, column, constraint_name: nil, validate: true)
+        if column_is_nullable?(table, column)
+          add_check_constraint(
+            table,
+            "#{column} IS NOT NULL",
+            not_null_constraint_name(table, column, name: constraint_name),
+            validate: validate
+          )
+        else
+          warning_message = <<~MESSAGE
+            NOT NULL check constraint was not created:
+            column #{table}.#{column} is already defined as `NOT NULL`
+          MESSAGE
+
+          Rails.logger.warn warning_message
+        end
+      end
+
+      def validate_not_null_constraint(table, column, constraint_name: nil)
+        validate_check_constraint(
+          table,
+          not_null_constraint_name(table, column, name: constraint_name)
+        )
+      end
+
+      def remove_not_null_constraint(table, column, constraint_name: nil)
+        remove_check_constraint(
+          table,
+          not_null_constraint_name(table, column, name: constraint_name)
+        )
+      end
+
+      def check_not_null_constraint_exists?(table, column, constraint_name: nil)
+        check_constraint_exists?(
+          table,
+          not_null_constraint_name(table, column, name: constraint_name)
+        )
+      end
+
       private
 
+      def statement_timeout_disabled?
+        # This is a string of the form "100ms" or "0" when disabled
+        connection.select_value('SHOW statement_timeout') == "0"
+      end
+
+      def column_is_nullable?(table, column)
+        # Check if table.column has not been defined with NOT NULL
+        check_sql = <<~SQL
+          SELECT c.is_nullable
+          FROM information_schema.columns c
+          WHERE c.table_name = '#{table}'
+          AND c.column_name = '#{column}'
+        SQL
+
+        connection.select_value(check_sql) == 'YES'
+      end
+
       def text_limit_name(table, column, name: nil)
         name.presence || check_constraint_name(table, column, 'max_length')
       end
 
+      def not_null_constraint_name(table, column, name: nil)
+        name.presence || check_constraint_name(table, column, 'not_null')
+      end
+
       def missing_schema_object_message(table, type, name)
         <<~MESSAGE
           Could not find #{type} "#{name}" on table "#{table}" which was referenced during the migration.
@@ -1383,7 +1451,7 @@ into similar problems in the future (e.g. when new tables are created).
 
         update_column_in_batches(table, new, Arel::Table.new(table)[old], batch_column_name: batch_column_name)
 
-        change_column_null(table, new, false) unless old_col.null
+        add_not_null_constraint(table, new) unless old_col.null
 
         copy_indexes(table, old, new)
         copy_foreign_keys(table, old, new)

spec/support/shared_examples/lib/gitlab/migration_helpers_shared_examples.rb

@@ -13,10 +13,11 @@ end
 RSpec.shared_examples 'performs validation' do |validation_option|
   it 'performs validation' do
     expect(model).to receive(:disable_statement_timeout).and_call_original
+    expect(model).to receive(:statement_timeout_disabled?).and_return(false)
     expect(model).to receive(:execute).with(/statement_timeout/)
     expect(model).to receive(:execute).ordered.with(/NOT VALID/)
     expect(model).to receive(:execute).ordered.with(/VALIDATE CONSTRAINT/)
-    expect(model).to receive(:execute).with(/RESET ALL/)
+    expect(model).to receive(:execute).ordered.with(/RESET ALL/)
 
     model.add_concurrent_foreign_key(*args, **options.merge(validation_option))
   end