Merge branch 'loose-fk-multi-partition-query' into 'master'

Query all partitions when loading loose FK records

See merge request gitlab-org/gitlab!74072

app/models/loose_foreign_keys/deleted_record.rb

@@ -4,29 +4,29 @@ class LooseForeignKeys::DeletedRecord < ApplicationRecord
   self.primary_key = :id
 
   scope :for_table, -> (table) { where(fully_qualified_table_name: table) }
-  scope :ordered_by_id, -> { order(:id, :primary_key_value) }
-  # This needs to be parameterized once we start adding partitions
-  scope :for_partition, -> { where(partition: 1) }
+  scope :consume_order, -> { order(:partition, :consume_after, :id) }
 
   enum status: { pending: 1, processed: 2 }, _prefix: :status
 
   def self.load_batch_for_table(table, batch_size)
     for_table(table)
-      .for_partition
       .status_pending
-      .ordered_by_id
+      .consume_order
       .limit(batch_size)
       .to_a
   end
 
-  def self.mark_records_processed_for_table_between(table, from_record, to_record)
-    from = from_record.id
-    to = to_record.id
+  def self.mark_records_processed(all_records)
+    # Run a query for each partition to optimize the row lookup by primary key (partition, id)
+    update_count = 0
 
-    for_table(table)
-      .for_partition
-      .status_pending
-      .where(id: from..to)
+    all_records.group_by(&:partition).each do |partition, records_within_partition|
+      update_count += status_pending
+        .where(partition: partition)
+        .where(id: records_within_partition.pluck(:id))
         .update_all(status: :processed)
     end
+
+    update_count
+  end
 end

app/services/loose_foreign_keys/batch_cleaner_service.rb

@@ -25,8 +25,7 @@ module LooseForeignKeys
       return if modification_tracker.over_limit?
 
       # At this point, all associations are cleaned up, we can update the status of the parent records
-      update_count = LooseForeignKeys::DeletedRecord
-        .mark_records_processed_for_table_between(deleted_parent_records.first.fully_qualified_table_name, deleted_parent_records.first, deleted_parent_records.last)
+      update_count = LooseForeignKeys::DeletedRecord.mark_records_processed(deleted_parent_records)
 
       deleted_records_counter.increment({ table: parent_klass.table_name, db_config_name: LooseForeignKeys::DeletedRecord.connection.pool.db_config.name }, update_count)
     end

db/migrate/20211109100050_add_consume_after_to_loose_fk_deleted_records.rb

+# frozen_string_literal: true
+
+class AddConsumeAfterToLooseFkDeletedRecords < Gitlab::Database::Migration[1.0]
+  enable_lock_retries!
+
+  def up
+    add_column :loose_foreign_keys_deleted_records, :consume_after, :datetime_with_timezone, default: -> { 'NOW()' }
+  end
+
+  def down
+    remove_column :loose_foreign_keys_deleted_records, :consume_after
+  end
+end

db/migrate/20211109101010_support_partition_query_in_loose_fk_table.rb

+# frozen_string_literal: true
+
+class SupportPartitionQueryInLooseFkTable < Gitlab::Database::Migration[1.0]
+  include Gitlab::Database::PartitioningMigrationHelpers
+
+  disable_ddl_transaction!
+
+  INDEX_NAME = 'index_loose_foreign_keys_deleted_records_for_partitioned_query'
+
+  def up
+    add_concurrent_partitioned_index :loose_foreign_keys_deleted_records,
+      %I[partition fully_qualified_table_name consume_after id],
+      where: 'status = 1',
+      name: INDEX_NAME
+  end
+
+  def down
+    remove_concurrent_partitioned_index_by_name :loose_foreign_keys_deleted_records, INDEX_NAME
+  end
+end

db/post_migrate/20211109112454_drop_old_loose_fk_deleted_records_index.rb

+# frozen_string_literal: true
+
+class DropOldLooseFkDeletedRecordsIndex < Gitlab::Database::Migration[1.0]
+  include Gitlab::Database::PartitioningMigrationHelpers
+
+  disable_ddl_transaction!
+
+  INDEX_NAME = 'index_loose_foreign_keys_deleted_records_for_loading_records'
+
+  def up
+    remove_concurrent_partitioned_index_by_name :loose_foreign_keys_deleted_records, INDEX_NAME
+  end
+
+  def down
+    add_concurrent_partitioned_index :loose_foreign_keys_deleted_records,
+      %I[fully_qualified_table_name id primary_key_value partition],
+      where: 'status = 1',
+      name: INDEX_NAME
+  end
+end

db/schema_migrations/20211109100050

+86aa6ad1759a00c2cc5cb6dc2e381aead2910a24f0e37933a5e72af56d08101a
\ No newline at end of file

db/schema_migrations/20211109101010

+30eb98b8fdb24bc5de357b0ec14a6b92d520db025c82bd7b9448f71542c7d7e3
\ No newline at end of file

db/schema_migrations/20211109112454

+1bc48cdae55eea5a5963edd3a138d7d6859afa6caafe0b793c553fdfabe9f488
\ No newline at end of file

db/structure.sql

@@ -1021,6 +1021,7 @@ CREATE TABLE loose_foreign_keys_deleted_records (
     status smallint DEFAULT 1 NOT NULL,
     created_at timestamp with time zone DEFAULT now() NOT NULL,
     fully_qualified_table_name text NOT NULL,
+    consume_after timestamp with time zone DEFAULT now(),
     CONSTRAINT check_1a541f3235 CHECK ((char_length(fully_qualified_table_name) <= 150))
 )
 PARTITION BY LIST (partition);
@@ -1041,6 +1042,7 @@ CREATE TABLE gitlab_partitions_static.loose_foreign_keys_deleted_records_1 (
     status smallint DEFAULT 1 NOT NULL,
     created_at timestamp with time zone DEFAULT now() NOT NULL,
     fully_qualified_table_name text NOT NULL,
+    consume_after timestamp with time zone DEFAULT now(),
     CONSTRAINT check_1a541f3235 CHECK ((char_length(fully_qualified_table_name) <= 150))
 );
 ALTER TABLE ONLY loose_foreign_keys_deleted_records ATTACH PARTITION gitlab_partitions_static.loose_foreign_keys_deleted_records_1 FOR VALUES IN ('1');
@@ -24073,6 +24075,10 @@ CREATE INDEX index_merge_request_stage_events_project_duration ON ONLY analytics
 
 CREATE INDEX index_006f943df6 ON gitlab_partitions_static.analytics_cycle_analytics_merge_request_stage_events_16 USING btree (stage_event_hash_id, project_id, end_event_timestamp, merge_request_id, start_event_timestamp) WHERE (end_event_timestamp IS NOT NULL);
 
+CREATE INDEX index_loose_foreign_keys_deleted_records_for_partitioned_query ON ONLY loose_foreign_keys_deleted_records USING btree (partition, fully_qualified_table_name, consume_after, id) WHERE (status = 1);
+
+CREATE INDEX index_01e3390fac ON gitlab_partitions_static.loose_foreign_keys_deleted_records_1 USING btree (partition, fully_qualified_table_name, consume_after, id) WHERE (status = 1);
+
 CREATE INDEX index_02749b504c ON gitlab_partitions_static.analytics_cycle_analytics_merge_request_stage_events_11 USING btree (stage_event_hash_id, project_id, end_event_timestamp, merge_request_id, start_event_timestamp) WHERE (end_event_timestamp IS NOT NULL);
 
 CREATE INDEX index_merge_request_stage_events_group_duration ON ONLY analytics_cycle_analytics_merge_request_stage_events USING btree (stage_event_hash_id, group_id, end_event_timestamp, merge_request_id, start_event_timestamp) WHERE (end_event_timestamp IS NOT NULL);
@@ -24391,10 +24397,6 @@ CREATE INDEX index_8a0fc3de4f ON gitlab_partitions_static.analytics_cycle_analyt
 
 CREATE INDEX index_8b9f9a19a4 ON gitlab_partitions_static.analytics_cycle_analytics_merge_request_stage_events_18 USING btree (stage_event_hash_id, group_id, end_event_timestamp, merge_request_id, start_event_timestamp) WHERE (end_event_timestamp IS NOT NULL);
 
-CREATE INDEX index_loose_foreign_keys_deleted_records_for_loading_records ON ONLY loose_foreign_keys_deleted_records USING btree (fully_qualified_table_name, id, primary_key_value, partition) WHERE (status = 1);
-
-CREATE INDEX index_8be8640437 ON gitlab_partitions_static.loose_foreign_keys_deleted_records_1 USING btree (fully_qualified_table_name, id, primary_key_value, partition) WHERE (status = 1);
-
 CREATE INDEX index_8fb48e72ce ON gitlab_partitions_static.analytics_cycle_analytics_issue_stage_events_26 USING btree (stage_event_hash_id, group_id, end_event_timestamp, issue_id, start_event_timestamp) WHERE (end_event_timestamp IS NOT NULL);
 
 CREATE INDEX index_9201b952a0 ON gitlab_partitions_static.analytics_cycle_analytics_issue_stage_events_13 USING btree (stage_event_hash_id, group_id, end_event_timestamp, issue_id, start_event_timestamp) WHERE (end_event_timestamp IS NOT NULL);
@@ -27911,6 +27913,8 @@ ALTER INDEX index_issue_stage_events_project_duration ATTACH PARTITION gitlab_pa
 
 ALTER INDEX index_merge_request_stage_events_project_duration ATTACH PARTITION gitlab_partitions_static.index_006f943df6;
 
+ALTER INDEX index_loose_foreign_keys_deleted_records_for_partitioned_query ATTACH PARTITION gitlab_partitions_static.index_01e3390fac;
+
 ALTER INDEX index_merge_request_stage_events_project_duration ATTACH PARTITION gitlab_partitions_static.index_02749b504c;
 
 ALTER INDEX index_merge_request_stage_events_group_duration ATTACH PARTITION gitlab_partitions_static.index_0287f5ba09;
@@ -28217,8 +28221,6 @@ ALTER INDEX index_issue_stage_events_project_in_progress_duration ATTACH PARTITI
 
 ALTER INDEX index_merge_request_stage_events_group_duration ATTACH PARTITION gitlab_partitions_static.index_8b9f9a19a4;
 
-ALTER INDEX index_loose_foreign_keys_deleted_records_for_loading_records ATTACH PARTITION gitlab_partitions_static.index_8be8640437;
-
 ALTER INDEX index_issue_stage_events_group_duration ATTACH PARTITION gitlab_partitions_static.index_8fb48e72ce;
 
 ALTER INDEX index_issue_stage_events_group_duration ATTACH PARTITION gitlab_partitions_static.index_9201b952a0;

spec/models/loose_foreign_keys/deleted_record_spec.rb

@@ -24,26 +24,9 @@ RSpec.describe LooseForeignKeys::DeletedRecord, type: :model do
     end
   end
 
-  describe '.mark_records_processed_for_table_between' do
-    it 'marks processed exactly one record' do
-      described_class.mark_records_processed_for_table_between(table, deleted_record_2, deleted_record_2)
-
-      expect(described_class.status_pending.count).to eq(3)
-      expect(described_class.status_processed.count).to eq(1)
-
-      processed_record = described_class.status_processed.first
-      expect(processed_record).to eq(deleted_record_2)
-    end
-
-    it 'deletes two records' do
-      described_class.mark_records_processed_for_table_between(table, deleted_record_2, deleted_record_4)
-
-      expect(described_class.status_pending.count).to eq(2)
-      expect(described_class.status_processed.count).to eq(2)
-    end
-
-    it 'deletes all records' do
-      described_class.mark_records_processed_for_table_between(table, deleted_record_1, deleted_record_4)
+  describe '.mark_records_processed' do
+    it 'updates all records' do
+      described_class.mark_records_processed([deleted_record_1, deleted_record_2, deleted_record_4])
 
       expect(described_class.status_pending.count).to eq(1)
       expect(described_class.status_processed.count).to eq(3)