Setup sliding partitioning strategy for LFK

app/models/loose_foreign_keys/deleted_record.rb

 # frozen_string_literal: true
 
 class LooseForeignKeys::DeletedRecord < ApplicationRecord
+  PARTITION_DURATION = 1.day
+
+  include PartitionedTable
+
   self.primary_key = :id
+  self.ignored_columns = %i[partition]
+
+  partitioned_by :partition, strategy: :sliding_list,
+    next_partition_if: -> (active_partition) do
+      return false if Feature.disabled?(:lfk_automatic_partition_creation, default_enabled: :yaml)
+
+      oldest_record_in_partition = LooseForeignKeys::DeletedRecord
+        .select(:id, :created_at)
+        .for_partition(active_partition)
+        .order(:id)
+        .limit(1)
+        .take
+
+      oldest_record_in_partition.present? && oldest_record_in_partition.created_at < PARTITION_DURATION.ago
+    end,
+    detach_partition_if: -> (partition) do
+      return false if Feature.disabled?(:lfk_automatic_partition_dropping, default_enabled: :yaml)
+
+      !LooseForeignKeys::DeletedRecord
+        .for_partition(partition)
+        .status_pending
+        .exists?
+    end
 
   scope :for_table, -> (table) { where(fully_qualified_table_name: table) }
+  scope :for_partition, -> (partition) { where(partition: partition) }
   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)
+    # selecting partition as partition_number to workaround the sliding partitioning column ignore
+    select(arel_table[Arel.star], arel_table[:partition].as('partition_number'))
+      .for_table(table)
       .status_pending
       .consume_order
       .limit(batch_size)
@@ -20,9 +50,9 @@ class LooseForeignKeys::DeletedRecord < ApplicationRecord
     # Run a query for each partition to optimize the row lookup by primary key (partition, id)
     update_count = 0
 
-    all_records.group_by(&:partition).each do |partition, records_within_partition|
+    all_records.group_by(&:partition_number).each do |partition, records_within_partition|
       update_count += status_pending
-        .where(partition: partition)
+        .for_partition(partition)
         .where(id: records_within_partition.pluck(:id))
         .update_all(status: :processed)
     end

config/feature_flags/development/lfk_automatic_partition_creation.yml

+---
+name: lfk_automatic_partition_creation
+introduced_by_url:
+rollout_issue_url: https://gitlab.com/gitlab-org/gitlab/-/issues/346907
+milestone: '14.6'
+type: development
+group: group::sharding
+default_enabled: false

config/feature_flags/development/lfk_automatic_partition_dropping.yml

+---
+name: lfk_automatic_partition_dropping
+introduced_by_url:
+rollout_issue_url: https://gitlab.com/gitlab-org/gitlab/-/issues/346908
+milestone: '14.6'
+type: development
+group: group::sharding
+default_enabled: false

config/initializers/postgres_partitioning.rb

@@ -2,7 +2,8 @@
 
 Gitlab::Database::Partitioning.register_models([
   AuditEvent,
-  WebHookLog
+  WebHookLog,
+  LooseForeignKeys::DeletedRecord
 ])
 
 if Gitlab.ee?

db/migrate/20211202094944_move_loose_fk_deleted_records_to_dynamic_schema.rb

+# frozen_string_literal: true
+
+class MoveLooseFkDeletedRecordsToDynamicSchema < Gitlab::Database::Migration[1.0]
+  enable_lock_retries!
+
+  def up
+    if table_exists?('gitlab_partitions_static.loose_foreign_keys_deleted_records_1')
+      execute 'ALTER TABLE gitlab_partitions_static.loose_foreign_keys_deleted_records_1 SET SCHEMA gitlab_partitions_dynamic'
+    end
+  end
+
+  def down
+    if table_exists?('gitlab_partitions_dynamic.loose_foreign_keys_deleted_records_1')
+      execute 'ALTER TABLE gitlab_partitions_dynamic.loose_foreign_keys_deleted_records_1 SET SCHEMA gitlab_partitions_static'
+    end
+  end
+end

db/schema_migrations/20211202094944

+2bca61880005c9303b2ff71747cde64d3418b6ef8ad2a9f114d584f4149e386b
\ No newline at end of file

db/structure.sql

@@ -125,6 +125,18 @@ CREATE TABLE incident_management_pending_issue_escalations (
 )
 PARTITION BY RANGE (process_at);
 
+CREATE TABLE loose_foreign_keys_deleted_records (
+    id bigint NOT NULL,
+    partition bigint DEFAULT 1 NOT NULL,
+    primary_key_value bigint NOT NULL,
+    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);
+
 CREATE TABLE verification_codes (
     created_at timestamp with time zone DEFAULT now() NOT NULL,
     visitor_id_code text NOT NULL,
@@ -1013,39 +1025,6 @@ CREATE TABLE gitlab_partitions_static.analytics_cycle_analytics_merge_request_st
 );
 ALTER TABLE ONLY analytics_cycle_analytics_merge_request_stage_events ATTACH PARTITION gitlab_partitions_static.analytics_cycle_analytics_merge_request_stage_events_31 FOR VALUES WITH (modulus 32, remainder 31);
 
-CREATE TABLE loose_foreign_keys_deleted_records (
-    id bigint NOT NULL,
-    partition bigint DEFAULT 1 NOT NULL,
-    primary_key_value bigint NOT NULL,
-    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);
-
-CREATE SEQUENCE loose_foreign_keys_deleted_records_id_seq
-    START WITH 1
-    INCREMENT BY 1
-    NO MINVALUE
-    NO MAXVALUE
-    CACHE 1;
-
-ALTER SEQUENCE loose_foreign_keys_deleted_records_id_seq OWNED BY loose_foreign_keys_deleted_records.id;
-
-CREATE TABLE gitlab_partitions_static.loose_foreign_keys_deleted_records_1 (
-    id bigint DEFAULT nextval('loose_foreign_keys_deleted_records_id_seq'::regclass) NOT NULL,
-    partition bigint DEFAULT 1 NOT NULL,
-    primary_key_value bigint NOT NULL,
-    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');
-
 CREATE TABLE product_analytics_events_experimental (
     id bigint NOT NULL,
     project_id integer NOT NULL,
@@ -15867,6 +15846,15 @@ CREATE SEQUENCE lists_id_seq
 
 ALTER SEQUENCE lists_id_seq OWNED BY lists.id;
 
+CREATE SEQUENCE loose_foreign_keys_deleted_records_id_seq
+    START WITH 1
+    INCREMENT BY 1
+    NO MINVALUE
+    NO MAXVALUE
+    CACHE 1;
+
+ALTER SEQUENCE loose_foreign_keys_deleted_records_id_seq OWNED BY loose_foreign_keys_deleted_records.id;
+
 CREATE TABLE member_tasks (
     id bigint NOT NULL,
     member_id bigint NOT NULL,
@@ -22397,12 +22385,6 @@ ALTER TABLE ONLY gitlab_partitions_static.analytics_cycle_analytics_merge_reques
 ALTER TABLE ONLY gitlab_partitions_static.analytics_cycle_analytics_merge_request_stage_events_31
     ADD CONSTRAINT analytics_cycle_analytics_merge_request_stage_events_31_pkey PRIMARY KEY (stage_event_hash_id, merge_request_id);
 
-ALTER TABLE ONLY loose_foreign_keys_deleted_records
-    ADD CONSTRAINT loose_foreign_keys_deleted_records_pkey PRIMARY KEY (partition, id);
-
-ALTER TABLE ONLY gitlab_partitions_static.loose_foreign_keys_deleted_records_1
-    ADD CONSTRAINT loose_foreign_keys_deleted_records_1_pkey PRIMARY KEY (partition, id);
-
 ALTER TABLE ONLY product_analytics_events_experimental
     ADD CONSTRAINT product_analytics_events_experimental_pkey PRIMARY KEY (id, project_id);
 
@@ -23492,6 +23474,9 @@ ALTER TABLE ONLY list_user_preferences
 ALTER TABLE ONLY lists
     ADD CONSTRAINT lists_pkey PRIMARY KEY (id);
 
+ALTER TABLE ONLY loose_foreign_keys_deleted_records
+    ADD CONSTRAINT loose_foreign_keys_deleted_records_pkey PRIMARY KEY (partition, id);
+
 ALTER TABLE ONLY member_tasks
     ADD CONSTRAINT member_tasks_pkey PRIMARY KEY (id);
 
@@ -24256,10 +24241,6 @@ 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);
@@ -26514,6 +26495,8 @@ CREATE INDEX index_lists_on_milestone_id ON lists USING btree (milestone_id);
 
 CREATE INDEX index_lists_on_user_id ON lists USING btree (user_id);
 
+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_member_tasks_on_member_id ON member_tasks USING btree (member_id);
 
 CREATE UNIQUE INDEX index_member_tasks_on_member_id_and_project_id ON member_tasks USING btree (member_id, project_id);
@@ -28164,8 +28147,6 @@ 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;
@@ -28674,8 +28655,6 @@ ALTER INDEX index_issue_stage_events_project_duration ATTACH PARTITION gitlab_pa
 
 ALTER INDEX index_issue_stage_events_group_in_progress_duration ATTACH PARTITION gitlab_partitions_static.index_ff8741d8d7;
 
-ALTER INDEX loose_foreign_keys_deleted_records_pkey ATTACH PARTITION gitlab_partitions_static.loose_foreign_keys_deleted_records_1_pkey;
-
 ALTER INDEX index_product_analytics_events_experimental_project_and_time ATTACH PARTITION gitlab_partitions_static.product_analytics_events_expe_project_id_collector_tstamp_idx10;
 
 ALTER INDEX index_product_analytics_events_experimental_project_and_time ATTACH PARTITION gitlab_partitions_static.product_analytics_events_expe_project_id_collector_tstamp_idx11;

spec/lib/gitlab/database/migration_helpers/loose_foreign_key_helpers_spec.rb

@@ -47,11 +47,16 @@ RSpec.describe Gitlab::Database::MigrationHelpers::LooseForeignKeyHelpers do
       record_to_be_deleted.delete
 
       expect(LooseForeignKeys::DeletedRecord.count).to eq(1)
-      deleted_record = LooseForeignKeys::DeletedRecord.all.first
+
+      arel_table = LooseForeignKeys::DeletedRecord.arel_table
+      deleted_record = LooseForeignKeys::DeletedRecord
+        .select(arel_table[Arel.star], arel_table[:partition].as('partition_number')) # aliasing the ignored partition column to partition_number
+        .all
+        .first
 
       expect(deleted_record.primary_key_value).to eq(record_to_be_deleted.id)
       expect(deleted_record.fully_qualified_table_name).to eq('public._test_loose_fk_test_table')
-      expect(deleted_record.partition).to eq(1)
+      expect(deleted_record.partition_number).to eq(1)
     end
 
     it 'stores multiple record deletions' do

spec/models/loose_foreign_keys/deleted_record_spec.rb

@@ -5,31 +5,148 @@ require 'spec_helper'
 RSpec.describe LooseForeignKeys::DeletedRecord, type: :model do
   let_it_be(:table) { 'public.projects' }
 
-  let_it_be(:deleted_record_1) { described_class.create!(partition: 1, fully_qualified_table_name: table, primary_key_value: 5) }
-  let_it_be(:deleted_record_2) { described_class.create!(partition: 1, fully_qualified_table_name: table, primary_key_value: 1) }
-  let_it_be(:deleted_record_3) { described_class.create!(partition: 1, fully_qualified_table_name: 'public.other_table', primary_key_value: 3) }
-  let_it_be(:deleted_record_4) { described_class.create!(partition: 1, fully_qualified_table_name: table, primary_key_value: 1) } # duplicate
+  describe 'class methods' do
+    let_it_be(:deleted_record_1) { described_class.create!(fully_qualified_table_name: table, primary_key_value: 5) }
+    let_it_be(:deleted_record_2) { described_class.create!(fully_qualified_table_name: table, primary_key_value: 1) }
+    let_it_be(:deleted_record_3) { described_class.create!(fully_qualified_table_name: 'public.other_table', primary_key_value: 3) }
+    let_it_be(:deleted_record_4) { described_class.create!(fully_qualified_table_name: table, primary_key_value: 1) } # duplicate
 
-  describe '.load_batch_for_table' do
-    it 'loads records and orders them by creation date' do
-      records = described_class.load_batch_for_table(table, 10)
+    describe '.load_batch_for_table' do
+      it 'loads records and orders them by creation date' do
+        records = described_class.load_batch_for_table(table, 10)
 
-      expect(records).to eq([deleted_record_1, deleted_record_2, deleted_record_4])
+        expect(records).to eq([deleted_record_1, deleted_record_2, deleted_record_4])
+      end
+
+      it 'supports configurable batch size' do
+        records = described_class.load_batch_for_table(table, 2)
+
+        expect(records).to eq([deleted_record_1, deleted_record_2])
+      end
     end
 
-    it 'supports configurable batch size' do
-      records = described_class.load_batch_for_table(table, 2)
+    describe '.mark_records_processed' do
+      it 'updates all records' do
+        records = described_class.load_batch_for_table(table, 10)
+        described_class.mark_records_processed(records)
 
-      expect(records).to eq([deleted_record_1, deleted_record_2])
+        expect(described_class.status_pending.count).to eq(1)
+        expect(described_class.status_processed.count).to eq(3)
+      end
     end
   end
 
-  describe '.mark_records_processed' do
-    it 'updates all records' do
-      described_class.mark_records_processed([deleted_record_1, deleted_record_2, deleted_record_4])
+  describe 'sliding_list partitioning' do
+    let(:connection) { described_class.connection }
+    let(:partition_manager) { Gitlab::Database::Partitioning::PartitionManager.new(described_class) }
+
+    describe 'next_partition_if callback' do
+      let(:active_partition) { described_class.partitioning_strategy.active_partition.value }
+
+      subject(:value) { described_class.partitioning_strategy.next_partition_if.call(active_partition) }
+
+      context 'when the partition is empty' do
+        it { is_expected.to eq(false) }
+      end
+
+      context 'when the partition has records' do
+        before do
+          described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 1, status: :processed)
+          described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 2, status: :pending)
+        end
+
+        it { is_expected.to eq(false) }
+      end
+
+      context 'when the first record of the partition is older than PARTITION_DURATION' do
+        before do
+          described_class.create!(
+            fully_qualified_table_name: 'public.table',
+            primary_key_value: 1,
+            created_at: (described_class::PARTITION_DURATION + 1.day).ago)
+
+          described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 2)
+        end
+
+        it { is_expected.to eq(true) }
+
+        context 'when the lfk_automatic_partition_creation FF is off' do
+          before do
+            stub_feature_flags(lfk_automatic_partition_creation: false)
+          end
+
+          it { is_expected.to eq(false) }
+        end
+      end
+    end
+
+    describe 'detach_partition_if callback' do
+      let(:active_partition) { described_class.partitioning_strategy.active_partition.value }
+
+      subject(:value) { described_class.partitioning_strategy.detach_partition_if.call(active_partition) }
+
+      context 'when the partition contains unprocessed records' do
+        before do
+          described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 1, status: :processed)
+          described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 2, status: :pending)
+        end
+
+        it { is_expected.to eq(false) }
+      end
+
+      context 'when the partition contains only processed records' do
+        before do
+          described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 1, status: :processed)
+          described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 2, status: :processed)
+        end
+
+        it { is_expected.to eq(true) }
+
+        context 'when the lfk_automatic_partition_dropping FF is off' do
+          before do
+            stub_feature_flags(lfk_automatic_partition_dropping: false)
+          end
+
+          it { is_expected.to eq(false) }
+        end
+      end
+    end
+
+    describe 'the behavior of the strategy' do
+      it 'moves records to new partitions as time passes', :freeze_time do
+        # We start with partition 1
+        expect(described_class.partitioning_strategy.current_partitions.map(&:value)).to eq([1])
+
+        # it's not a day old yet so no new partitions are created
+        partition_manager.sync_partitions
+
+        expect(described_class.partitioning_strategy.current_partitions.map(&:value)).to eq([1])
+
+        # add one record so the next partition will be created
+        described_class.create!(fully_qualified_table_name: 'public.table', primary_key_value: 1)
+
+        # after traveling forward a day
+        travel(described_class::PARTITION_DURATION + 1.second)
+
+        # a new partition is created
+        partition_manager.sync_partitions
+
+        expect(described_class.partitioning_strategy.current_partitions.map(&:value)).to eq([1, 2])
+
+        # and we can insert to the new partition
+        expect { described_class.create!(fully_qualified_table_name: table, primary_key_value: 5) }.not_to raise_error
+
+        # after processing old records
+        LooseForeignKeys::DeletedRecord.for_partition(1).update_all(status: :processed)
+
+        partition_manager.sync_partitions
+
+        # the old one is removed
+        expect(described_class.partitioning_strategy.current_partitions.map(&:value)).to eq([2])
 
-      expect(described_class.status_pending.count).to eq(1)
-      expect(described_class.status_processed.count).to eq(3)
+        # and we only have the newly created partition left.
+        expect(described_class.count).to eq(1)
+      end
     end
   end
 end

spec/services/loose_foreign_keys/batch_cleaner_service_spec.rb

@@ -90,7 +90,7 @@ RSpec.describe LooseForeignKeys::BatchCleanerService do
 
       described_class.new(parent_table: '_test_loose_fk_parent_table',
                           loose_foreign_key_definitions: loose_foreign_key_definitions,
-                          deleted_parent_records: LooseForeignKeys::DeletedRecord.status_pending.all
+                          deleted_parent_records: LooseForeignKeys::DeletedRecord.load_batch_for_table('public._test_loose_fk_parent_table', 100)
                          ).execute
     end