Merge branch '325615/finalize_ci_job_artifacts_bigint-part-ii' into 'master'

Finalize job_id conversion to bigint for ci_job_artifacts

See merge request gitlab-org/gitlab!67774

db/post_migrate/20210809143931_finalize_job_id_conversion_to_bigint_for_ci_job_artifacts.rb

+# frozen_string_literal: true
+
+class FinalizeJobIdConversionToBigintForCiJobArtifacts < ActiveRecord::Migration[6.1]
+  include Gitlab::Database::MigrationHelpers
+
+  disable_ddl_transaction!
+
+  TABLE_NAME = 'ci_job_artifacts'
+
+  def up
+    ensure_batched_background_migration_is_finished(
+      job_class_name: 'CopyColumnUsingBackgroundMigrationJob',
+      table_name: TABLE_NAME,
+      column_name: 'id',
+      job_arguments: [%w[id job_id], %w[id_convert_to_bigint job_id_convert_to_bigint]]
+    )
+
+    swap
+  end
+
+  def down
+    swap
+  end
+
+  private
+
+  def swap
+    # This is to replace the existing "index_ci_job_artifacts_on_expire_at_and_job_id" btree (expire_at, job_id)
+    add_concurrent_index TABLE_NAME, [:expire_at, :job_id_convert_to_bigint], name: 'index_ci_job_artifacts_on_expire_at_and_job_id_bigint'
+    # This is to replace the existing "index_ci_job_artifacts_on_job_id_and_file_type" btree (job_id, file_type)
+    add_concurrent_index TABLE_NAME, [:job_id_convert_to_bigint, :file_type], name: 'index_ci_job_artifacts_on_job_id_and_file_type_bigint', unique: true
+
+    # # Add a FK on `job_id_convert_to_bigint` to `ci_builds(id)`, the old FK (fk_rails_c5137cb2c1)
+    # # is removed below since it won't be dropped automatically.
+    fk_ci_builds_job_id = concurrent_foreign_key_name(TABLE_NAME, :job_id, prefix: 'fk_rails_')
+    fk_ci_builds_job_id_tmp = "#{fk_ci_builds_job_id}_tmp"
+
+    add_concurrent_foreign_key TABLE_NAME, :ci_builds,
+      column: :job_id_convert_to_bigint,
+      name: fk_ci_builds_job_id_tmp,
+      on_delete: :cascade,
+      reverse_lock_order: true
+
+    with_lock_retries(raise_on_exhaustion: true) do
+      # We'll need  ACCESS EXCLUSIVE lock on the related tables,
+      # lets make sure it can be acquired from the start
+
+      execute "LOCK TABLE ci_builds, #{TABLE_NAME} IN ACCESS EXCLUSIVE MODE"
+
+      temp_name = 'job_id_tmp'
+      execute "ALTER TABLE #{quote_table_name(TABLE_NAME)} RENAME COLUMN #{quote_column_name(:job_id)} TO #{quote_column_name(temp_name)}"
+      execute "ALTER TABLE #{quote_table_name(TABLE_NAME)} RENAME COLUMN #{quote_column_name(:job_id_convert_to_bigint)} TO #{quote_column_name(:job_id)}"
+      execute "ALTER TABLE #{quote_table_name(TABLE_NAME)} RENAME COLUMN #{quote_column_name(temp_name)} TO #{quote_column_name(:job_id_convert_to_bigint)}"
+
+      # We need to update the trigger function in order to make PostgreSQL to
+      # regenerate the execution plan for it. This is to avoid type mismatch errors like
+      # "type of parameter 15 (bigint) does not match that when preparing the plan (integer)"
+      function_name = Gitlab::Database::UnidirectionalCopyTrigger.on_table(TABLE_NAME).name([:id, :job_id], [:id_convert_to_bigint, :job_id_convert_to_bigint])
+      execute "ALTER FUNCTION #{quote_table_name(function_name)} RESET ALL"
+
+      # Swap defaults
+      change_column_default TABLE_NAME, :job_id, nil
+      change_column_default TABLE_NAME, :job_id_convert_to_bigint, 0
+
+      # Rename the rest of the indexes (we already hold an exclusive lock, so no need to use DROP INDEX CONCURRENTLY here
+      execute 'DROP INDEX index_ci_job_artifacts_on_expire_at_and_job_id'
+      rename_index TABLE_NAME, 'index_ci_job_artifacts_on_expire_at_and_job_id_bigint', 'index_ci_job_artifacts_on_expire_at_and_job_id'
+      execute 'DROP INDEX index_ci_job_artifacts_on_job_id_and_file_type'
+      rename_index TABLE_NAME, 'index_ci_job_artifacts_on_job_id_and_file_type_bigint', 'index_ci_job_artifacts_on_job_id_and_file_type'
+
+      # Drop original FK on the old int4 `job_id` (fk_rails_c5137cb2c1)
+      remove_foreign_key TABLE_NAME, name: fk_ci_builds_job_id
+
+      # We swapped the columns but the FK for job_id is still using the temporary name for the job_id_convert_to_bigint column
+      # So we have to also swap the FK name now that we dropped the other one with the same
+      rename_constraint(TABLE_NAME, fk_ci_builds_job_id_tmp, fk_ci_builds_job_id)
+    end
+  end
+end

db/schema_migrations/20210809143931

+37cac2c3c5c5c22a34e0a77733c5330a32101090ac47b46260123c3362a9e36f
\ No newline at end of file

db/structure.sql

@@ -10787,7 +10787,7 @@ ALTER SEQUENCE ci_instance_variables_id_seq OWNED BY ci_instance_variables.id;
 CREATE TABLE ci_job_artifacts (
     id integer NOT NULL,
     project_id integer NOT NULL,
-    job_id integer NOT NULL,
+    job_id_convert_to_bigint integer DEFAULT 0 NOT NULL,
     file_type integer NOT NULL,
     size bigint,
     created_at timestamp with time zone NOT NULL,
@@ -10799,7 +10799,7 @@ CREATE TABLE ci_job_artifacts (
     file_format smallint,
     file_location smallint,
     id_convert_to_bigint bigint DEFAULT 0 NOT NULL,
-    job_id_convert_to_bigint bigint DEFAULT 0 NOT NULL,
+    job_id bigint NOT NULL,
     CONSTRAINT check_27f0f6dbab CHECK ((file_store IS NOT NULL))
 );