gitlab-org--gitlab-foss/lib/tasks/gitlab/db.rake

372 lines
15 KiB
Ruby

# frozen_string_literal: true
databases = ActiveRecord::Tasks::DatabaseTasks.setup_initial_database_yaml
def each_database(databases, include_geo: false)
ActiveRecord::Tasks::DatabaseTasks.for_each(databases) do |database|
next if !include_geo && database == 'geo'
yield database
end
end
namespace :gitlab do
namespace :db do
desc 'GitLab | DB | Manually insert schema migration version on all configured databases'
task :mark_migration_complete, [:version] => :environment do |_, args|
mark_migration_complete(args[:version])
end
namespace :mark_migration_complete do
each_database(databases) do |database_name|
desc "Gitlab | DB | Manually insert schema migration version on #{database_name} database"
task database_name, [:version] => :environment do |_, args|
mark_migration_complete(args[:version], only_on: database_name)
end
end
end
def mark_migration_complete(version, only_on: nil)
if version.to_i == 0
puts 'Must give a version argument that is a non-zero integer'.color(:red)
exit 1
end
Gitlab::Database::EachDatabase.each_database_connection(only: only_on) do |connection, name|
connection.execute("INSERT INTO schema_migrations (version) VALUES (#{connection.quote(version)})")
puts "Successfully marked '#{version}' as complete on database #{name}".color(:green)
rescue ActiveRecord::RecordNotUnique
puts "Migration version '#{version}' is already marked complete on database #{name}".color(:yellow)
end
end
desc 'GitLab | DB | Drop all tables on all configured databases'
task drop_tables: :environment do
drop_tables
end
namespace :drop_tables do
each_database(databases) do |database_name|
desc "GitLab | DB | Drop all tables on the #{database_name} database"
task database_name => :environment do
drop_tables(only_on: database_name)
end
end
end
def drop_tables(only_on: nil)
Gitlab::Database::EachDatabase.each_database_connection(only: only_on) do |connection, name|
# In PostgreSQLAdapter, data_sources returns both views and tables, so use tables instead
tables = connection.tables
# Removes the entry from the array
tables.delete 'schema_migrations'
# Truncate schema_migrations to ensure migrations re-run
connection.execute('TRUNCATE schema_migrations') if connection.table_exists? 'schema_migrations'
# Drop any views
connection.views.each do |view|
connection.execute("DROP VIEW IF EXISTS #{connection.quote_table_name(view)} CASCADE")
end
# Drop tables with cascade to avoid dependent table errors
# PG: http://www.postgresql.org/docs/current/static/ddl-depend.html
# Add `IF EXISTS` because cascade could have already deleted a table.
tables.each { |t| connection.execute("DROP TABLE IF EXISTS #{connection.quote_table_name(t)} CASCADE") }
# Drop all extra schema objects GitLab owns
Gitlab::Database::EXTRA_SCHEMAS.each do |schema|
connection.execute("DROP SCHEMA IF EXISTS #{connection.quote_table_name(schema)} CASCADE")
end
end
end
desc 'GitLab | DB | Configures the database by running migrate, or by loading the schema and seeding if needed'
task configure: :environment do
databases_with_tasks = ActiveRecord::Base.configurations.configs_for(env_name: Rails.env)
databases_loaded = []
if databases_with_tasks.size == 1
next unless databases_with_tasks.first.name == 'main'
connection = Gitlab::Database.database_base_models['main'].connection
databases_loaded << configure_database(connection)
else
Gitlab::Database.database_base_models.each do |name, model|
next unless databases_with_tasks.any? { |db_with_tasks| db_with_tasks.name == name }
databases_loaded << configure_database(model.connection, database_name: name)
end
end
Rake::Task['db:seed_fu'].invoke if databases_loaded.present? && databases_loaded.all?
end
def configure_database(connection, database_name: nil)
database_name = ":#{database_name}" if database_name
load_database = connection.tables.count <= 1
if load_database
Gitlab::Database.add_post_migrate_path_to_rails(force: true)
Rake::Task["db:schema:load#{database_name}"].invoke
else
Rake::Task["db:migrate#{database_name}"].invoke
end
load_database
end
desc 'GitLab | DB | Run database migrations and print `unattended_migrations_completed` if action taken'
task unattended: :environment do
no_database = !ActiveRecord::Base.connection.schema_migration.table_exists?
needs_migrations = ActiveRecord::Base.connection.migration_context.needs_migration?
if no_database || needs_migrations
Rake::Task['gitlab:db:configure'].invoke
puts "unattended_migrations_completed"
else
puts "unattended_migrations_static"
end
end
desc 'GitLab | DB | Sets up EE specific database functionality'
if Gitlab.ee?
task setup_ee: %w[db:drop:geo db:create:geo db:schema:load:geo db:migrate:geo]
else
task :setup_ee
end
desc 'This adjusts and cleans db/structure.sql - it runs after db:structure:dump'
task :clean_structure_sql do |task_name|
ActiveRecord::Base.configurations.configs_for(env_name: ActiveRecord::Tasks::DatabaseTasks.env).each do |db_config|
structure_file = ActiveRecord::Tasks::DatabaseTasks.dump_filename(db_config.name)
schema = File.read(structure_file)
File.open(structure_file, 'wb+') do |io|
Gitlab::Database::SchemaCleaner.new(schema).clean(io)
end
end
# Allow this task to be called multiple times, as happens when running db:migrate:redo
Rake::Task[task_name].reenable
end
# Inform Rake that custom tasks should be run every time rake db:structure:dump is run
#
# Rails 6.1 deprecates db:structure:dump in favor of db:schema:dump
Rake::Task['db:structure:dump'].enhance do
Rake::Task['gitlab:db:clean_structure_sql'].invoke
end
# Inform Rake that custom tasks should be run every time rake db:schema:dump is run
Rake::Task['db:schema:dump'].enhance do
Rake::Task['gitlab:db:clean_structure_sql'].invoke
end
ActiveRecord::Tasks::DatabaseTasks.for_each(databases) do |name|
# Inform Rake that custom tasks should be run every time rake db:structure:dump is run
#
# Rails 6.1 deprecates db:structure:dump in favor of db:schema:dump
Rake::Task["db:structure:dump:#{name}"].enhance do
Rake::Task['gitlab:db:clean_structure_sql'].invoke
end
Rake::Task["db:schema:dump:#{name}"].enhance do
Rake::Task['gitlab:db:clean_structure_sql'].invoke
end
end
desc 'Create missing dynamic database partitions'
task create_dynamic_partitions: :environment do
Gitlab::Database::Partitioning.sync_partitions
end
namespace :create_dynamic_partitions do
each_database(databases) do |database_name|
desc "Create missing dynamic database partitions on the #{database_name} database"
task database_name => :environment do
Gitlab::Database::Partitioning.sync_partitions(only_on: database_name)
end
end
end
# This is targeted towards deploys and upgrades of GitLab.
# Since we're running migrations already at this time,
# we also check and create partitions as needed here.
Rake::Task['db:migrate'].enhance do
Rake::Task['gitlab:db:create_dynamic_partitions'].invoke
end
# We'll temporarily skip this enhancement for geo, since in some situations we
# wish to setup the geo database before the other databases have been setup,
# and partition management attempts to connect to the main database.
each_database(databases) do |database_name|
Rake::Task["db:migrate:#{database_name}"].enhance do
Rake::Task["gitlab:db:create_dynamic_partitions:#{database_name}"].invoke
end
end
# When we load the database schema from db/structure.sql
# we don't have any dynamic partitions created. We don't really need to
# because application initializers/sidekiq take care of that, too.
# However, the presence of partitions for a table has influence on their
# position in db/structure.sql (which is topologically sorted).
#
# Other than that it's helpful to create partitions early when bootstrapping
# a new installation.
Rake::Task['db:schema:load'].enhance do
Rake::Task['gitlab:db:create_dynamic_partitions'].invoke
end
# We'll temporarily skip this enhancement for geo, since in some situations we
# wish to setup the geo database before the other databases have been setup,
# and partition management attempts to connect to the main database.
each_database(databases) do |database_name|
# :nocov:
Rake::Task["db:schema:load:#{database_name}"].enhance do
Rake::Task["gitlab:db:create_dynamic_partitions:#{database_name}"].invoke
end
# :nocov:
end
desc "Clear all connections"
task :clear_all_connections do
ActiveRecord::Base.clear_all_connections!
end
Rake::Task['db:test:purge'].enhance(['gitlab:db:clear_all_connections'])
Rake::Task['db:drop'].enhance(['gitlab:db:clear_all_connections'])
# During testing, db:test:load restores the database schema from scratch
# which does not include dynamic partitions. We cannot rely on application
# initializers here as the application can continue to run while
# a rake task reloads the database schema.
Rake::Task['db:test:load'].enhance do
# Due to bug in `db:test:load` if many DBs are used
# the `ActiveRecord::Base.connection` might be switched to another one
# This is due to `if should_reconnect`:
# https://github.com/rails/rails/blob/a81aeb63a007ede2fe606c50539417dada9030c7/activerecord/lib/active_record/railties/databases.rake#L622
ActiveRecord::Base.establish_connection :main # rubocop: disable Database/EstablishConnection
Rake::Task['gitlab:db:create_dynamic_partitions'].invoke
end
desc "Reindex database without downtime to eliminate bloat"
task reindex: :environment do
unless Gitlab::Database::Reindexing.enabled?
puts "This feature (database_reindexing) is currently disabled.".color(:yellow)
exit
end
Gitlab::Database::Reindexing.invoke
end
namespace :reindex do
each_database(databases) do |database_name|
desc "Reindex #{database_name} database without downtime to eliminate bloat"
task database_name => :environment do
unless Gitlab::Database::Reindexing.enabled?
puts "This feature (database_reindexing) is currently disabled.".color(:yellow)
exit
end
Gitlab::Database::Reindexing.invoke(database_name)
end
end
end
desc 'Enqueue an index for reindexing'
task :enqueue_reindexing_action, [:index_name, :database] => :environment do |_, args|
model = Gitlab::Database.database_base_models[args.fetch(:database, Gitlab::Database::PRIMARY_DATABASE_NAME)]
Gitlab::Database::SharedModel.using_connection(model.connection) do
queued_action = Gitlab::Database::PostgresIndex.find(args[:index_name]).queued_reindexing_actions.create!
puts "Queued reindexing action: #{queued_action}"
puts "There are #{Gitlab::Database::Reindexing::QueuedAction.queued.size} queued actions in total."
end
unless Feature.enabled?(:database_reindexing, type: :ops, default_enabled: :yaml)
puts <<~NOTE.color(:yellow)
Note: database_reindexing feature is currently disabled.
Enable with: Feature.enable(:database_reindexing)
NOTE
end
end
desc 'Check if there have been user additions to the database'
task active: :environment do
if ActiveRecord::Base.connection.migration_context.needs_migration?
puts "Migrations pending. Database not active"
exit 1
end
# A list of projects that GitLab creates automatically on install/upgrade
# gc = Gitlab::CurrentSettings.current_application_settings
seed_projects = [Gitlab::CurrentSettings.current_application_settings.self_monitoring_project]
if (Project.count - seed_projects.count {|x| !x.nil? }).eql?(0)
puts "No user created projects. Database not active"
exit 1
end
puts "Found user created projects. Database active"
exit 0
end
namespace :migration_testing do
desc 'Run migrations with instrumentation'
task up: :environment do
Gitlab::Database::Migrations::Runner.up.run
end
desc 'Run down migrations in current branch with instrumentation'
task down: :environment do
Gitlab::Database::Migrations::Runner.down.run
end
desc 'Sample traditional background migrations with instrumentation'
task :sample_background_migrations, [:duration_s] => [:environment] do |_t, args|
duration = args[:duration_s]&.to_i&.seconds || 30.minutes # Default of 30 minutes
Gitlab::Database::Migrations::Runner.background_migrations.run_jobs(for_duration: duration)
end
end
desc 'Run all pending batched migrations'
task execute_batched_migrations: :environment do
Gitlab::Database::EachDatabase.each_database_connection do |connection, name|
Gitlab::Database::BackgroundMigration::BatchedMigration.with_status(:active).queue_order.each do |migration|
Gitlab::AppLogger.info("Executing batched migration #{migration.id} on database #{name} inline")
Gitlab::Database::BackgroundMigration::BatchedMigrationRunner.new(connection: connection).run_entire_migration(migration)
end
end
end
desc 'Run migration as gitlab non-superuser'
task :reset_as_non_superuser, [:username] => :environment do |_, args|
username = args.fetch(:username, 'gitlab')
puts "Migrate using username #{username}"
Rake::Task['db:drop'].invoke
Rake::Task['db:create'].invoke
ActiveRecord::Base.configurations.configs_for(env_name: ActiveRecord::Tasks::DatabaseTasks.env).each do |db_config|
ActiveRecord::Base.establish_connection(db_config.configuration_hash.merge(username: username)) # rubocop: disable Database/EstablishConnection
Gitlab::Database.check_for_non_superuser
Rake::Task['db:migrate'].invoke
end
end
# Only for development environments,
# we execute pending data migrations inline for convenience.
Rake::Task['db:migrate'].enhance do
if Rails.env.development? && Gitlab::Database::BackgroundMigration::BatchedMigration.table_exists?
Rake::Task['gitlab:db:execute_batched_migrations'].invoke
end
end
end
end