mirror of
https://github.com/mastodon/mastodon.git
synced 2024-11-26 23:39:37 +00:00
da13fa5021
* rubocop: quit being so picky * rubocop: miscellany * rubocop: prefer present to blank
76 lines
2.4 KiB
Ruby
76 lines
2.4 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
require_relative '../mastodon/snowflake'
|
|
|
|
def each_schema_load_environment
|
|
# If we're in development, also run this for the test environment.
|
|
# This is a somewhat hacky way to do this, so here's why:
|
|
# 1. We have to define this before we load the schema, or we won't
|
|
# have a timestamp_id function when we get to it in the schema.
|
|
# 2. db:setup calls db:schema:load_if_ruby, which calls
|
|
# db:schema:load, which we define above as having a prerequisite
|
|
# of this task.
|
|
# 3. db:schema:load ends up running
|
|
# ActiveRecord::Tasks::DatabaseTasks.load_schema_current, which
|
|
# calls a private method `each_current_configuration`, which
|
|
# explicitly also does the loading for the `test` environment
|
|
# if the current environment is `development`, so we end up
|
|
# needing to do the same, and we can't even use the same method
|
|
# to do it.
|
|
|
|
if Rails.env.development?
|
|
test_conf = ActiveRecord::Base.configurations['test']
|
|
|
|
if test_conf['database']&.present?
|
|
ActiveRecord::Base.establish_connection(:test)
|
|
yield
|
|
ActiveRecord::Base.establish_connection(Rails.env.to_sym)
|
|
end
|
|
end
|
|
|
|
yield
|
|
end
|
|
|
|
namespace :db do
|
|
namespace :migrate do
|
|
desc 'Setup the db or migrate depending on state of db'
|
|
task setup: :environment do
|
|
begin
|
|
if ActiveRecord::Migrator.current_version.zero?
|
|
Rake::Task['db:migrate'].invoke
|
|
Rake::Task['db:seed'].invoke
|
|
end
|
|
rescue ActiveRecord::NoDatabaseError
|
|
Rake::Task['db:setup'].invoke
|
|
else
|
|
Rake::Task['db:migrate'].invoke
|
|
end
|
|
end
|
|
end
|
|
|
|
# Before we load the schema, define the timestamp_id function.
|
|
# Idiomatically, we might do this in a migration, but then it
|
|
# wouldn't end up in schema.rb, so we'd need to figure out a way to
|
|
# get it in before doing db:setup as well. This is simpler, and
|
|
# ensures it's always in place.
|
|
Rake::Task['db:schema:load'].enhance ['db:define_timestamp_id']
|
|
|
|
# After we load the schema, make sure we have sequences for each
|
|
# table using timestamp IDs.
|
|
Rake::Task['db:schema:load'].enhance do
|
|
Rake::Task['db:ensure_id_sequences_exist'].invoke
|
|
end
|
|
|
|
task :define_timestamp_id do
|
|
each_schema_load_environment do
|
|
Mastodon::Snowflake.define_timestamp_id
|
|
end
|
|
end
|
|
|
|
task :ensure_id_sequences_exist do
|
|
each_schema_load_environment do
|
|
Mastodon::Snowflake.ensure_id_sequences_exist
|
|
end
|
|
end
|
|
end
|