that may mean, for every ci we have to manually restart failed job in the future?