1. fetch need_scheduler routine load job 2. caculate current concurrent task number of job 3. divide kafka partition into tasks