Core principle description of Task assignment algorithm

1. Create a Taskset object for each task that the stage generates
   

2. Dual for loop, traverse all taskset, and each localization level

3. The localization levels are:

   1.process_local: Process localization RDD partition and task Enter the same executor, the fastest

   2.node_local:rdd's partition and task are not on a executor, but on a worker

   3.no_perf: There is no so-called localization level

   4.rack_local: Rack localization, at least RDD partition and task on a rack

   5. Any localization level

4. Task first chooses the best localization level

5. Call Tasksetmanager's Resourceoffer method to find out which tasks can be started on this executor,

6. Put the Taskset array to the specified executor start task
   

Core principle description of Task assignment algorithm