1, Stage: Virtual parallel task set, all tasks on the same Stage have the same shuffle dependency. The stages are divided according to the shuffle mark. One stage with multiple Rdd, first with an rdd and one stage with multiple task stages through Shuffledependency division, one stage is narrowdependency stage type There are two kinds, shufflemapstage and Resultstage. -Shufflemapstage The result of this phase task is the input to the next stage task. You need to keep track of the node where each partition resides. Intermediate process during task execution, which saves the output data of a task for fetch by the next reduce. This phase can be submitted separately. -The Resultstage results stage executes the RDD action directly. Compute functions are applied to some partitions (not necessarily in all partitions, such as first (), take (3)). 2, Task: Runs on a node, a real task contains an RDD the entire transformation process from each partition of the last Rdd to find his dependence, that is its task; the number of partitions in the last Rdd is the task number for this phase is the Spark execution unit, and there are two Type. -Shuffelmaptask is composed of multiple shufflemaptask in Shufflemapstage. -The Resulttask Resultstage is composed of multiple resulttask, which results in the task being sent back to driver. 3, job an action is a job4, application An application can contain multiple jobs.
4, 2 core components