resilient distributed datasets 读后笔记

1.Formally, an RDD is a read-only, partitioned collection of records. RDDs can be only created through deterministic operations on either (1) a dataset in stable storage or (2) other existing RDDs.

2.RDD是延迟加载的，就是说直到action被触发，才真正有动作。

 

3. RDD之间的关系分为narrow dependency 和 wide dependency，看图很好理解

4.spark的scheuler会把程序逻辑和RDD变成DAG图来，分stage执行