apache-flink - 弗林克 : ValueState vs ReducingState/AggregatingState

Question

背景

• ValueState文档指出:partitioned single-value state .
• ReducingState文档指出:combined using a reduce function .
• AggregatingState文档指出:eagerly pre-aggregated .
• ValueState延伸State而两者ReducingState和AggregatingState延伸MergingState .

问题

1. 各州何时合并？
2. 我应该如何为特定问题选择正确的状态原语？
3. 什么机制调用reduce和aggregate函数？它是否跳过非 MergingState ？

Answer 1

Fabian Hueske 不久前(2018 年 5 月 6 日)回答了我关于 AggregateFunctions 中合并的问题。他说:

The only situation when merge() is called in a DataStream job (that I am aware of) is when session windows get merged. For example when you define a session window with 30 minute gap and you receive the following records R1, 12:00:00 R2, 12:05:00 R3, 12:40:00 R4, 12:20:00

In this case, Flink R1 will create a new window W1, R2 will be assigned to W1, R3 > creates a new window W2, and R4 connects and merges W1 and W2.

我认为您其他问题的部分答案是 ValueState 是通用(键控)状态。因此，当您实现通用函数时，您最终会使用它，而不是聚合器或缩减器(带有组合器)。