Spark SQL 指南

DISTRIBUTE BY 子句

描述

DISTRIBUTE BY 子句用于根据输入表达式重新分区数据。与 CLUSTER BY 子句不同，这不会对每个分区中的数据进行排序。

语法

DISTRIBUTE BY { expression [ , ... ] }

参数

示例

CREATE TABLE person (name STRING, age INT);
INSERT INTO person VALUES
    ('Zen Hui', 25),
    ('Anil B', 18),
    ('Shone S', 16),
    ('Mike A', 25),
    ('John A', 18),
    ('Jack N', 16);

-- Reduce the number of shuffle partitions to 2 to illustrate the behavior of `DISTRIBUTE BY`.
-- It's easier to see the clustering and sorting behavior with less number of partitions.
SET spark.sql.shuffle.partitions = 2;

-- Select the rows with no ordering. Please note that without any sort directive, the result
-- of the query is not deterministic. It's included here to just contrast it with the
-- behavior of `DISTRIBUTE BY`. The query below produces rows where age columns are not
-- clustered together.
SELECT age, name FROM person;
+---+-------+
|age|   name|
+---+-------+
| 16|Shone S|
| 25|Zen Hui|
| 16| Jack N|
| 25| Mike A|
| 18| John A|
| 18| Anil B|
+---+-------+

-- Produces rows clustered by age. Persons with same age are clustered together.
-- Unlike `CLUSTER BY` clause, the rows are not sorted within a partition.
SELECT age, name FROM person DISTRIBUTE BY age;
+---+-------+
|age|   name|
+---+-------+
| 25|Zen Hui|
| 25| Mike A|
| 18| John A|
| 18| Anil B|
| 16|Shone S|
| 16| Jack N|
+---+-------+