#hive partition
partition是hive提供的一种机制:用户通过指定一个或多个partition key,决定数据存放方式,进而优化数据的查询
一个表可以指定多个partition key,每个partition在hive中以文件夹的形式存在。
**实例(static partition):**
~~~
编辑文件:/home/work/data/test3.txt; /home/work/data/test4.txt;
$ cat /home/work/data/test3.txt
1,zxm
2,ljz
3,cds
4,mac
5,android
6,symbian
7,wp
$ cat /home/work/data/test4.txt
8,zxm
9,ljz
10,cds
11,mac
12,android
13,symbian
14,wp
~~~
建表:
~~~
hive> create table student_tmp(id INT, name STRING)
> partitioned by(academy STRING, class STRING)
> row format delimited fields terminated by ',';
OK
Time taken: 6.505 seconds
id,name是真实列,partition列academy和class是伪列
~~~
load数据:(此处直接load数据进partition,在hive 0.6之前的版本,必须先创建好partition,数据才能导入)
~~~
hive> load data local inpath '/home/work/data/test3.txt' into table student_tmp partition(academy='computer', class='034');
Copying data from file:/home/work/data/test3.txt
Copying file: file:/home/work/data/test3.txt
Loading data to table default.student_tmp partition (academy=computer, class=034)
OK
Time taken: 0.898 seconds
hive> load data local inpath '/home/work/data/test3.txt' into table student_tmp partition(academy='physics', class='034');
Copying data from file:/home/work/data/test3.txt
Copying file: file:/home/work/data/test3.txt
Loading data to table default.student_tmp partition (academy=physics, class=034)
OK
Time taken: 0.256 seconds
~~~
查看hive文件结构:
~~~
$ hadoop fs -ls /user/hive/warehouse/student_tmp/
Found 2 items
drwxr-xr-x - work supergroup 0 2012-07-30 18:47 /user/hive/warehouse/student_tmp/academy=computer
drwxr-xr-x - work supergroup 0 2012-07-30 19:00 /user/hive/warehouse/student_tmp/academy=physics
$ hadoop fs -ls /user/hive/warehouse/student_tmp/academy=computer
Found 1 items
drwxr-xr-x - work supergroup 0 2012-07-30 18:47 /user/hive/warehouse/student_tmp/academy=computer/class=034
~~~
查询数据:
~~~
hive> select * from student_tmp where academy='physics';
OK
1 zxm physics 034
2 ljz physics 034
3 cds physics 034
4 mac physics 034
5 android physics 034
6 symbian physics 034
7 wp physics 034
Time taken: 0.139 seconds
~~~
以上是static partition的示例,static partition即由用户指定数据所在的partition,在load数据时,指定partition(academy='computer', class='034');
static partition常适用于使用处理时间作为partition key的例子。
但是,我们也常常会遇到需要向分区表中插入大量数据,并且插入前不清楚数据归宿的partition,此时,我们需要dynamic partition。
使用动态分区需要设置hive.exec.dynamic.partition参数值为true。
可以设置部分列为dynamic partition列,例如:partition(academy='computer', class);
也可以设置所有列为dynamic partition列,例如partition(academy, class);
设置所有列为dynamic partition列时,需要设置hive.exec.dynamic.partition.mode=nonstrict
需要注意的是,主分区为dynamic partition列,而副分区为static partition列是不允许的,例如partition(academy, class=‘034’);是不允许的
**示例(dynamic partition):**
~~~
建表
hive> create table student(id INT, name STRING)
> partitioned by(academy STRING, class STRING)
> row format delimited fields terminated by ',';
OK
Time taken: 0.393 seconds
~~~
设置参数
~~~
hive> set hive.exec.dynamic.partition.mode=nonstrict;
hive> set hive.exec.dynamic.partition=true;
~~~
导入数据:
~~~
hive> insert overwrite table student partition(academy, class)
> select id,name,academy,class from student_tmp
> where class='034';
Total MapReduce jobs = 2
.........
OK
Time taken: 29.616 seconds
~~~
查询数据:
~~~
hive> select * from student where academy='physics';
OK
1 zxm physics 034
2 ljz physics 034
3 cds physics 034
4 mac physics 034
5 android physics 034
6 symbian physics 034
7 wp physics 034
Time taken: 0.165 seconds
~~~
查看文件:
~~~
$ hadoop fs -ls /user/hive/warehouse/student/
Found 2 items
drwxr-xr-x - work supergroup 0 2012-07-30 19:22 /user/hive/warehouse/student/academy=computer
drwxr-xr-x - work supergroup 0 2012-07-30 19:22 /user/hive/warehouse/student/academy=physics
~~~
**总结:**
hive partition是通过将数据拆分成不同的partition放入不同的文件,从而减少查询操作时数据处理规模的手段。
例如,Hive Select查询中,如果没有建partition,则会扫描整个表内容,这样计算量巨大。如果我们在相应维度做了partition,则处理数据规模可能会大大减少。
附partition相关参数:
hive.exec.dynamic.partition(缺省false):
设置为true允许使用dynamic partition
hive.exec.dynamic.partition.mode(缺省strick):设置dynamic partition模式(nostrict允许所有partition列都为dynamic partition,strict不允许)
hive.exec.max.dynamic.partitions.pernode (缺省100):每一个mapreduce job允许创建的分区的最大数量,如果超过了这个数量就会报错
hive.exec.max.dynamic.partitions (缺省1000):一个dml语句允许创建的所有分区的最大数量
hive.exec.max.created.files (缺省100000):所有的mapreduce job允许创建的文件的最大数量
reference:
[Dynamic Partitions](https://cwiki.apache.org/Hive/dynamicpartitions.html)
[hive中简单介绍分区表(partition table),含动态分区(dynamic partition)与静态分区(static partition)](http://blog.sina.com.cn/s/blog_6ff05a2c0100tah0.html)