05-Hive动态分区

大家好！砸门又见面了。我先作一个自我介绍吧。我是一个打算学习大数据一万小时的谢老师，目前学习了800多小时了，现在正在努力。

今天来玩的实验是：Hive分区表的动态分区

分区是在处理大型事实表时常用的方法。分区的好处在于缩小查询扫描范围，从而提高速度。分区分为两种：静态分区static partition和动态分区dynamic partition。静态分区和动态分区的区别在于导入数据时，是手动输入分区名称，还是通过数据来判断数据分区。对于大数据批量导入来说，显然采用动态分区更为简单方便。

动态分区不需要为不同的分区添加不同的插入语句。那什么时候使用动态分区呢？其实，当你不知道要怎么分区，这个分区要通过查询语句里得出来，通过查出来的数据建立分区。换句话说就是分区数据不确定的时候，就适合使用动态分区。

有几个参数，要认识一下：

set hive.exec.dynamic.partition=true// 使用动态分区
set hive.exec.dynamic.partition.mode=nonstrick//无限制模式
如果模式是stric，则必须有一个静态分区，且放在最前面
set hive.exec.max.dynamic.partitions.pernode=10000;//每个节点生成动态分区的最大个数
set hive.exec.max.created.files=150000;//一个任务最多可以创建的文件数目
set dfs.datanode.max.xcievers=8192;//限定一次最多打开的文件数

1 接下来我们来实战一下怎么动态分区：

hive> create table d_part(
> name string
> )
> partitioned by(value string)
> row format delimited fields terminated by '\t'
> lines terminated by '\n'
> stored as textfile;
OK
Time taken: 0.243 seconds
hive>

注意到什么？partitioned by后面没有指定具体的值了。

hive> show partitions d_part;
OK
Time taken: 0.304 seconds
hive> select * from d_part;
OK
Time taken: 0.096 seconds
hive>

没有数据对不对？我们来插入一些数据，并且让他动态分区。

hive> set hive.exec.dynamic.partition=true;
hive> set hive.exec.dynamic.partition.mode=nonstrick;
hive> insert overwrite table d_part partition(value)
> select name,
> addr as value
> from testtext;
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Job running in-process (local Hadoop)
Hadoop job information for null: number of mappers: 1; number of reducers: 0
2016-06-02 05:16:16,951 null map = 0%,  reduce = 0%
2016-06-02 05:16:32,117 null map = 100%,  reduce = 0%, Cumulative CPU 1.07 sec
2016-06-02 05:16:33,226 null map = 100%,  reduce = 0%, Cumulative CPU 1.07 sec
2016-06-02 05:16:34,277 null map = 100%,  reduce = 0%, Cumulative CPU 1.07 sec
MapReduce Total cumulative CPU time: 1 seconds 70 msec
Ended Job = job_1464828076391_0016
Execution completed successfully
Mapred Local Task Succeeded . Convert the Join into MapJoin
Loading data to table default.d_part partition (value=null)
Loading partition {value=78}
Loading partition {value=46}
Loading partition {value=89}
Partition default.d_part{value=46} stats: [num_files: 1, num_rows: 0, total_size: 4, raw_data_size: 0]
Partition default.d_part{value=78} stats: [num_files: 1, num_rows: 0, total_size: 6, raw_data_size: 0]
Partition default.d_part{value=89} stats: [num_files: 1, num_rows: 0, total_size: 7, raw_data_size: 0]
Table default.d_part stats: [num_partitions: 3, num_files: 3, num_rows: 0, total_size: 17, raw_data_size: 0]
OK
Time taken: 41.974 seconds
hive>

查看一下数据加载了没？

hive> select * from d_part;
OK
wer 46
weree   78
wer 89
rr  89
Time taken: 0.732 seconds
hive> show partitions d_part;
OK
value=46
value=78
value=89
Time taken: 0.142 seconds
hive>

为什么会有三个partition呢？其实addr as value就说清楚了，addr的值就作为分区值，分别是46,78,89（两个89分到一起了）.不信你看：

hive> desc testtext;
OK
name    string  name value
addr    string  addr value
Time taken: 0.155 seconds
hive>  select * from testtext;
OK
wer 46
wer 89
weree   78
rr  89
Time taken: 0.162 seconds

我们再看web端的情况：

搞啥子哟，图片居然上传不上来！！

(此处本该有图)

不生气不生气，待会补上咯！

2 接下来我们要创建两个分区：

hive> create table d_part2(
> name string
> )
> partitioned by(value string,dt string)
> row format delimited fields terminated by '\t'
> lines terminated by '\n'
> stored as textfile;
OK
Time taken: 0.092 seconds
hive> set hive.exec.dynamic.partition=true;
hive>

好，停一下。是不是第二次见到

set hive.exec.dynamic.partition=true;

了？dynamic是动态的意思，明白了吧？

接着干嘛？插入数据：

hive> insert overwrite table d_part2 partition(value,dt)
> select 'test' as name,
> addr as value,
> name as dt
> from testtext;
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Job running in-process (local Hadoop)
Hadoop job information for null: number of mappers: 1; number of reducers: 0
2016-06-02 05:45:36,828 null map = 0%,  reduce = 0%
2016-06-02 05:45:52,339 null map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec
2016-06-02 05:45:53,443 null map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec
2016-06-02 05:45:54,493 null map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec
MapReduce Total cumulative CPU time: 1 seconds 300 msec
Ended Job = job_1464828076391_0017
Execution completed successfully
Mapred Local Task Succeeded . Convert the Join into MapJoin
Loading data to table default.d_part2 partition (value=null, dt=null)
Loading partition {value=89, dt=wer}
Loading partition {value=89, dt=rr}
Loading partition {value=46, dt=wer}
Loading partition {value=78, dt=weree}
Partition default.d_part2{value=46, dt=wer} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Partition default.d_part2{value=78, dt=weree} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Partition default.d_part2{value=89, dt=rr} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Partition default.d_part2{value=89, dt=wer} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Table default.d_part2 stats: [num_partitions: 4, num_files: 4, num_rows: 0, total_size: 20, raw_data_size: 0]
OK
Time taken: 40.971 seconds
hive>

各位看官，这里两个分区哦，value 分区下有子分区dt.两个分区都是根据数据的值来分的。

addr as value, name as dt。你知道d_part2表下面的数据是啥吗？select ‘test’ as name, 都是test。

来验证一下：

hive> show partitions d_part2;
OK
value=46/dt=wer
value=78/dt=weree
value=89/dt=rr
value=89/dt=wer
Time taken: 0.293 seconds
hive> select * from d_part2;
OK
test    46  wer
test    78  weree
test    89  rr
test    89  wer
Time taken: 0.435 seconds
hive>

学了这两个实验基本就入门了。砸门还是应该多练习一下，对吧？熟能生巧，巧能生能。

hive> create table d_part3(
> name string
> )
> partitioned by(value string,dt string)
> row format delimited fields terminated by '\t'
> lines terminated by '\n'
> stored as textfile;
OK
Time taken: 0.121 seconds
hive> set hive.exec.dynamic.partition=true;

插入数据：

hive> insert overwrite table d_part3 partition(value,dt)
> select 'test' as name,
> addr as value,
> name as dt
> from testtext;
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Job running in-process (local Hadoop)
Hadoop job information for null: number of mappers: 1; number of reducers: 0
2016-06-02 05:54:27,509 null map = 0%,  reduce = 0%
2016-06-02 05:54:37,411 null map = 100%,  reduce = 0%, Cumulative CPU 0.99 sec
2016-06-02 05:54:38,501 null map = 100%,  reduce = 0%, Cumulative CPU 0.99 sec
2016-06-02 05:54:39,555 null map = 100%,  reduce = 0%, Cumulative CPU 0.99 sec
MapReduce Total cumulative CPU time: 990 msec
Ended Job = job_1464828076391_0018
Execution completed successfully
Mapred Local Task Succeeded . Convert the Join into MapJoin
Loading data to table default.d_part3 partition (value=null, dt=null)
Loading partition {value=78, dt=weree}
Loading partition {value=46, dt=wer}
Loading partition {value=89, dt=wer}
Loading partition {value=89, dt=rr}
Partition default.d_part3{value=46, dt=wer} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Partition default.d_part3{value=78, dt=weree} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Partition default.d_part3{value=89, dt=rr} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Partition default.d_part3{value=89, dt=wer} stats: [num_files: 1, num_rows: 0, total_size: 5, raw_data_size: 0]
Table default.d_part3 stats: [num_partitions: 4, num_files: 4, num_rows: 0, total_size: 20, raw_data_size: 0]
OK
Time taken: 36.737 seconds
hive> show partitions d_part3;
OK
value=46/dt=wer
value=78/dt=weree
value=89/dt=rr
value=89/dt=wer
Time taken: 0.264 seconds
hive>

好了，一遍练不会练两边。

这里给大家布置一个作业：什么情况下设置动态分区会产生灾难？欢迎留言呀！

有点累了，休息一下。如果你看到此文，想进一步学习或者和我沟通，加我微信公众号：五十年后