1. ORDER BY 和SORT BY 语句

order by 会对输入做全局排序，因此只有一个reducer（多个reducer无法保证全局有序）
只有一个reducer，会导致当输入规模较大时，需要较长的计算时间。

 

sort by不是全局排序，其在数据进入reducer前完成排序.因此，如果用sort by进行排序，并且设置mapred.reduce.tasks>1， 则sort by只保证每个reducer的输出有序，不保证全局有序。

 

l order by : 全局排序，启动一个reducer进行处理。

l sort by ：局部排序，启动多个reducer进行处理，一般业务都是获取topN的排名，一般可以使用sort by进行局部排序，在通过limit n对局部结果进行汇总

Hive提供了关键字limit，在可以获取topN，获取网站页面访问的top10的实现方案：

通过一下方案，我们可以发现order by 启动两个job，sort by 启动3个job。sort by可以启动多个reduce，每个reduce做局部排序，但是这对于sort by limit N已经够用了

方案1： 使用order by 进行排序

select * from (

select referrerpage ,count(1) refer_count from data_collect  group by referrerpage 

) t

order  by refer_count desc limit 10;

 

结果：

t.referrerpage  t.refer_count

http%3A//jf.10086.cn/   4714

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchMyExchangeWareInfo       1684

http%3A//jf.10086.cn/index.html 1171

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchWareInfo%26pager.offset%3D12     923

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchMyExchangeWareInfo%26pager.offset%3D12   794

http%3A//jf.10086.cn/ware/allClass.jsp  699

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchWareInfo%26pager.offset%3D24     685

http%3A//jf.10086.cn/portal/order/web/UserOrderAction%3Faction%3DdirectExchangeWare     680

http%3A//jf.10086.cn/rank/0_0_0_0.html  673

 

日志分析：

hive (jfyun)> select * from (

            > select referrerpage ,count(1) refer_count from data_collect  group by referrerpage 

            > ) t

            > order  by refer_count desc limit 10;

Automatically selecting local only mode for query

Total jobs = 2

 

Launching Job 1 out of 2

Number of reduce tasks not specified. Estimated from input data size: 1

In order to change the average load for a reducer (in bytes):

  set hive.exec.reducers.bytes.per.reducer=<number>

In order to limit the maximum number of reducers:

  set hive.exec.reducers.max=<number>

In order to set a constant number of reducers:

  set mapreduce.job.reduces=<number>

Job running in-process (local Hadoop)

Hadoop job information for null: number of mappers: 0; number of reducers: 0

2016-01-06 21:08:49,862 null map = 0%,  reduce = 0%

2016-01-06 21:08:53,345 null map = 100%,  reduce = 0%

2016-01-06 21:08:54,462 null map = 100%,  reduce = 100%

Ended Job = job_local785531207_0001

Execution completed successfully

MapredLocal task succeeded

 

Launching Job 2 out of 2

Number of reduce tasks determined at compile time: 1

In order to change the average load for a reducer (in bytes):

  set hive.exec.reducers.bytes.per.reducer=<number>

In order to limit the maximum number of reducers:

  set hive.exec.reducers.max=<number>

In order to set a constant number of reducers:

  set mapreduce.job.reduces=<number>

 

Job running in-process (local Hadoop)

Hadoop job information for null: number of mappers: 0; number of reducers: 0

2016-01-06 21:09:03,129 null map = 0%,  reduce = 0%

2016-01-06 21:09:04,385 null map = 100%,  reduce = 0%

2016-01-06 21:09:05,459 null map = 100%,  reduce = 100%

Ended Job = job_local127113388_0001

Execution completed successfully

MapredLocal task succeeded

OK

 

方案2： 使用sort by 进行排序

select * from (
select referrerpage ,count(1) refer_count from data_collect  group by referrerpage 
) t
sort by refer_count desc limit 10;

 

 

结果：

t.referrerpage  t.refer_count

http%3A//jf.10086.cn/   4714

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchMyExchangeWareInfo       1684

http%3A//jf.10086.cn/index.html 1171

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchWareInfo%26pager.offset%3D12     923

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchMyExchangeWareInfo%26pager.offset%3D12   794

http%3A//jf.10086.cn/ware/allClass.jsp  699

http%3A//jf.10086.cn/portal/ware/web/SearchWareAction%3Faction%3DsearchWareInfo%26pager.offset%3D24     685

http%3A//jf.10086.cn/portal/order/web/UserOrderAction%3Faction%3DdirectExchangeWare     680

http%3A//jf.10086.cn/rank/0_0_0_0.html  673

 

日志分析：

hive (jfyun)> select * from (
            > select referrerpage ,count(1) refer_count from data_collect  group by referrerpage 
            > ) t
            > sort by refer_count desc limit 10;
Automatically selecting local only mode for query
Total jobs = 3
 
Launching Job 1 out of 3
Number of reduce tasks not specified. Estimated from input data size: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
 
Job running in-process (local Hadoop)
Hadoop job information for null: number of mappers: 0; number of reducers: 0
2016-01-06 21:12:28,743 null map = 0%,  reduce = 0%
2016-01-06 21:12:32,163 null map = 100%,  reduce = 0%
2016-01-06 21:12:33,285 null map = 100%,  reduce = 100%
Ended Job = job_local1285072626_0001
Execution completed successfully
MapredLocal task succeeded
Launching Job 2 out of 3
 
Number of reduce tasks not specified. Estimated from input data size: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Job running in-process (local Hadoop)
Hadoop job information for null: number of mappers: 0; number of reducers: 0
2016-01-06 21:12:43,206 null map = 0%,  reduce = 0%
2016-01-06 21:12:44,397 null map = 100%,  reduce = 0%
2016-01-06 21:12:45,518 null map = 100%,  reduce = 100%
Ended Job = job_local1012382962_0001
Execution completed successfully
MapredLocal task succeeded
Launching Job 3 out of 3
 
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
 
Job running in-process (local Hadoop)
Hadoop job information for null: number of mappers: 0; number of reducers: 0
2016-01-06 21:12:53,969 null map = 100%,  reduce = 100%
Ended Job = job_local931416756_0001
Execution completed successfully
MapredLocal task succeede

 

2. 包含SORT BY 的DISTRIBUTE BY

distribute by按照指定的字段对数据进行划分到不同的输出reduce / 文件中，根据map函数输出的key通过hash函数计算哈希数值，然后得到的哈希数值将键-值对均匀分发到多个reduer中去。

 

需求：获取每个商品类别和国家下的商品个数，并按照商品类别和国家下的商品销售取前三名。

数据：

类别 国家 商品名称 销售数量

movies us movies_us_1 100

movies us movies_us_2 150

movies us movies_us_3 200

movies us movies_us_4 300

movies gb movies_gb_1 100

movies gb movies_gb_2 150

movies gb movies_gb_3 200

movies gb movies_gb_4 300

office gb office_gb_1 30

office gb office_gb_2 40

office gb office_gb_3 50

office gb office_gb_4 60

office us office_us_1 30

office us office_us_2 50

office us office_us_3 60

office us office_us_4 70

数据表

create external table p_rank_demo(category string,country string,product string,sales int) row format delimited fields terminated by '\t';

 

加载数据

load data local inpath '/home/hadoop/p_rank_demo' overwrite into table p_rank_demo;

 

按照商品类别和国家下的商品销售排名(按照category,country分组，并通过sort by每个分组的结果排序)

select  category,country,product, sales,rank() over(distribute by category,country sort by category,country, sales desc)  as rk from p_rank_demo;

 

或者通过

select  category,country,product, sales,rank() over(partition by category,country order by sales desc)  as rk from p_rank_demo;

 

输出通过的结果（两者的区别： distribute by 具有相同的类别和国家的记录都发送到同一个reducer上，这样就可以统计

出每个类别和国家的销售商品的排名了。若业务是全局排序，可以通过partition by order by 获取排名。）

category        country product sales   rk

movies  gb      movies_gb_4     300     1

movies  gb      movies_gb_3     200     2

movies  gb      movies_gb_2     150     3

movies  gb      movies_gb_1     100     4

movies  us      movies_us_4     300     1

movies  us      movies_us_3     200     2

movies  us      movies_us_1     100     3

movies  us      movies_us_2     100     3

office  gb      office_gb_4     60      1

office  gb      office_gb_3     50      2

office  gb      office_gb_2     40      3

office  gb      office_gb_1     30      4

office  us      office_us_4     70      1

office  us      office_us_3     60      2

office  us      office_us_2     50      3

office  us      office_us_1     30      4

 

按照商品类别和国家下的商品销售排名前三输出

select * from (select  category,country,product, sales,rank() over(distribute by category,country sort by category,country, sales desc)  as rk from p_rank_demo) t where t.rk <= 3;

 

输出结果：

t.category      t.country       t.product       t.sales t.rk

movies  gb      movies_gb_4     300     1

movies  gb      movies_gb_3     200     2

movies  gb      movies_gb_2     150     3

movies  us      movies_us_4     300     1

movies  us      movies_us_3     200     2

movies  us      movies_us_1     100     3

movies  us      movies_us_2     100     3

office  gb      office_gb_4     60      1

office  gb      office_gb_3     50      2

office  gb      office_gb_2     40      3

office  us      office_us_4     70      1

office  us      office_us_3     60      2

office  us      office_us_2     50      3

 

3. CLUSTER BY 

使用distribute  by  A  sort by A 语句同cluster by A语句相同，即：分区和排序属于一个字段。但是cluster by 排序只能是倒序排序，不能指定排序规则为asc 或者desc。