hive 排序特性研究

分手后的思念是犯贱 2022-08-07 10:52 132阅读 0赞

1.  排序定义： 所谓排序就是使一串记录，按照其中的某个或某些关键字，递增或是递减的排列。  
2. hive 中排序相关内容   
2.1 order by   
order by 会对输入做全局排序，故只有一个reducer,若数据的规模比较大时，需要较长的计算时间。hive中order by 也是对一个结果集进行排序，不同于关系型数据库是底层架构。hive的hive-site.xml配置文件中的参数hive.mapred.mode控制着hive的执行方式，若选择strict，则order by 则需要指定limit（若有分区还有指定哪个分区） ；若为nostrict，则与关系型数据库差不多。由于order by 执行时，只有一个reducer ，如果结果集过大，那执行时间相对会比较漫长。  
注：若不想修改配置文件，可临时执行：set hive.mapred.mode=nonstrict 或set hive.mapred.mode=strict;也可以在当前会话中达到同样的效果。  
测试：  
\--未开启strict模式,即nostrict模式  
hive> select id,devid,devname from tb\_in\_base order by devid;   
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15431, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15431  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15431  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 1  
2013-08-05 16:33:21,817 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 16:33:23,828 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:24,834 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:25,843 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:26,849 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:27,855 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:28,860 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:29,873 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:30,880 Stage-1 map = 100%,  reduce = 33%, Cumulative CPU 0.94 sec  
2013-08-05 16:33:31,888 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.51 sec  
2013-08-05 16:33:32,893 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.51 sec  
2013-08-05 16:33:33,899 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.51 sec  
MapReduce Total cumulative CPU time: 2 seconds 510 msec  
Ended Job = job\_201307151509\_15431  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 1   Cumulative CPU: 2.51 sec   HDFS Read: 559 HDFS Write: 138 SUCCESS  
Total MapReduce CPU Time Spent: 2 seconds 510 msec  
OK  
1       121212  test1  
2       131313  test2  
3       141414  test3  
4       151515  test5  
5       161616  test6  
6       171717  test7  
8       191919  test9overwrite  
8       191919  test9overwrite  
Time taken: 16.872 seconds  
  
结果说明：没有开启严格模式时，order by 与关系型数据库效果类似。  
\--开启scrict模式，且未在order by 后面加limit    
hive> select id,devid from tb\_in\_base order by devid;  
FAILED: Error in semantic analysis: 1:41 In strict mode, if ORDER BY is specified, LIMIT must also be specified. Error encountered near token 'devid'  
注：没有指定limit 报错  
\--开启scrict模式，且未在order by 后面加limit  ，且未指定分区  
hive> select \* from tb\_in\_base;  
FAILED: Error in semantic analysis: No partition predicate found for Alias "tb\_in\_base" Table "tb\_in\_base"  
结果说明：严格模式下，无法直接进行查询。  
\--开启scrict模式，且未在order by 后面加limit  ，且指定分区  
hive> select \* from tb\_in\_base where job\_time=030729 order by devid limit 2;  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15432, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15432  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15432  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 1  
2013-08-05 16:47:32,900 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 16:47:34,920 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:35,927 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:36,934 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:37,941 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:38,946 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:39,953 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:40,959 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:41,965 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.39 sec  
2013-08-05 16:47:42,971 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.05 sec  
2013-08-05 16:47:43,977 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.05 sec  
2013-08-05 16:47:44,983 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.05 sec  
MapReduce Total cumulative CPU time: 3 seconds 50 msec  
Ended Job = job\_201307151509\_15432  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 1   Cumulative CPU: 3.05 sec   HDFS Read: 458 HDFS Write: 44 SUCCESS  
Total MapReduce CPU Time Spent: 3 seconds 50 msec  
OK  
1       121212  test1   030729  
2       131313  test2   030729  
Time taken: 17.597 seconds  
结果说明：严格模式下，使用order by 不仅需要指定limit 数量，若有表分区还需要指定表分区。  
  
2.2 sort by   
sort可以控制每个reduce产生的文件都是排序，再对多个排序的好的文件做二次归并排序。sort by 特点如下：  
 1) . sort by 基本受hive.mapred.mode是否为strict、nonstrict的影响，但若有分区需要指定分区。  
 2). sort by 的数据在同一个reduce中数据是按指定字段排序。  
 3). sort by 可以指定执行的reduce个数，如：set mapred.reduce.tasks=5 ,对输出的数据再执行归并排序，即可以得到全部结果。  
\-- 在hive.mapred.mode为nonstrict时  
hive> select id,devid from tb\_in\_base sort by devid;   
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15434, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15434  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15434  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 1  
2013-08-05 17:15:38,244 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 17:15:40,258 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:41,269 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:42,275 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:43,281 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:44,286 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:45,292 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:46,298 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:47,305 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.3 sec  
2013-08-05 17:15:48,311 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.07 sec  
2013-08-05 17:15:49,317 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.07 sec  
2013-08-05 17:15:50,323 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.07 sec  
MapReduce Total cumulative CPU time: 3 seconds 70 msec  
Ended Job = job\_201307151509\_15434  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 1   Cumulative CPU: 3.07 sec   HDFS Read: 559 HDFS Write: 72 SUCCESS  
Total MapReduce CPU Time Spent: 3 seconds 70 msec  
OK  
1       121212  
2       131313  
3       141414  
4       151515  
5       161616  
6       171717  
8       191919  
8       191919  
Time taken: 15.969 seconds  
  
  
\---开启hive.mapred.mode为strict模式  
hive>  set hive.mapred.mode=strict;  
hive> select \* from tb\_in\_base order by devid;   //注：测试set hive.mapred.mode=strict;是否生效  
  
  
FAILED: Error in semantic analysis: 1:34 In strict mode, if ORDER BY is specified, LIMIT must also be specified. Error encountered near token 'devid'  
  
\---开启hive.mapred.mode为strict模式，查询时未指定分区  
  
hive> select \* from tb\_in\_base sort by devid;   
FAILED: Error in semantic analysis: No partition predicate found for Alias "tb\_in\_base" Table "tb\_in\_base"  
  
\---开启hive.mapred.mode为strict模式，查询时指定分区  
hive> select id,devid,job\_time from tb\_in\_base where job\_time=030729  sort by devid;   
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15438, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15438  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15438  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 1  
2013-08-05 17:22:02,631 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 17:22:05,645 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:06,651 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:07,656 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:08,662 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:09,668 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:10,673 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:11,679 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:12,685 Stage-1 map = 100%,  reduce = 33%, Cumulative CPU 0.89 sec  
2013-08-05 17:22:13,691 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.61 sec  
2013-08-05 17:22:14,696 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.61 sec  
2013-08-05 17:22:15,704 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.61 sec  
MapReduce Total cumulative CPU time: 2 seconds 610 msec  
Ended Job = job\_201307151509\_15438  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 1   Cumulative CPU: 2.61 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 2 seconds 610 msec  
OK  
1       121212  030729  
2       131313  030729  
3       141414  030729  
4       151515  030729  
5       161616  030729  
6       171717  030729  
8       191919  030729  
Time taken: 18.044 seconds  
结果说明：严格模式下，sort by 不指定limit 数，可以正常执行。sort by 受hive.mapred.mode=sctrict 的影响较小。  
\---set mapred.reduce.tasks=2，使用sort by    
hive> set mapred.reduce.tasks=2;                                  
hive> insert overwrite local directory '/tmp/hivetest/sortby' select id,devid,job\_time from tb\_in\_base where job\_time=030729 sort by devid;  //注：将查询结果写入到指定目录中  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
Number of reduce tasks not specified. Defaulting to jobconf value of: 2  //注：说明有两个reducer  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15466, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15466  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15466  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 2  
2013-08-05 18:03:48,298 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 18:03:50,307 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:51,312 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:52,317 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:53,322 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:54,327 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:55,333 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:56,338 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:57,343 Stage-1 map = 100%,  reduce = 17%, Cumulative CPU 1.33 sec  
2013-08-05 18:03:58,351 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.57 sec  
2013-08-05 18:03:59,356 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.57 sec  
2013-08-05 18:04:00,362 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.57 sec  
MapReduce Total cumulative CPU time: 4 seconds 570 msec  
Ended Job = job\_201307151509\_15466  
Copying data to local directory /tmp/hivetest/sortby  
Copying data to local directory /tmp/hivetest/sortby  
7 Rows loaded to /tmp/hivetest/sortby  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 2   Cumulative CPU: 4.57 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 4 seconds 570 msec  
OK  
Time taken: 16.712 seconds  
  
查看/tmp/hivetest/sort下查询结果  
结果说明：sort by 可以通过mapred.reduce.task指定reduce个数，可以将查询后的数据分别写入两个不同的reduce文件中。  
\---set mapred.reduce.tasks=2，使用order by    
hive> set mapred.reduce.tasks=2;      
hive>  insert overwrite local directory '/tmp/hivetest/orderby' select id,devid,job\_time from tb\_in\_base where job\_time=030729 order by devid;  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
Number of reduce tasks determined at compile time: 1  //注：说明只有一个reducer  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15469, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15469  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15469  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 1  
2013-08-05 18:14:22,528 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 18:14:24,538 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:25,548 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:26,562 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:27,568 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:28,574 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:29,581 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:30,587 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:31,592 Stage-1 map = 100%,  reduce = 33%, Cumulative CPU 1.05 sec  
2013-08-05 18:14:32,598 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.56 sec  
2013-08-05 18:14:33,604 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.56 sec  
2013-08-05 18:14:34,611 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 2.56 sec  
MapReduce Total cumulative CPU time: 2 seconds 560 msec  
Ended Job = job\_201307151509\_15469  
Copying data to local directory /tmp/hivetest/orderby  
Copying data to local directory /tmp/hivetest/orderby  
7 Rows loaded to /tmp/hivetest/orderby  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 1   Cumulative CPU: 2.56 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 2 seconds 560 msec  
OK  
Time taken: 16.804 seconds  
  
查看/tmp/hivetest/orderby下查询结果   
结果说明：设置mapred.reduce.task=2，使用order by做排序，所有的数据都是写入同一个文件中。由此可说明不论设计多少个reduce任务数，order by 只使用一个reduce。   
注：可以用limit 子句大大减少数据量，使用limit  n ，后，传输到reduce商的数据记录就减少到n\* map，否则可能由于数据过大可能出不了结果。  
疑问：使用limit n之后，数据是减少了，但是对于统计数据的话，少了数据的统计还有意义吗？  
2.3 distribute by   
distribute by 是控制在map端如何拆分给reduce端。根据distribute by 后面的列及reduce个数进行数据分发，默认采用hash算法。distribute可以使用length方法会根据string类型的长度划分到不同的reduce中，最终输出到不同的文件中。 length 是内建函数，也可以指定其他的函数或这使用自定义函数。  
注：对distribute by 进行测试，一定要分配多个reduce进行处理，否则无法看到distribute by 效果。  
\--使用id进行排序且set mapred.reduce.tasks=2;   
hive> set mapred.reduce.tasks=2;        
hive> insert overwrite local directory '/tmp/hivetest/distributeby'  select id,devid,job\_time from tb\_in\_base where job\_time=030729 and id<8  distribute   by id ;  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
Number of reduce tasks not specified. Defaulting to jobconf value of: 2  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15499, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15499  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15499  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 2  
2013-08-05 18:37:14,681 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 18:37:16,691 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:37:17,697 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:37:18,703 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:37:19,710 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:37:20,717 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:37:21,727 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:37:22,733 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.33 sec  
2013-08-05 18:37:23,739 Stage-1 map = 100%,  reduce = 50%, Cumulative CPU 3.1 sec  
2013-08-05 18:37:24,745 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.89 sec  
2013-08-05 18:37:25,751 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.89 sec  
2013-08-05 18:37:26,757 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.89 sec  
MapReduce Total cumulative CPU time: 4 seconds 890 msec  
Ended Job = job\_201307151509\_15499  
Copying data to local directory /tmp/hivetest/distributeby  
Copying data to local directory /tmp/hivetest/distributeby  
7 Rows loaded to /tmp/hivetest/distributeby  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 2   Cumulative CPU: 4.89 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 4 seconds 890 msec  
OK  
Time taken: 16.785 seconds  
查看写入的查询数据:  
结果说明：distribute by采用hash算法，将查询的结果写入不同的reduce文件中。数据分配到哪个reduce文件中，是在map端控制的。   
\--使用job\_time进行排序且set mapred.reduce.tasks=2;   
hive> set mapred.reduce.tasks=2;        
hive> insert overwrite local directory '/tmp/hivetest/distributeby'  select id,devid,job\_time from tb\_in\_base where job\_time=030729 distribute by job\_time;  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
Number of reduce tasks not specified. Defaulting to jobconf value of: 2  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15500, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15500  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15500  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 2  
2013-08-05 18:42:07,764 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 18:42:10,778 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:11,784 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:12,789 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:13,795 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:14,801 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:15,810 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:16,816 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:17,821 Stage-1 map = 100%,  reduce = 33%, Cumulative CPU 0.61 sec  
2013-08-05 18:42:18,827 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.89 sec  
2013-08-05 18:42:19,833 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.89 sec  
2013-08-05 18:42:20,839 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 3.89 sec  
MapReduce Total cumulative CPU time: 3 seconds 890 msec  
Ended Job = job\_201307151509\_15500  
Copying data to local directory /tmp/hivetest/distributeby  
Copying data to local directory /tmp/hivetest/distributeby  
7 Rows loaded to /tmp/hivetest/distributeby  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 2   Cumulative CPU: 3.89 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 3 seconds 890 msec  
OK  
Time taken: 16.678 seconds  
查看查询后的结果：  
结果说明：hash值一样的数据被分配到同一个reduce中。  
额外验证:  
\---sort  by 是否也像distribute by 一样使用hash算法  
set mapred.reduce.tasks=2;  
hive>  insert overwrite local directory '/tmp/hivetest/sortby'  select id,devid,job\_time from tb\_in\_base where job\_time=030729 sort by id;  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
Number of reduce tasks not specified. Defaulting to jobconf value of: 2  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15501, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15501  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15501  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 2  
2013-08-05 18:57:33,616 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 18:57:35,625 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:36,631 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:37,636 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:38,642 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:39,648 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:40,653 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:41,659 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:42,669 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.16 sec  
2013-08-05 18:57:43,675 Stage-1 map = 100%,  reduce = 67%, Cumulative CPU 3.02 sec  
2013-08-05 18:57:44,681 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.68 sec  
2013-08-05 18:57:45,687 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.68 sec  
2013-08-05 18:57:46,693 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.68 sec  
MapReduce Total cumulative CPU time: 4 seconds 680 msec  
Ended Job = job\_201307151509\_15501  
Copying data to local directory /tmp/hivetest/sortby  
Copying data to local directory /tmp/hivetest/sortby  
7 Rows loaded to /tmp/hivetest/sortby  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 2   Cumulative CPU: 4.68 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 4 seconds 680 msec  
OK  
查询结果：  
  
结果说明：sort by  不支持通过hash算法将数据分配到不同的reduce文件。  
2.4  cluster by   
cluster by 除了distribute by 的功能外，还会对该字段进行排序，所以cluster by = distribute by +sort by   
\---cluster by 是否可以指定asc 或desc  
hive> select id,devid,job\_time from tb\_in\_base where job\_time=030729 cluster by id desc;  
FAILED: Parse Error: line 1:77 mismatched input 'desc' expecting EOF near 'id'  
  
hive> select id,devid,job\_time from tb\_in\_base where job\_time=030729 cluster by id asc;   
FAILED: Parse Error: line 1:77 mismatched input 'asc' expecting EOF near 'id'  
注：cluster by 默认倒序排序  
hive> set mapred.reduce.tasks=2;                                                             
hive> insert overwrite local directory '/tmp/hivetest/clusterby' select id,devid,job\_time from tb\_in\_base where job\_time=030729 cluster by id;  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
Number of reduce tasks not specified. Defaulting to jobconf value of: 2  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15532, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15532  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15532  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 2  
2013-08-05 19:41:15,138 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 19:41:17,147 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:18,153 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:19,158 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:20,163 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:21,169 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:22,174 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:23,180 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:24,186 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.28 sec  
2013-08-05 19:41:25,193 Stage-1 map = 100%,  reduce = 67%, Cumulative CPU 2.5 sec  
2013-08-05 19:41:26,199 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.34 sec  
2013-08-05 19:41:27,205 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.34 sec  
2013-08-05 19:41:28,210 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.34 sec  
MapReduce Total cumulative CPU time: 4 seconds 340 msec  
Ended Job = job\_201307151509\_15532  
Copying data to local directory /tmp/hivetest/clusterby  
Copying data to local directory /tmp/hivetest/clusterby  
7 Rows loaded to /tmp/hivetest/clusterby  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 2   Cumulative CPU: 4.34 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 4 seconds 340 msec  
OK  
Time taken: 17.637 seconds  
写入文件后的查询结果：  
结果分析：采用hash算法，根据hash值将查询的结果写入不同的reduce文件中。  
hive> set mapred.reduce.tasks=2;  
hive> insert overwrite local directory '/tmp/hivetest/clusterby' select id,devid,job\_time from tb\_in\_base where job\_time=030729 cluster by job\_time;  
Total MapReduce jobs = 1  
Launching Job 1 out of 1  
Number of reduce tasks not specified. Defaulting to jobconf value of: 2  
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 mapred.reduce.tasks=<number>  
Starting Job = job\_201307151509\_15533, Tracking URL = http://mwtec-50:50030/jobdetails.jsp?jobid=job\_201307151509\_15533  
Kill Command = /home/hadoop/hadoop-0.20.2/bin/hadoop job  -Dmapred.job.tracker=mwtec-50:9002 -kill job\_201307151509\_15533  
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 2  
2013-08-05 19:43:38,722 Stage-1 map = 0%,  reduce = 0%  
2013-08-05 19:43:40,732 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:41,738 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:42,743 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:43,748 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:44,754 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:45,759 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:46,765 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:47,770 Stage-1 map = 100%,  reduce = 17%, Cumulative CPU 1.5 sec  
2013-08-05 19:43:48,776 Stage-1 map = 100%,  reduce = 67%, Cumulative CPU 3.02 sec  
2013-08-05 19:43:49,781 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.55 sec  
2013-08-05 19:43:50,787 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 4.55 sec  
MapReduce Total cumulative CPU time: 4 seconds 550 msec  
Ended Job = job\_201307151509\_15533  
Copying data to local directory /tmp/hivetest/clusterby  
Copying data to local directory /tmp/hivetest/clusterby  
7 Rows loaded to /tmp/hivetest/clusterby  
MapReduce Jobs Launched:   
Job 0: Map: 1  Reduce: 2   Cumulative CPU: 4.55 sec   HDFS Read: 458 HDFS Write: 112 SUCCESS  
Total MapReduce CPU Time Spent: 4 seconds 550 msec  
OK  
Time taken: 16.613 seconds  
写入文件后的查询结果：  
结果说明：cluster by 对其指定的列，进行hash，若hash值相同则被分配到相同的reduce文件中。  
 3 . 总结分析  
1).  order by 只有一个reduce负责对所有的数据进行排序，若大数据量，则需要较长的时间。建议在小的数据集中使用order by 进行排序。  
2).  order by 可以通过设置hive.mapred.mode参数控制执行方式，若选择strict，则order by 则需要指定limit（若有分区还有指定哪个分区） ；若为nostrict，则与关系型数据库差不多。  
3). sort by 基本上不受hive.mapred.mode影响，可以通过mapred.reduce.task 指定reduce个数，查询后的数据被分发到相关的reduce中。  
4). sort by 的数据在进入reduce前就完成排序，如果要使用sort by 是行排序，并且设置map.reduce.tasks>1,则sort by 才能保证每个reducer输出有序，不能保证全局数据有序。  
5). distribute by 采集hash算法，在map端将查询的结果中hash值相同的结果分发到对应的reduce文件中。  
6). distribute by 可以使用length方法会根据string类型的长度划分到不同的reduce中，最终输出到不同的文件中。 length 是内建函数，也可以指定其他的函数或这使用自定义函数。  
7). cluster by 除了distribute by 的功能外，还会对该字段进行排序，所以cluster by = distribute by +sort by 。