概述
桶的概念，主要是为性能考虑，可以理解为对分区内列，进行再次划分，提高性能。在底层，一个桶其实是一个文件。如果桶划分过多，会导致文件数量暴增，一旦达到系统文件数量的上限，就杯具了。哪种是最优数量，这个哥也不知道。
桶表是对数据进行哈希取值，然后放到不同文件中存储。
数据加载到桶表时，会对字段取hash值，然后与桶的数量取模。把数据放到对应的文件中。
物理上，每个桶就是表(或分区）目录里的一个文件，一个作业产生的桶(输出文件)和reduce任务个数相同。
作用
桶表专门用于抽样查询，是很专业性的，不是日常用来存储数据的表，需要抽样查询时，才创建和使用桶表。
实验
创建
[22:39:03]hive (zmgdb)> create table bucket_t1(id string)
[22:39:26] > clustered by(id) into 6 buckets;
[22:39:27]OK
[22:39:27]Time taken: 0.546 seconds
clustered by：以哪个字段分桶。对id进行哈希取值，随机 地放到4个桶里。
-----------------------------
准备数据
[root@hello110 data]# vi bucket_test
1
2
3
4
5
6
.............
.........
导入数据
正确的导入方式：从日常保存数据的表insert
[21:27:45]hive (zmgdb)> create table t2(id string);
[21:27:45]OK
[21:27:45]Time taken: 0.073 seconds
[21:28:24]hive (zmgdb)> load data local inpath '/data/bucket_test' into table t2;
[21:28:24]Loading data to table zmgdb.t2
[21:28:25]OK
从日常表导入
[22:39:47]hive (zmgdb)> insert overwrite table bucket_t1 select id from t2;
hive会启动mapreduce
[22:39:48]WARNING: Hive-on-MR is deprecated in Hive 2 and may not be available in the future versions. Consider using a different execution engine (i.e. spark, tez) or using Hive 1.X releases.
[22:39:48]Query ID = hadoop_20160922063946_34bf30c4-3f23-43e9-ad8f-edd5ee214948
[22:39:48]Total jobs = 1
[22:39:48]Launching Job 1 out of 1
[22:39:48]Number of reduce tasks determined at compile time: 6
[22:39:48]In order to change the average load for a reducer (in bytes):
[22:39:48] set hive.exec.reducers.bytes.per.reducer=<number>
[22:39:48]In order to limit the maximum number of reducers:
[22:39:48] set hive.exec.reducers.max=<number>
[22:39:48]In order to set a constant number of reducers:
[22:39:48] set mapreduce.job.reduces=<number>
[22:39:51]Starting Job = job_1474497386931_0001, Tracking URL = http://hello110:8088/proxy/application_1474497386931_0001/
[22:39:51]Kill Command = /home/hadoop/app/hadoop-2.7.2/bin/hadoop job -kill job_1474497386931_0001
[22:39:59]Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 6
[22:39:59]2016-09-22 06:39:59,419 Stage-1 map = 0%, reduce = 0%
[22:40:06]2016-09-22 06:40:05,828 Stage-1 map = 100%, reduce = 0%, Cumulative CPU 1.63 sec
[22:40:12]2016-09-22 06:40:12,347 Stage-1 map = 100%, reduce = 17%, Cumulative CPU 3.48 sec
[22:40:16]2016-09-22 06:40:15,739 Stage-1 map = 100%, reduce = 33%, Cumulative CPU 5.4 sec
[22:40:17]2016-09-22 06:40:16,807 Stage-1 map = 100%, reduce = 50%, Cumulative CPU 7.52 sec
[22:40:19]2016-09-22 06:40:18,929 Stage-1 map = 100%, reduce = 83%, Cumulative CPU 11.35 sec
[22:40:20]2016-09-22 06:40:19,991 Stage-1 map = 100%, reduce = 100%, Cumulative CPU 13.19 sec
[22:40:21]MapReduce Total cumulative CPU time: 13 seconds 190 msec
[22:40:21]Ended Job = job_1474497386931_0001
[22:40:21]Loading data to table zmgdb.bucket_t1
[22:40:22]MapReduce Jobs Launched:
[22:40:22]Stage-Stage-1: Map: 1 Reduce: 6 Cumulative CPU: 13.19 sec HDFS Read: 25355 HDFS Write: 1434 SUCCESS
[22:40:22]Total MapReduce CPU Time Spent: 13 seconds 190 msec
[22:40:22]OK
[22:40:22]id
[22:40:22]Time taken: 34.91 seconds
错误的导入方式：从文件load data
hive (zmgdb)> create table bucket_t2 like bucket_t1;
OK
Time taken: 0.707 seconds
hive (zmgdb)> load data local inpath '/data/bucket_test' into table bucket_t2;
Loading data to table zmgdb.bucket_t2
OK
Time taken: 1.485 seconds
没有启动mapreduce对数据进行哈希取值，只是简单的原样导入，没有起到抽样查询的目的。通过select * from 比较会发现bucket_t1的数据和bucket_t2的数据顺序是不同的，bucket_t2的表顺序与原数据文件顺序一致，没有做过哈希取值。
查询
select * from bucket_table tablesample(bucket x out of y on column);
tablesample是抽样语句
语法解析：TABLESAMPLE(BUCKET x OUT OF y on 字段)
y必须是table总bucket数的倍数或者因子。
hive根据y的大小，决定抽样的比例。
例如，table总共分了64份，当y=32时，抽取(64/32=)2个bucket的数据，当y=128时，抽取(64/128=)1/2个bucket的数据。x表示从哪个bucket开始抽取。
例如，table总bucket数为32，tablesample(bucket 3 out of 16)，表示总共抽取（32/16=）2个bucket的数据，分别为第3个bucket和第（3+16=）19个bucket的数据。如果是y=64，则抽取半个第3个桶的值。
[22:44:31]hive (zmgdb)> select * from bucket_t1 tablesample (bucket 1 out of 6 on id);
[22:44:31]OK
[22:44:31]bucket_t1.id
[22:44:31]6
[22:44:31]iu
[22:44:31]0
[22:44:31]6
[22:44:31]hj
[22:44:31]6
[22:44:31]6
[22:44:31]51
[22:44:31]
[22:44:31]
[22:44:31]r
[22:44:31]99
[22:44:31]0
[22:44:31]57
[22:44:31]loo
[22:44:31]r
[22:44:31]r
[22:44:31]r
[22:44:31]60
[22:44:31]66
[22:44:31]75
[22:44:31]6
[22:44:31]84
[22:44:31]x
[22:44:31]24
[22:44:31]93
[22:44:31]99
[22:44:31]105
[22:44:31]f
[22:44:31]r
[22:44:31]114
[22:44:31]0
[22:44:31]123
[22:44:31]129
[22:44:31]132
[22:44:31]x
[22:44:31]138
[22:44:31]141
[22:44:31]147
[22:44:31]33
[22:44:31]150
[22:44:31]156
[22:44:31]r
[22:44:31]f
[22:44:31]39
[22:44:31]15
[22:44:31]r
[22:44:31]ddd
[22:44:31]
[22:44:31]06
[22:44:31]hj
[22:44:31]f
[22:44:31]l
[22:44:31]f
[22:44:31]f
[22:44:31]f
[22:44:31]f
[22:44:31]42
[22:44:31]f
[22:44:31]r
[22:44:31]r
[22:44:31]f
[22:44:31]f
[22:44:31]r
[22:44:31]48
[22:44:31]6
[22:44:31]Time taken: 0.142 seconds, Fetched:66 row(s)
[22:44:43]hive (zmgdb)> select * from bucket_t1 tablesample (bucket 1 out of 60 on id);
[22:44:43]OK
[22:44:43]bucket_t1.id
[22:44:43]
[22:44:43]
[22:44:43]loo
[22:44:43]x
[22:44:43]114
[22:44:43]132
[22:44:43]x
[22:44:43]150
[22:44:43]ddd
[22:44:43]
[22:44:43]Time taken: 0.064 seconds, Fetched: 10 row(s)
原文参考：https://blog.csdn.net/zengmingen/article/details/52620913