Hive索引

1、 Hive索引概述

Hive的索引目的是提高Hive表指定列的查询速度。

没有索引时，类似'WHERE tab1.col1 = 10' 的查询，Hive会加载整张表或分区，然后处理所有的rows，但是如果在字段col1上面存在索引时，那么只会加载和处理文件的一部分。

与其他传统数据库一样，增加索引在提升查询速度时，会消耗额外资源去创建索引和需要更多的磁盘空间存储索引。

Hive 0.7.0版本中，加入了索引。Hive 0.8.0版本中增加了bitmap索引。

2、 索引相关的配置参数

hive.index.compact.file.ignore.hdfs

Default Value: false

Added In: Hive 0.7.0 withHIVE-1889

在索引文件中存储的hdfs地址将在运行时被忽略，如果开启的话；如果数据被迁移，那么索引文件依然可用，默认是false

hive.optimize.index.filter

Default Value: false

Added In: Hive 0.8.0 withHIVE-1644

是否自动使用索引, 默认是false

hive.optimize.index.filter.compact.minsize

Default Value: 5368709120

Added In: Hive 0.8.0 withHIVE-1644

压缩索引自动应用的最小输入大小

hive.optimize.index.filter.compact.maxsize

Default Value: -1

Added In: Hive 0.8.0 withHIVE-1644

压缩索引自动应用的最大输入大小,负值代表正无穷

hive.index.compact.query.max.size

Default Value: 10737418240

Added In: Hive 0.8.0 withHIVE-2096

一个使用压缩索引做的查询能取到的最大数据量，默认是10737418240 个byte；负值代表无穷大；

hive.index.compact.query.max.entries

Default Value: 10000000

Added In: Hive 0.8.0 withHIVE-2096

使用压缩索引查询时能读到的最大索引项数，默认是10000000；负值代表无穷大；

hive.exec.concatenate.check.index

Default Value: true

Added In: Hive 0.8.0 withHIVE-2125

如果设置为true，那么在做ALTER TABLE tbl_name CONCATENATE on a table/partition（有索引） 操作时，抛出错误；可以帮助用户避免index的删除和重建；

hive.optimize.index.groupby

Default Value: false

Added In: Hive 0.8.1 withHIVE-1694

hive.index.compact.binary.search

Default Value: true

Added In: Hive 0.8.1with HIVE-2535

在索引表中是否开启二分搜索进行索引项查询，默认是true；

3、 索引示例

注意：在Hive 0.12.0以及之前版本中，索引名称在create index和drop index语句中是大小写敏感的。然而，alter index 需要一个小写的索引名字。

此bug在Hive 0.13.0解决，此版本开始使索引名字大小写不敏感。

对于Hive 0.13.0之前的版本，最好使用小写的索引名字。

下面介绍索引的常见用法：

A、 Create/build，show和drop index

create index table01_index ontable table01(column2) as 'COMPACT' with deferred rebuild;

show index on table01;

drop index table01_index ontable01;

B、 Create then build，show formatted和drop index

create index table02_index ontable table02(column3) as 'compact' with deferred rebuild;

alter index table02_index ontable02 rebuild;

show formatted index ontable02;

drop index table02_index ontable02;

C、 创建bitmap索引，build,show 和drop

createindex table03_index on table table03 (column4) as 'bitmap' with deferred rebuild;

alter index table03_index ontable03 rebuild;

show formatted index ontable03;

drop index table03_index on table03;

D、 在一张新表上创建索引

createindex table04_index on table table04 (column5) as 'compact'with deferred rebuild in tabletable04_index_table;

E、 创建索引，存储格式为RCFile

create index table05_index ontable table05 (column6) as 'compact' with deferred rebuildstored as rcfile;

F、 创建索引，存储格式为TextFile

create index table06_index ontable table06 (column7) as 'compact' with deferredrebuild row format delimited fields terminated by '\t' stored as textfile;

G、 创建带有索引属性的索引

create index table07_index ontable table07 (column8) as 'compact' with deferred rebuild idxproperties("prop1"="value1", "prop2"="value2");

H、 创建带有表属性的索引

create index table08_index ontable table08 (column9) as 'compact' withdeferred rebuild tblproperties("prop3"="value3", "prop4"="value4");

I、 如果索引存在，则删除

drop index if exists table09_indexon table09;

J、 在分区上重建索引

alter index table10_index on table10partition (columnx='valueq', columny='valuer') rebuild;

4、 索引测试

(1) 查询表中行数

hive (hive)> select count(1)from userbook;

4409365

(2) 表中未创建索引前查询

hive (hive)> select * fromuserbook where book_id = '15999998838';

Query ID =hadoop_20150627165551_595da79a-0e27-453b-9142-7734912934c4

Total jobs = 1

Launching Job 1 out of 1

Number of reduce tasks is setto 0 since there's no reduce operator

Starting Job =job_1435392961740_0012, Tracking URL =http://gpmaster:8088/proxy/application_1435392961740_0012/

Kill Command =/home/hadoop/hadoop-2.6.0/bin/hadoop job -kill job_1435392961740_0012

Hadoop job information forStage-1: number of mappers: 2; number of reducers: 0

2015-06-27 16:56:04,666 Stage-1map = 0%, reduce = 0%

2015-06-27 16:56:28,974 Stage-1map = 50%, reduce = 0%, Cumulative CPU4.36 sec

2015-06-27 16:56:31,123 Stage-1map = 78%, reduce = 0%, Cumulative CPU6.21 sec

2015-06-27 16:56:34,698 Stage-1map = 100%, reduce = 0%, Cumulative CPU7.37 sec

MapReduce Total cumulative CPUtime: 7 seconds 370 msec

Ended Job =job_1435392961740_0012

MapReduce Jobs Launched:

Stage-Stage-1: Map: 2 Cumulative CPU: 7.37 sec HDFS Read: 348355875 HDFS Write: 76 SUCCESS

Total MapReduce CPU Time Spent:7 seconds 370 msec

OK

userbook.book_id userbook.book_name userbook.author userbook.public_date userbook.address

15999998838 uviWfFJ KwCrDOA 2009-12-27 3b74416d-eb69-48e2-9d0d-09275064691b

Time taken: 45.678 seconds, Fetched: 1 row(s)

(3) 创建索引

hive (hive)> create indexuserbook_bookid_idx on table userbook(book_id) as 'COMPACT' WITH DEFERREDREBUILD;

(4) 创建索引后再执行查询

hive (hive)> select * fromuserbook where book_id = '15999998838';

Query ID =hadoop_20150627170019_5bb5514a-4c8e-4c47-9347-ed0657e1f2ff

Total jobs = 1

Launching Job 1 out of 1

Number of reduce tasks is setto 0 since there's no reduce operator

Starting Job =job_1435392961740_0013, Tracking URL = http://gpmaster:8088/proxy/application_1435392961740_0013/

Kill Command =/home/hadoop/hadoop-2.6.0/bin/hadoop job -kill job_1435392961740_0013

Hadoop job information forStage-1: number of mappers: 2; number of reducers: 0

2015-06-27 17:00:30,429 Stage-1map = 0%, reduce = 0%

2015-06-27 17:00:54,003 Stage-1map = 50%, reduce = 0%, Cumulative CPU7.43 sec

2015-06-27 17:00:56,181 Stage-1map = 78%, reduce = 0%, Cumulative CPU9.66 sec

2015-06-27 17:00:58,417 Stage-1map = 100%, reduce = 0%, Cumulative CPU10.83 sec

MapReduce Total cumulative CPUtime: 10 seconds 830 msec

Ended Job =job_1435392961740_0013

MapReduce Jobs Launched:

Stage-Stage-1: Map: 2 Cumulative CPU: 10.83 sec HDFS Read: 348356271 HDFS Write: 76 SUCCESS

Total MapReduce CPU Time Spent:10 seconds 830 msec

OK

userbook.book_id userbook.book_name userbook.author userbook.public_date userbook.address

15999998838 uviWfFJ KwCrDOA 2009-12-27 3b74416d-eb69-48e2-9d0d-09275064691b

Time taken: 40.549 seconds, Fetched: 1 row(s)

可以看到创建索引后，速度还是稍快一点的。

其实对于这种简单的查询，通过我们的设置，可以不用启动Map/Reduce的，而是启动Fetch task，直接从HDFS文件中filter过滤出需要的数据，需要设置如下参数：

set hive.fetch.task.conversion=more;

hive (hive)> select * fromuserbook where book_id = '15999998838';

OK

userbook.book_id userbook.book_name userbook.author userbook.public_date userbook.address

15999998838 uviWfFJ KwCrDOA 2009-12-27 3b74416d-eb69-48e2-9d0d-09275064691b

Time taken: 0.093 seconds,Fetched: 1 row(s)

可以看到速度更快了，毕竟省略掉了开启MR任务，执行效率提高不少。