原帖地址: http://blog.csdn/nsrainbow/article/details/41748863 最新课程请关注原作者博客
声明
- 本文基于Centos 6.x + CDH 5.x
Hive是什么
Hive 提供了一个让大家可以使用sql去查询数据的途径。让大家可以在hadoop上写sql语句。但是最好不要拿Hive进行实时的查询。因为Hive的实现原理是把sql语句转化为多个Map Reduce任务所以Hive非常慢,官方文档说Hive 适用于高延时性的场景而且很费资源。
举个简单的例子,可以像这样去查询
hive> select * from h_employee;
OK
1 1 peter
2 2 paul
Time taken: 9.289 seconds, Fetched: 2 row(s)
这个h_employee不一定是一个数据库表,有可能只是一个针对csv文件的元数据映射。
Hive 安装
相比起很多教程先介绍概念,我喜欢先动手装上,然后用例子来介绍概念。我们先来安装一下Hive
先确认是否已经安装了对应的yum源,如果没有照这个教程里面写的安装cdh的yum源http://blog.csdn/nsrainbow/article/details/36629339
hive 基本包
yum install hive -y
hive metastore
yum install hive-metastore
hive服务端
yum install hive-server2 -y
如果要跟hbase通讯就安装 hive-hbase
yum install hive-hbase -y
Hive metastore 服务
3种模式
hive metastore(元数据) 服务用来存储 Hive 表的元数据和分区。下面会介绍metastore的概念,现在先搞定安装再说。hive 存储 metastore有3种模式内置存储模式
内置存储用的是derby作为数据库,但是这个derby很挫啊,一个纯java的数据库,同时只能有一个会话,存粹测试玩玩。所以我们说下第二种模式
本地存储模式
在这种模式下,hive metastore 服务跟HiveServer进程共用一个进程,但是会另起一个线程来运行元数据数据库,这个线程有可能在另外一个机器上。内置的metastore服务跟metastore数据库之间通过JDBC交互。比上一个方案更进一步了,但是还是不够好,因为hive metastore跟HiveServer还共用一个进程呢,于是来介绍下CDH强烈推荐的第三种模式
远程模式
使用mysql作为metastore数据库
我们选择mysql作为metastore的数据库安装mysql
如果你的机器上已经安装过mysql可以跳过这一步yum install mysql-server
启动服务
service mysqld start
添加到自启动
chkconfig mysqld on
初始化mysql的一些参数,比如root用户的密码等
$ sudo /usr/bin/mysql_secure_installation
[...]
Enter current password for root (enter for none):
OK, successfully used password, moving on...
[...]
Set root password? [Y/n] y
New password:
Re-enter new password:
Remove anonymous users? [Y/n] Y
[...]
Disallow root login remotely? [Y/n] N
[...]
Remove test database and access to it [Y/n] Y
[...]
Reload privilege tables now? [Y/n] Y
All done!
安装mysql JDBC驱动
$ sudo yum install mysql-connector-java
$ ln -s /usr/share/java/mysql-connector-java.jar /usr/lib/hive/lib/mysql-connector-java.jar
第二步是把驱动建立一个软链到hive的lib库里面,让hive可以加载
创建metastore需要的用户和库
创建metastore库$ mysql -u root -p
Enter password:
mysql> CREATE DATABASE metastore;
mysql> USE metastore;
mysql> SOURCE /usr/lib/hive/scripts/metastore/upgrade/mysql/hive-schema-0.13.0.mysql.sql;
创建hive用户
官方给的例子是
mysql> CREATE USER 'hive'@'metastorehost' IDENTIFIED BY 'mypassword';
...
mysql> REVOKE ALL PRIVILEGES, GRANT OPTION FROM 'hive'@'metastorehost';
mysql> GRANT ALL PRIVILEGES ON metastore.* TO 'hive'@'metastorehost';
mysql> FLUSH PRIVILEGES;
这边metastorehost换成你metastore的机器的host名字,mypassword换成你想设定的密码
在本例子中是这样
mysql> CREATE USER 'hive'@'%' IDENTIFIED BY 'hive';
mysql> REVOKE ALL PRIVILEGES, GRANT OPTION FROM 'hive'@'%';
mysql> GRANT ALL PRIVILEGES ON metastore.* TO 'hive'@'%';
mysql> FLUSH PRIVILEGES;
mysql> quit;
配置hive
编辑 /usr/lib/hive/conf/hive-site.xml- 假设你安装mysql的机器名叫host1,在 javax.jdo.option.ConnectionURL 中配置上jdbc连接
- hive.metastore.uris 这个参数必须用ip,不懂为什么
- hive.metastore.schema.verification 官方建议用true,官方说新旧版本的hive数据结构差别很大,要打开验证,免得出错
<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<configuration>
<property>
<name>javax.jdo.option.ConnectionURL</name>
<value>jdbc:mysql://host1/metastore</value>
<description>the URL of the MySQL database</description>
</property>
<property>
<name>javax.jdo.option.ConnectionDriverName</name>
<value>com.mysql.jdbc.Driver</value>
</property>
<property>
<name>javax.jdo.option.ConnectionUserName</name>
<value>hive</value>
</property>
<property>
<name>javax.jdo.option.ConnectionPassword</name>
<value>hive</value>
</property>
<property>
<name>datanucleus.autoCreateSchema</name>
<value>false</value>
</property>
<property>
<name>datanucleus.fixedDatastore</name>
<value>true</value>
</property>
<property>
<name>datanucleus.autoStartMechanism</name>
<value>SchemaTable</value>
</property>
<property>
<name>hive.metastore.uris</name>
<value>thrift://192.168.199.126:9083</value>
<description>IP address (or fully-qualified domain name) and port of the metastore host</description>
</property>
<property>
<name>hive.metastore.schema.verification</name>
<value>true</value>
</property>
</configuration>
配置HiveServer2
编辑 /etc/hive/conf/hive-site.xml 增加或者修改这两项<property>
<name>hive.support.concurrency</name>
<description>Enable Hive's Table Lock Manager Service</description>
<value>true</value>
</property>
<property>
<name>hive.zookeeper.quorum</name>
<description>Zookeeper quorum used by Hive's Table Lock Manager</description>
<value>host1,host2</value>
</property>
如果你修改了zookeeper 的默认端口就增加或修改这个属性
<property>
<name>hive.zookeeper.client.port</name>
<value>2222</value>
<description>
The port at which the clients will connect.
</description>
</property>
启动服务
启动顺序是 hive-metastore -> hive-server2service hive-metastore start
service hive-server2 start
启动的时候遇到问题
我遇到了一个问题,启动的时候报错Starting Hive Metastore Server
Error creating temp dir in hadoop.tmp.dir /data/hdfs/tmp due to Permission denied
给 /tmp 文件夹一个写权限就好了
cd /data/hdfs
chmod a+rwx tmp
测试是否安装成功
使用hive进入客户端$ hive
hive>
hive> show tables;
OK
Time taken: 10.345 seconds
Hive使用
metastore
Hive 中建立的表都叫metastore表。这些表并不真实的存储数据,而是定义真实数据跟hive之间的映射,就像传统数据库中表的meta信息,所以叫做metastore。实际存储的时候可以定义的存储模式有四种:
- 内部表(默认)
- 分区表
- 桶表
- 外部表
CREATE TABLE worker(id INT, name STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '\054';
这个语句的意思是建立一个worker的内部表,内部表是默认的类型,所以不用写存储的模式。并且使用逗号作为分隔符存储
建表语句支持的类型
基本数据类型tinyint / smalint / int /bigint
float / double
boolean
string
复杂数据类型
Array/Map/Struct
没有date /datetime
建完的表存在哪里呢?
在 /user/hive/warehouse 里面,可以通过hdfs来查看建完的表位置$ hdfs dfs -ls /user/hive/warehouse
Found 11 items
drwxrwxrwt - root supergroup 0 2014-12-02 14:42 /user/hive/warehouse/h_employee
drwxrwxrwt - root supergroup 0 2014-12-02 14:42 /user/hive/warehouse/h_employee2
drwxrwxrwt - wlsuser supergroup 0 2014-12-04 17:21 /user/hive/warehouse/h_employee_export
drwxrwxrwt - root supergroup 0 2014-08-18 09:20 /user/hive/warehouse/h_http_access_logs
drwxrwxrwt - root supergroup 0 2014-06-30 10:15 /user/hive/warehouse/hbase_apache_access_log
drwxrwxrwt - username supergroup 0 2014-06-27 17:48 /user/hive/warehouse/hbase_table_1
drwxrwxrwt - username supergroup 0 2014-06-30 09:21 /user/hive/warehouse/hbase_table_2
drwxrwxrwt - username supergroup 0 2014-06-30 09:43 /user/hive/warehouse/hive_apache_accesslog
drwxrwxrwt - root supergroup 0 2014-12-02 15:12 /user/hive/warehouse/hive_employee
一个文件夹对应一个metastore表
Hive 各种类型表使用
内部表
CREATE TABLE workers( id INT, name STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '\054';
通过这样的语句就建立了一个内部表叫 workers,并且分隔符是逗号, \054 是ASCII 码
我们可以通过 show tables; 来看看有多少表,其实hive的很多语句是模仿mysql的,当你们不知道语句的时候,把mysql的语句拿来基本可以用。除了limit比较怪,这个后面会说
hive> show tables;
OK
h_employee
h_employee2
h_employee_export
h_http_access_logs
hive_employee
workers
Time taken: 0.371 seconds, Fetched: 6 row(s)
建立完后,我们试着插入几条数据。这边要告诉大家Hive不支持单句插入的语句,必须批量,所以不要指望能用insert into workers values (1,'jack') 这样的语句插入数据。hive支持的插入数据的方式有两种:
- 从文件读取数据
- 从别的表读出数据插入(insert from select)
$ cat workers.csv
1,jack
2,terry
3,michael
用LOAD DATA 导入到Hive的表中
hive> LOAD DATA LOCAL INPATH '/home/alex/workers.csv' INTO TABLE workers;
Copying data from file:/home/alex/workers.csv
Copying file: file:/home/alex/workers.csv
Loading data to table default.workers
Table default.workers stats: [num_partitions: 0, num_files: 1, num_rows: 0, total_size: 25, raw_data_size: 0]
OK
Time taken: 0.655 seconds
注意
- 不要少了那个 LOCAL , LOAD DATA LOCAL INPATH 跟 LOAD DATA INPATH 的区别是一个是从你本地磁盘上找源文件,一个是从hdfs上找文件
- 如果加上OVERWRITE可以再导入之前先清空表,比如 LOAD DATA LOCAL INPATH '/home/alex/workers.csv' OVERWRITE INTO TABLE workers;
hive> select * from workers;
OK
1 jack
2 terry
3 michael
Time taken: 0.177 seconds, Fetched: 3 row(s)
我们去看下导入后在hive内部表是怎么存的
# hdfs dfs -ls /user/hive/warehouse/workers/
Found 1 items
-rwxrwxrwt 2 root supergroup 25 2014-12-08 15:23 /user/hive/warehouse/workers/workers.csv
原来就是原封不动的把文件拷贝进去啊!就是这么土! 我们可以试验再放一个文件 workers2.txt (我故意把扩展名换一个,其实hive是不看扩展名的)
# cat workers2.txt
4,peter
5,kate
6,ted
导入
hive> LOAD DATA LOCAL INPATH '/home/alex/workers2.txt' INTO TABLE workers;
Copying data from file:/home/alex/workers2.txt
Copying file: file:/home/alex/workers2.txt
Loading data to table default.workers
Table default.workers stats: [num_partitions: 0, num_files: 2, num_rows: 0, total_size: 46, raw_data_size: 0]
OK
Time taken: 0.79 seconds
去看下文件的存储结构
# hdfs dfs -ls /user/hive/warehouse/workers/
Found 2 items
-rwxrwxrwt 2 root supergroup 25 2014-12-08 15:23 /user/hive/warehouse/workers/workers.csv
-rwxrwxrwt 2 root supergroup 21 2014-12-08 15:29 /user/hive/warehouse/workers/workers2.txt
多出来一个workers2.txt 再用sql查询下
hive> select * from workers;
OK
1 jack
2 terry
3 michael
4 peter
5 kate
6 ted
Time taken: 0.144 seconds, Fetched: 6 row(s)
分区表
分区表是用来加速查询的,比如你的数据非常多,但是你的应用场景是基于这些数据做日报表,那你就可以根据日进行分区,当你要做2014-05-05的报表的时候只需要加载2014-05-05这一天的数据就行了。我们来创建一个分区表来看下create table partition_employee(id int, name string)
partitioned by(daytime string)
row format delimited fields TERMINATED BY '\054';
可以看到分区的属性,并不是任何一个列 我们先建立2个测试数据文件,分别对应两天的数据
# cat 2014-05-05
22,kitty
33,lily
# cat 2014-05-06
14,sami
45,micky
导入到分区表里面
hive> LOAD DATA LOCAL INPATH '/home/alex/2014-05-05' INTO TABLE partition_employee partition(daytime='2014-05-05');
Copying data from file:/home/alex/2014-05-05
Copying file: file:/home/alex/2014-05-05
Loading data to table default.partition_employee partition (daytime=2014-05-05)
Partition default.partition_employee{daytime=2014-05-05} stats: [num_files: 1, num_rows: 0, total_size: 21, raw_data_size: 0]
Table default.partition_employee stats: [num_partitions: 1, num_files: 1, num_rows: 0, total_size: 21, raw_data_size: 0]
OK
Time taken: 1.154 seconds
hive> LOAD DATA LOCAL INPATH '/home/alex/2014-05-06' INTO TABLE partition_employee partition(daytime='2014-05-06');
Copying data from file:/home/alex/2014-05-06
Copying file: file:/home/alex/2014-05-06
Loading data to table default.partition_employee partition (daytime=2014-05-06)
Partition default.partition_employee{daytime=2014-05-06} stats: [num_files: 1, num_rows: 0, total_size: 21, raw_data_size: 0]
Table default.partition_employee stats: [num_partitions: 2, num_files: 2, num_rows: 0, total_size: 42, raw_data_size: 0]
OK
Time taken: 0.763 seconds
导入的时候通过 partition 来指定分区。
查询的时候通过指定分区来查询
hive> select * from partition_employee where daytime='2014-05-05';
OK
22 kitty 2014-05-05
33 lily 2014-05-05
Time taken: 0.173 seconds, Fetched: 2 row(s)
我的查询语句并没有什么特别的语法,hive 会自动判断你的where语句中是否包含分区的字段。而且可以使用大于小于等运算符
hive> select * from partition_employee where daytime>='2014-05-05';
OK
22 kitty 2014-05-05
33 lily 2014-05-05
14 sami 2014-05-06
45 mick' 2014-05-06
Time taken: 0.273 seconds, Fetched: 4 row(s)
我们去看看存储的结构
# hdfs dfs -ls /user/hive/warehouse/partition_employee
Found 2 items
drwxrwxrwt - root supergroup 0 2014-12-08 15:57 /user/hive/warehouse/partition_employee/daytime=2014-05-05
drwxrwxrwt - root supergroup 0 2014-12-08 15:57 /user/hive/warehouse/partition_employee/daytime=2014-05-06
我们试试二维的分区表
create table p_student(id int, name string)
partitioned by(daytime string,country string)
row format delimited fields TERMINATED BY '\054';
查入一些数据
# cat 2014-09-09-CN
1,tammy
2,eric
# cat 2014-09-10-CN
3,paul
4,jolly
# cat 2014-09-10-EN
44,ivan
66,billy
导入hive
hive> LOAD DATA LOCAL INPATH '/home/alex/2014-09-09-CN' INTO TABLE p_student partition(daytime='2014-09-09',country='CN');
Copying data from file:/home/alex/2014-09-09-CN
Copying file: file:/home/alex/2014-09-09-CN
Loading data to table default.p_student partition (daytime=2014-09-09, country=CN)
Partition default.p_student{daytime=2014-09-09, country=CN} stats: [num_files: 1, num_rows: 0, total_size: 19, raw_data_size: 0]
Table default.p_student stats: [num_partitions: 1, num_files: 1, num_rows: 0, total_size: 19, raw_data_size: 0]
OK
Time taken: 0.736 seconds
hive> LOAD DATA LOCAL INPATH '/home/alex/2014-09-10-CN' INTO TABLE p_student partition(daytime='2014-09-10',country='CN');
Copying data from file:/home/alex/2014-09-10-CN
Copying file: file:/home/alex/2014-09-10-CN
Loading data to table default.p_student partition (daytime=2014-09-10, country=CN)
Partition default.p_student{daytime=2014-09-10, country=CN} stats: [num_files: 1, num_rows: 0, total_size: 19, raw_data_size: 0]
Table default.p_student stats: [num_partitions: 2, num_files: 2, num_rows: 0, total_size: 38, raw_data_size: 0]
OK
Time taken: 0.691 seconds
hive> LOAD DATA LOCAL INPATH '/home/alex/2014-09-10-EN' INTO TABLE p_student partition(daytime='2014-09-10',country='EN');
Copying data from file:/home/alex/2014-09-10-EN
Copying file: file:/home/alex/2014-09-10-EN
Loading data to table default.p_student partition (daytime=2014-09-10, country=EN)
Partition default.p_student{daytime=2014-09-10, country=EN} stats: [num_files: 1, num_rows: 0, total_size: 21, raw_data_size: 0]
Table default.p_student stats: [num_partitions: 3, num_files: 3, num_rows: 0, total_size: 59, raw_data_size: 0]
OK
Time taken: 0.622 seconds
看看存储结构
# hdfs dfs -ls /user/hive/warehouse/p_student
Found 2 items
drwxr-xr-x - root supergroup 0 2014-12-08 16:10 /user/hive/warehouse/p_student/daytime=2014-09-09
drwxr-xr-x - root supergroup 0 2014-12-08 16:10 /user/hive/warehouse/p_student/daytime=2014-09-10
# hdfs dfs -ls /user/hive/warehouse/p_student/daytime=2014-09-09
Found 1 items
drwxr-xr-x - root supergroup 0 2014-12-08 16:10 /user/hive/warehouse/p_student/daytime=2014-09-09/country=CN
查询一下数据
hive> select * from p_student;
OK
1 tammy 2014-09-09 CN
2 eric 2014-09-09 CN
3 paul 2014-09-10 CN
4 jolly 2014-09-10 CN
44 ivan 2014-09-10 EN
66 billy 2014-09-10 EN
Time taken: 0.228 seconds, Fetched: 6 row(s)
hive> select * from p_student where daytime='2014-09-10' and country='EN';
OK
44 ivan 2014-09-10 EN
66 billy 2014-09-10 EN
Time taken: 0.224 seconds, Fetched: 2 row(s)
桶表
桶表是根据某个字段的hash值,来将数据扔到不同的“桶”里面。外国人有个习惯,就是分类东西的时候摆几个桶,上面贴不同的标签,所以他们取名的时候把这种表形象的取名为桶表。 桶表表专门用于采样分析下面这个例子是官网教程直接拷贝下来的,因为分区表跟桶表是可以同时使用的,所以这个例子中同时使用了分区跟桶两种特性
CREATE TABLE b_student(id INT, name STRING)
PARTITIONED BY(dt STRING, country STRING)
CLUSTERED BY(id) SORTED BY(name) INTO 4 BUCKETS
row format delimited
fields TERMINATED BY '\054';
意思是根据userid来进行计算hash值,用viewTIme来排序存储 做数据跟导入的过程我就不在赘述了,这是导入后的数据
hive> select * from b_student;
OK
1 tammy 2014-09-09 CN
2 eric 2014-09-09 CN
3 paul 2014-09-10 CN
4 jolly 2014-09-10 CN
34 allen 2014-09-11 EN
Time taken: 0.727 seconds, Fetched: 5 row(s)
从4个桶中采样抽取一个桶的数据
hive> select * from b_student tablesample(bucket 1 out of 4 on id);
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Starting Job = job_1406097234796_0041, Tracking URL = http://hadoop01:8088/proxy/application_1406097234796_0041/
Kill Command = /usr/lib/hadoop/bin/hadoop job -kill job_1406097234796_0041
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 0
2014-12-08 17:35:56,995 Stage-1 map = 0%, reduce = 0%
2014-12-08 17:36:06,783 Stage-1 map = 100%, reduce = 0%, Cumulative CPU 2.9 sec
2014-12-08 17:36:07,845 Stage-1 map = 100%, reduce = 0%, Cumulative CPU 2.9 sec
MapReduce Total cumulative CPU time: 2 seconds 900 msec
Ended Job = job_1406097234796_0041
MapReduce Jobs Launched:
Job 0: Map: 1 Cumulative CPU: 2.9 sec HDFS Read: 482 HDFS Write: 22 SUCCESS
Total MapReduce CPU Time Spent: 2 seconds 900 msec
OK
4 jolly 2014-09-10 CN
外部表
外部表就是存储不是由hive来存储的,比如可以依赖Hbase来存储,hive只是做一个映射而已。我用Hbase来举例先建立一张Hbase表叫 employee
hbase(main):005:0> create 'employee','info'
0 row(s) in 0.4740 seconds
=> Hbase::Table - employee
hbase(main):006:0> put 'employee',1,'info:id',1
0 row(s) in 0.2080 seconds
hbase(main):008:0> scan 'employee'
ROW COLUMN+CELL
1 column=info:id, timestamp=1417591291730, value=1
1 row(s) in 0.0610 seconds
hbase(main):009:0> put 'employee',1,'info:name','peter'
0 row(s) in 0.0220 seconds
hbase(main):010:0> scan 'employee'
ROW COLUMN+CELL
1 column=info:id, timestamp=1417591291730, value=1
1 column=info:name, timestamp=1417591321072, value=peter
1 row(s) in 0.0450 seconds
hbase(main):011:0> put 'employee',2,'info:id',2
0 row(s) in 0.0370 seconds
hbase(main):012:0> put 'employee',2,'info:name','paul'
0 row(s) in 0.0180 seconds
hbase(main):013:0> scan 'employee'
ROW COLUMN+CELL
1 column=info:id, timestamp=1417591291730, value=1
1 column=info:name, timestamp=1417591321072, value=peter
2 column=info:id, timestamp=1417591500179, value=2
2 column=info:name, timestamp=1417591512075, value=paul
2 row(s) in 0.0440 seconds
建立外部表进行映射
hive> CREATE EXTERNAL TABLE h_employee(key int, id int, name string)
> STORED BY 'org.apache.hadoop.hive.hbase.HBaseStorageHandler'
> WITH SERDEPROPERTIES ("hbase.columns.mapping" = ":key, info:id,info:name")
> TBLPROPERTIES ("hbase.table.name" = "employee");
OK
Time taken: 0.324 seconds
hive> select * from h_employee;
OK
1 1 peter
2 2 paul
Time taken: 1.129 seconds, Fetched: 2 row(s)
查询语法
具体语法可以参考官方手册https://cwiki.apache/confluence/display/Hive/Tutorial 我只说几个比较奇怪的点显示条数
展示x条数据,用的还是limit,比如hive> select * from h_employee limit 1
> ;
OK
1 1 peter
Time taken: 0.284 seconds, Fetched: 1 row(s)
但是不支持起点,比如offset
参考资料
- http://www.cloudera/content/cloudera/en/documentation/core/v5-2-x/topics/cdh_ig_hiveserver2_configure.html
更多推荐
Alex 的 Hadoop 菜鸟教程: 第10课 Hive 安装和使用教程
发布评论