[Sqoop]Sqoop使用

Sqoop的本质还是一个命令行工具，和HDFS，MapReduce相比，并没有什么高深的理论。

我们可以通过sqoop help命令来查看sqoop的命令选项，如下：

1. 16/11/13 20:10:17 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. usage: sqoop COMMAND [ARGS]
4. Available commands:
5.  codegen            Generate code to interact with database records
6.  create-hive-table  Import a table definition into Hive
7.  eval               Evaluate a SQL statement and display the results
8.  export             Export an HDFS directory to a database table
9.  help               List available commands
10.  import             Import a table from a database to HDFS
11.  import-all-tables  Import tables from a database to HDFS
12.  import-mainframe   Import datasets from a mainframe server to HDFS
13.  job                Work with saved jobs
14.  list-databases     List available databases on a server
15.  list-tables        List available tables in a database
16.  merge              Merge results of incremental imports
17.  metastore          Run a standalone Sqoop metastore
18.  version            Display version information
20. See 'sqoop help COMMAND' for information on a specific command.

其中使用频率最高的选项还是import 和 export 选项。

1. codegen

将关系型数据库表的记录映射为一个Java文件，Java class类以及相关的jar包，该命令将数据库表的记录映射为一个Java文件，在该Java文件中对应有表的各个字段。生成的jar和class文件在Metastore功能使用时会用到。该命令选项的参数如下图所示：

举例：

1. sqoop codegen --connect jdbc:mysql://localhost:3306/test --table order_info -outdir /home/xiaosi/test/ --username root -password root

上面实例以test数据库的order_info表来生成Java代码，其中-outdir指定了Java代码生成的路径

运行结果信息如下：

1. 16/11/13 21:50:34 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. Enter password:
3. 16/11/13 21:50:38 INFO manager.MySQLManager: Preparing to use a MySQL streaming resultset.
4. 16/11/13 21:50:38 INFO tool.CodeGenTool: Beginning code generation
5. 16/11/13 21:50:38 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `order_info` AS t LIMIT 1
6. 16/11/13 21:50:38 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `order_info` AS t LIMIT 1
7. 16/11/13 21:50:38 INFO orm.CompilationManager: HADOOP_MAPRED_HOME is /opt/hadoop-2.7.2
8. 注: /tmp/sqoop-xiaosi/compile/ea41fe40e1f12f6b052ad9fe4a5d9710/order_info.java使用或覆盖了已过时的 API。
9. 注: 有关详细信息, 请使用 -Xlint:deprecation 重新编译。
10. 16/11/13 21:50:39 INFO orm.CompilationManager: Writing jar file: /tmp/sqoop-xiaosi/compile/ea41fe40e1f12f6b052ad9fe4a5d9710/order_info.jar

我们还可以使用-bindir指定编译成的class文件以及将生成文件打包为jar的jar包文件输出路径：

1. 16/11/13 21:53:55 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. Enter password:
3. 16/11/13 21:53:58 INFO manager.MySQLManager: Preparing to use a MySQL streaming resultset.
4. 16/11/13 21:53:58 INFO tool.CodeGenTool: Beginning code generation
5. 16/11/13 21:53:58 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `order_info` AS t LIMIT 1
6. 16/11/13 21:53:58 INFO manager.SqlManager: Executing SQL statement: SELECT t.* FROM `order_info` AS t LIMIT 1
7. 16/11/13 21:53:58 INFO orm.CompilationManager: HADOOP_MAPRED_HOME is /opt/hadoop-2.7.2
8. 注: /home/xiaosi/data/order_info.java使用或覆盖了已过时的 API。
9. 注: 有关详细信息, 请使用 -Xlint:deprecation 重新编译。
10. 16/11/13 21:53:59 INFO orm.CompilationManager: Writing jar file: /home/xiaosi/data/order_info.jar

上面实例指定编译成的class文件（order_info.class）以及将生成文件打包为jar的jar包文件（order_info.jar）输出路径为/home/xiaosi/data路径，java文件（order_info.java）路径为/home/xiaosi/test

2. create-hive-table

这个命令上一篇文章[Sqoop导入与导出]中已经使用过了，作用就是生成与关系数据库表的表结构对应的Hive表。该命令选项的参数如下图所示：

举例：

1. sqoop create-hive-table --connect jdbc:mysql://localhost:3306/test --table employee --username root -password root --fields-terminated-by ','

3. eval

eval命令选项可以让Sqoop使用SQL语句对关系性数据库进行操作，在使用import这种工具进行数据导入的时候，可以预先了解相关的SQL语句是否正确，并能将结果显示在控制台。

3.1 选择查询评估计算

使用eval工具，我们可以评估计算任何类型的SQL查询。我们以test数据库的order_info表为例子：

1. sqoop eval --connect jdbc:mysql://localhost:3306/test --username root --query "select * from order_info limit 3" -P

运行结果信息：

1. 16/11/13 22:25:19 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. Enter password:
3. 16/11/13 22:25:22 INFO manager.MySQLManager: Preparing to use a MySQL streaming resultset.
4. ------------------------------------------------------------
5. | id                   | order_time           | business   |
6. ------------------------------------------------------------
7. | 358574046793404      | 2016-04-05           | flight     |
8. | 358574046794733      | 2016-08-03           | hotel      |
9. | 358574050631177      | 2016-05-08           | vacation   |
10. ------------------------------------------------------------

3.2 插入评估计算

Sqoop的eval工具可以适用于两个模拟和定义的SQL语句。这意味着，我们可以使用eval的INSERT语句了。下面的命令用于在test数据库的order_info表中插入新行：

1. sqoop eval --connect jdbc:mysql://localhost:3306/test --username root --query "insert into order_info (id, order_time, business) values('358574050631166', '2016-11-13', 'hotel')" -P

运行结果信息输出：

1. 16/11/13 22:29:42 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. Enter password:
3. 16/11/13 22:29:44 INFO manager.MySQLManager: Preparing to use a MySQL streaming resultset.
4. 16/11/13 22:29:44 INFO tool.EvalSqlTool: 1 row(s) updated.

1. mysql> select * from order_info where id = "358574050631166";
2. +-----------------+------------+----------+
3. | id              | order_time | business |
4. +-----------------+------------+----------+
5. | 358574050631166 | 2016-11-13 | hotel    |
6. +-----------------+------------+----------+
7. 1 row in set (0.00 sec)

4. export 

从HDFS中将数据导出到关系性数据库中。该命令选项的参数如下图所示：

举例：

在HDFS文件中的员工数据的一个例子，数据如下：

1. hadoop fs -text /user/xiaosi/employee/* | less
2. yoona,qunar,创新事业部
3. xiaosi,qunar,创新事业部
4. jim,ali,淘宝
5. kom,ali,淘宝
6. lucy,baidu,搜索
7. jim,ali,淘宝

在将HDFS中数据导出到关系性数据库时，必须在关系性数据库中新建一张来接受数据的表，如下：

1. CREATE TABLE `employee` (
2.  `name` varchar(255) DEFAULT NULL,
3.  `company` varchar(255) DEFAULT NULL,
4.  `depart` varchar(255) DEFAULT NULL
5. );

下面执行导出操作，命令如下：

1. sqoop export --connect jdbc:mysql://localhost:3306/test --table employee --export-dir /user/xiaosi/employee --username root -m 1 --fields-terminated-by ',' -P

运行结果信息输出：

1. 16/11/13 23:40:49 INFO mapreduce.Job: The url to track the job: http://localhost:8080/
2. 16/11/13 23:40:49 INFO mapreduce.Job: Running job: job_local611430785_0001
3. 16/11/13 23:40:49 INFO mapred.LocalJobRunner: OutputCommitter set in config null
4. 16/11/13 23:40:49 INFO mapred.LocalJobRunner: OutputCommitter is org.apache.sqoop.mapreduce.NullOutputCommitter
5. 16/11/13 23:40:49 INFO mapred.LocalJobRunner: Waiting for map tasks
6. 16/11/13 23:40:49 INFO mapred.LocalJobRunner: Starting task: attempt_local611430785_0001_m_000000_0
7. 16/11/13 23:40:49 INFO mapred.Task:  Using ResourceCalculatorProcessTree : [ ]
8. 16/11/13 23:40:49 INFO mapred.MapTask: Processing split: Paths:/user/xiaosi/employee/part-m-00000:0+120
9. 16/11/13 23:40:49 INFO Configuration.deprecation: map.input.file is deprecated. Instead, use mapreduce.map.input.file
10. 16/11/13 23:40:49 INFO Configuration.deprecation: map.input.start is deprecated. Instead, use mapreduce.map.input.start
11. 16/11/13 23:40:49 INFO Configuration.deprecation: map.input.length is deprecated. Instead, use mapreduce.map.input.length
12. 16/11/13 23:40:49 INFO mapreduce.AutoProgressMapper: Auto-progress thread is finished. keepGoing=false
13. 16/11/13 23:40:49 INFO mapred.LocalJobRunner:
14. 16/11/13 23:40:49 INFO mapred.Task: Task:attempt_local611430785_0001_m_000000_0 is done. And is in the process of committing
15. 16/11/13 23:40:49 INFO mapred.LocalJobRunner: map
16. 16/11/13 23:40:49 INFO mapred.Task: Task 'attempt_local611430785_0001_m_000000_0' done.
17. 16/11/13 23:40:49 INFO mapred.LocalJobRunner: Finishing task: attempt_local611430785_0001_m_000000_0
18. 16/11/13 23:40:49 INFO mapred.LocalJobRunner: map task executor complete.
19. 16/11/13 23:40:50 INFO mapreduce.Job: Job job_local611430785_0001 running in uber mode : false
20. 16/11/13 23:40:50 INFO mapreduce.Job:  map 100% reduce 0%
21. 16/11/13 23:40:50 INFO mapreduce.Job: Job job_local611430785_0001 completed successfully
22. 16/11/13 23:40:50 INFO mapreduce.Job: Counters: 20
23. File System Counters
24. FILE: Number of bytes read=22247825
25. FILE: Number of bytes written=22732498
26. FILE: Number of read operations=0
27. FILE: Number of large read operations=0
28. FILE: Number of write operations=0
29. HDFS: Number of bytes read=126
30. HDFS: Number of bytes written=0
31. HDFS: Number of read operations=12
32. HDFS: Number of large read operations=0
33. HDFS: Number of write operations=0
34. Map-Reduce Framework
35. Map input records=6
36. Map output records=6
37. Input split bytes=136
38. Spilled Records=0
39. Failed Shuffles=0
40. Merged Map outputs=0
41. GC time elapsed (ms)=0
42. Total committed heap usage (bytes)=245366784
43. File Input Format Counters
44. Bytes Read=0
45. File Output Format Counters
46. Bytes Written=0
47. 16/11/13 23:40:50 INFO mapreduce.ExportJobBase: Transferred 126 bytes in 2.3492 seconds (53.6344 bytes/sec)
48. 16/11/13 23:40:50 INFO mapreduce.ExportJobBase: Exported 6 records.

1. mysql> select name, company from employee;
2. +--------+---------+
3. | name   | company |
4. +--------+---------+
5. | yoona  | qunar   |
6. | xiaosi | qunar   |
7. | jim    | ali     |
8. | kom    | ali     |
9. | lucy   | baidu   |
10. | jim    | ali     |
11. +--------+---------+
12. 6 rows in set (0.00 sec)

5. import 

将数据表中的数据导入HDFS或者Hive中，该命令选项的参数如下图所示：

举例：

1. sqoop import --connect jdbc:mysql://localhost:3306/test --target-dir /user/xiaosi/data/order_info --query 'select * from order_info where $CONDITIONS' -m 1 --username root -P

如上代码从查询结果中导入数据到HDFS中，存储路径由--target-dir参数指定。这里，使用了--query选项，不能同时与--table选项使用。同时，变量$CONDITIONS必须在WHERE语句之后，供Sqoop进程运行命令过程中使用。

运行结果信息如下：

1. 16/11/14 12:08:50 INFO mapreduce.Job: The url to track the job: http://localhost:8080/
2. 16/11/14 12:08:50 INFO mapreduce.Job: Running job: job_local127577466_0001
3. 16/11/14 12:08:50 INFO mapred.LocalJobRunner: OutputCommitter set in config null
4. 16/11/14 12:08:50 INFO output.FileOutputCommitter: File Output Committer Algorithm version is 1
5. 16/11/14 12:08:50 INFO mapred.LocalJobRunner: OutputCommitter is org.apache.hadoop.mapreduce.lib.output.FileOutputCommitter
6. 16/11/14 12:08:50 INFO mapred.LocalJobRunner: Waiting for map tasks
7. 16/11/14 12:08:50 INFO mapred.LocalJobRunner: Starting task: attempt_local127577466_0001_m_000000_0
8. 16/11/14 12:08:50 INFO output.FileOutputCommitter: File Output Committer Algorithm version is 1
9. 16/11/14 12:08:50 INFO mapred.Task:  Using ResourceCalculatorProcessTree : [ ]
10. 16/11/14 12:08:50 INFO db.DBInputFormat: Using read commited transaction isolation
11. 16/11/14 12:08:50 INFO mapred.MapTask: Processing split: 1=1 AND 1=1
12. 16/11/14 12:08:50 INFO db.DBRecordReader: Working on split: 1=1 AND 1=1
13. 16/11/14 12:08:50 INFO db.DBRecordReader: Executing query: select * from order_info where ( 1=1 ) AND ( 1=1 )
14. 16/11/14 12:08:50 INFO mapreduce.AutoProgressMapper: Auto-progress thread is finished. keepGoing=false
15. 16/11/14 12:08:50 INFO mapred.LocalJobRunner:
16. 16/11/14 12:08:51 INFO mapred.Task: Task:attempt_local127577466_0001_m_000000_0 is done. And is in the process of committing
17. 16/11/14 12:08:51 INFO mapred.LocalJobRunner:
18. 16/11/14 12:08:51 INFO mapred.Task: Task attempt_local127577466_0001_m_000000_0 is allowed to commit now
19. 16/11/14 12:08:51 INFO output.FileOutputCommitter: Saved output of task 'attempt_local127577466_0001_m_000000_0' to hdfs://localhost:9000/user/xiaosi/data/order_info/_temporary/0/task_local127577466_0001_m_000000
20. 16/11/14 12:08:51 INFO mapred.LocalJobRunner: map
21. 16/11/14 12:08:51 INFO mapred.Task: Task 'attempt_local127577466_0001_m_000000_0' done.
22. 16/11/14 12:08:51 INFO mapred.LocalJobRunner: Finishing task: attempt_local127577466_0001_m_000000_0
23. 16/11/14 12:08:51 INFO mapred.LocalJobRunner: map task executor complete.
24. 16/11/14 12:08:51 INFO mapreduce.Job: Job job_local127577466_0001 running in uber mode : false
25. 16/11/14 12:08:51 INFO mapreduce.Job:  map 100% reduce 0%
26. 16/11/14 12:08:51 INFO mapreduce.Job: Job job_local127577466_0001 completed successfully
27. 16/11/14 12:08:51 INFO mapreduce.Job: Counters: 20
28. File System Counters
29. FILE: Number of bytes read=22247784
30. FILE: Number of bytes written=22732836
31. FILE: Number of read operations=0
32. FILE: Number of large read operations=0
33. FILE: Number of write operations=0
34. HDFS: Number of bytes read=0
35. HDFS: Number of bytes written=3710
36. HDFS: Number of read operations=4
37. HDFS: Number of large read operations=0
38. HDFS: Number of write operations=3
39. Map-Reduce Framework
40. Map input records=111
41. Map output records=111
42. Input split bytes=87
43. Spilled Records=0
44. Failed Shuffles=0
45. Merged Map outputs=0
46. GC time elapsed (ms)=0
47. Total committed heap usage (bytes)=245366784
48. File Input Format Counters
49. Bytes Read=0
50. File Output Format Counters
51. Bytes Written=3710
52. 16/11/14 12:08:51 INFO mapreduce.ImportJobBase: Transferred 3.623 KB in 2.5726 seconds (1.4083 KB/sec)
53. 16/11/14 12:08:51 INFO mapreduce.ImportJobBase: Retrieved 111 records.

我们可以查看HDFS由参数--target-dir指定的路径查看导入的数据：

1. hadoop fs -text /user/xiaosi/data/order_info/* | less
2. 358574046793404,2016-04-05,flight
3. 358574046794733,2016-08-03,hotel
4. 358574050631177,2016-05-08,vacation
5. 358574050634213,2015-04-28,train
6. 358574050634692,2016-04-05,tuan
7. 358574050650524,2015-07-26,hotel
8. 358574050654773,2015-01-23,flight
9. 358574050668658,2015-01-23,hotel
10. 358574050730771,2016-11-06,train
11. 358574050731241,2016-05-08,car
12. 358574050743865,2015-01-23,vacation
13. 358574050767666,2015-04-28,train
14. 358574050767971,2015-07-26,flight
15. 358574050808288,2016-05-08,hotel
16. 358574050816828,2015-01-23,hotel
17. 358574050818220,2015-04-28,car
18. 358574050821877,2013-08-03,flight

再看一个例子：

1. sqoop import --connect jdbc:mysql://localhost:3306/test --table order_info --columns "business,id,order_time"  -m 1 --username root -P

HDFS上会在/user/xiaosi/目录下新增一个目录order_info，与关系性数据库的表名一致，内容如下：

1. flight,358574046793404,2016-04-05
2. hotel,358574046794733,2016-08-03
3. vacation,358574050631177,2016-05-08
4. train,358574050634213,2015-04-28
5. tuan,358574050634692,2016-04-05

6. import-all-tables

将数据库里的所有表导入HDFS中，每个表在HDFS中对应一个独立的目录。该命令选项的参数如下图所示：

7. list-databases

该命令选项可以列出关系性数据库的所有数据库名，命令如下：

1. sqoop list-databases --connect jdbc:mysql://localhost:3306 --username root -P

运行结果信息如下：

1. 16/11/14 14:30:11 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. Enter password:
3. 16/11/14 14:30:14 INFO manager.MySQLManager: Preparing to use a MySQL streaming resultset.
4. information_schema
5. hive_db
6. mysql
7. performance_schema
8. phpmyadmin
9. test

8. list-tables

该命令选项可以列出关系性数据库的某一个数据库的所有表名，命令如下：

1. sqoop list-tables --connect jdbc:mysql://localhost:3306/test --username root -P

运行结果信息如下：

1. 16/11/14 14:32:08 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. Enter password:
3. 16/11/14 14:32:10 INFO manager.MySQLManager: Preparing to use a MySQL streaming resultset.
4. PageView
5. book
6. bookID
7. cc
8. city_click
9. country
10. country2
11. cup
12. employee
13. flightOrder
14. hotel_book_info
15. hotel_info
16. order_info
17. stu
18. stu2
19. stu3
20. stuInfo
21. student

9. merge

该命令选项的作用是将HDFS上的两份数据进行合并，在合并的同时进行数据去重。该命令选项的参数如下图所示：

例如，在HDFS的路径/user/xiaosi/old下由一份导入数据，如下：

1. id name
2. 1 a
3. 2 b
4. 3 c

在HDFS的路径/user/xiaosi/new下也有一份数据，但是在导入时间在第一份之后，如下：

1. id name
2. 1 a2
3. 2 b
4. 3 c

那么合并的结果为：

1. id name
2. 1 a2
3. 2 b
4. 3 c

运行如下命令：

1. sqoop merge -new-data /user/xiaosi/new/part-m-00000 -onto /user/xiaosi/old/part-m-00000 -target-dir /user/xiaosi/final -jar-file /home/xiaosi/test/testmerge.jar -class-name testmerge -merge-key id

备注：

在一份数据集中，多行不应具有相同的主键，否则会发生数据丢失。

10. metastore

记录Sqoop作业的元数据信息，如果不启动Metastore实例，则默认的元数据存储目录为~/.sqoop。如果要更改存储目录，可以在配置文件sqoop-site.xml中进行更改。

启动Metastore实例：

1. sqoop metastore

运行结果信息如下：

1. 16/11/14 14:44:40 INFO sqoop.Sqoop: Running Sqoop version: 1.4.6
2. 16/11/14 14:44:40 WARN hsqldb.HsqldbMetaStore: The location for metastore data has not been explicitly set. Placing shared metastore files in /home/xiaosi/.sqoop/shared-metastore.db
3. [Server@52308be6]: [Thread[main,5,main]]: checkRunning(false) entered
4. [Server@52308be6]: [Thread[main,5,main]]: checkRunning(false) exited
5. [Server@52308be6]: [Thread[main,5,main]]: setDatabasePath(0,file:/home/xiaosi/.sqoop/shared-metastore.db)
6. [Server@52308be6]: [Thread[main,5,main]]: checkRunning(false) entered
7. [Server@52308be6]: [Thread[main,5,main]]: checkRunning(false) exited
8. [Server@52308be6]: [Thread[main,5,main]]: setDatabaseName(0,sqoop)
9. [Server@52308be6]: [Thread[main,5,main]]: putPropertiesFromString(): [hsqldb.write_delay=false]
10. [Server@52308be6]: [Thread[main,5,main]]: checkRunning(false) entered
11. [Server@52308be6]: [Thread[main,5,main]]: checkRunning(false) exited
12. [Server@52308be6]: Initiating startup sequence...
13. [Server@52308be6]: Server socket opened successfully in 3 ms.
14. [Server@52308be6]: Database [index=0, id=0, db=file:/home/xiaosi/.sqoop/shared-metastore.db, alias=sqoop] opened sucessfully in 153 ms.
15. [Server@52308be6]: Startup sequence completed in 157 ms.
16. [Server@52308be6]: 2016-11-14 14:44:40.414 HSQLDB server 1.8.0 is online
17. [Server@52308be6]: To close normally, connect and execute SHUTDOWN SQL
18. [Server@52308be6]: From command line, use [Ctrl]+[C] to abort abruptly
19. 16/11/14 14:44:40 INFO hsqldb.HsqldbMetaStore: Server started on port 16000 with protocol HSQL

11. job

该命令选项可以生产一个Sqoop的作业，但是不会立即执行，需要手动执行，该命令选项目的在于尽可能的服用Sqoop命令。该命令选项的参数如下图所示：

举例：

上面代码实现一个job，显示关系性数据库test数据库中所有的表。

上面代码执行我们已经定义好的Job，输出结果信息如下：

备注：

-- 和 list-tables（Job 所要执行的Sqoop命令） 不能挨着。



来自于《Hadoop海量数据处理  技术详解与项目实战》