从本质理解Oracle数据泵（expdp/impdp）原理

原创范伟林 2022-02-16

2647

一、前言

相信数据泵我们会经常用到，但是没有理解其原理和普通exp/imp的区别，往往遇到一些错误就很难进行排查；

其实数据泵的本质就是后台存储过程作业，由dw进程进行工作，此进程可以并发多个运行，相关作业状态可以由dba_datapump_jobs视图查看。在终端关闭或者ctrl c啥的依然不会影响作业的运行，若要进行作业管理需要attach；

利用数据泵可以做到逻辑的备份，数据迁移，传输表空间等，DBA比较常用的工具；

下面我们做个简单案例，环境为12c版本。

二、数据导出(expdp)

数据的导出为数据库的存储过程作业，所以需要用到目录对象，定义导出dump文件、日志文件等的路径信息；
导出任务发起，作业会数据库自动创建，作业会自动创建master表记录作业信息，作业状态也会记录在dba_datapump_jobs视图
数据导出源端为数据库，目标端为dump文件，也就是灵活将数据库的相关对象写入到dump物理文件，理解链接关系，跟目标库是无关系的
当然需要用到exp_full_database角色权限，目录对象的执行读写权限

1、查看涉及的表空间信息

表：

set line 300 pages 100
col tablespace_name for a40
col owner for a30
select distinct tablespace_name,owner from dba_tables
where owner in (select username from dba_users where account_status='OPEN' and username not in ('SYS','SYSTEM','DBSNMP'))
order by 2;

索引：

set line 300 pages 100
col tablespace_name for a40
col owner for a30
select distinct tablespace_name,owner from dba_indexes
where owner in (select username from dba_users where account_status='OPEN' and username not in ('SYS','SYSTEM','DBSNMP'))
order by 2;

Note：规划需要导出的schema，找出其数据涉及的表空间，方便后续的数据导入规划

2、查看数据库逻辑大小

select sum(bytes)/1024/1024/1024 from dba_segments;

Note：也可以事先查出比较大的表，例如千万级别以上的，然后在导的过程中看进度也心里有数，也可以单独拎出来导

3、创建目录对象

准备目录
mkdir /backup/dumpfile
chown oracle.oinstall /backup/dumpfile

创建
create directory mydir as '/backup/dumpfile';

Note：这里默认是用sys进行创建，普通用户需要进行授权

4、导出数据

创建参数文件

vi /home/oracle/wl_full.par

USERID=system/*******
DIRECTORY=dumpdir
DUMPFILE=wl20220216_%U.dmp
LOGFILE=wl20220216_exp.log
JOB_NAME=wl_export_full
LOGTIME=all
SCHEMA=wl
EXCLUDE=statistics
COMPRESSION=all
PARALLEL=8
CLUSTER=no

Note：也可以直接写导出参数，看个人习惯；参数比较多，可以得知数据泵很灵活功能强大，可以-help查看所有参数，这里列举几个参数；

USERID 为用户凭证，普通用户操作需要授权
DIRECTORY 为目录对象
DUMPFILE 为导出文件名称
LOGFILE 为导出日志名称
JOB_NAME 为作业名称，自定义作业名称方便管理
LOGTIME all为记录每个环节的时间
SCHEMA 为模式名称，通俗讲也就是导出哪个业务用户的数据
EXCLUDE 为排除不导出的内容，这里列举为统计信息，当然可以是其他的对象信息
COMPRESSION 为导出数据的压缩级别，all为数据和元数据全部压缩，最高压缩类型，当然可以结合压缩算法COMPRESSION_ALGORITHM达到更高级别的压缩
PARALLEL 为并行度，也就是定义dw进程的个数
CLUSTER 为RAC特有参数，定义作业是否在每个节点运行

执行任务

expdp PARFILE=/home/oracle/wl_full.par

5、查看作业日志也状态

观察导出日志

也就是LOGFILE参数定义的日志文件

tail -1000f wl20220216_exp.log

Export: Release 12.2.0.1.0 - Production on Thu Oct 15 11:54:07 2020

Copyright (c) 1982, 2017, Oracle and/or its affiliates.  All rights reserved.
;;; 
Connected to: Oracle Database 12c Enterprise Edition Release 12.2.0.1.0 - 64bit Production
15-OCT-20 11:54:19.635: ;;; **************************************************************************
15-OCT-20 11:54:19.640: ;;; Parfile values:
15-OCT-20 11:54:19.645: ;;;  parfile:  parallel=8
15-OCT-20 11:54:19.650: ;;;  parfile:  cluster=N
15-OCT-20 11:54:19.655: ;;;  parfile:  schemas=wl
15-OCT-20 11:54:19.661: ;;;  parfile:  logtime=ALL
15-OCT-20 11:54:19.666: ;;;  parfile:  logfile=wl20220216_exp.log
15-OCT-20 11:54:19.671: ;;;  parfile:  dumpfile=wl20220216_%U.dmp
15-OCT-20 11:54:19.754: ;;;  parfile:  directory=mydir
15-OCT-20 11:54:19.760: ;;; **************************************************************************
15-OCT-20 11:54:20.427: FLASHBACK automatically enabled to preserve database integrity.
15-OCT-20 11:54:21.601: Starting "SYS"."SYS_EXPORT_SCHEMA_01":  "/******** AS SYSDBA" parfile=/home/oracle/wl_full.par
15-OCT-20 11:54:24.230: Processing object type SCHEMA_EXPORT/SYSTEM_GRANT
15-OCT-20 11:54:24.273: Processing object type SCHEMA_EXPORT/ROLE_GRANT
15-OCT-20 11:54:24.274: Processing object type SCHEMA_EXPORT/DEFAULT_ROLE
15-OCT-20 11:54:24.275: Processing object type SCHEMA_EXPORT/USER
15-OCT-20 11:54:25.968: Processing object type SCHEMA_EXPORT/PRE_SCHEMA/PROCACT_SCHEMA
15-OCT-20 11:54:26.146: Processing object type SCHEMA_EXPORT/DB_LINK
15-OCT-20 11:54:26.773: Processing object type SCHEMA_EXPORT/SYNONYM/SYNONYM
15-OCT-20 11:54:26.900: Processing object type SCHEMA_EXPORT/TABLE/INDEX/STATISTICS/FUNCTIONAL_INDEX/INDEX_STATISTICS
15-OCT-20 11:54:26.904: Processing object type SCHEMA_EXPORT/PACKAGE/PACKAGE_BODY
15-OCT-20 11:54:27.075: Processing object type SCHEMA_EXPORT/TABLE/INDEX/STATISTICS/INDEX_STATISTICS
15-OCT-20 11:54:27.222: Processing object type SCHEMA_EXPORT/TABLE/STATISTICS/TABLE_STATISTICS
15-OCT-20 11:54:27.621: Processing object type SCHEMA_EXPORT/SEQUENCE/SEQUENCE
15-OCT-20 11:54:28.311: Processing object type SCHEMA_EXPORT/TABLE/TABLE_DATA
15-OCT-20 11:54:29.010: Processing object type SCHEMA_EXPORT/TABLE/PROCACT_INSTANCE
15-OCT-20 11:54:29.999: Processing object type SCHEMA_EXPORT/TABLE/GRANT/OWNER_GRANT/OBJECT_GRANT
15-OCT-20 11:54:31.714: Processing object type SCHEMA_EXPORT/PACKAGE/PACKAGE_SPEC
15-OCT-20 11:54:31.830: Processing object type SCHEMA_EXPORT/FUNCTION/FUNCTION
15-OCT-20 11:54:31.969: Processing object type SCHEMA_EXPORT/PROCEDURE/PROCEDURE
15-OCT-20 11:54:32.145: Processing object type SCHEMA_EXPORT/PACKAGE/COMPILE_PACKAGE/PACKAGE_SPEC/ALTER_PACKAGE_SPEC
15-OCT-20 11:54:32.157: Processing object type SCHEMA_EXPORT/FUNCTION/ALTER_FUNCTION
15-OCT-20 11:54:32.163: Processing object type SCHEMA_EXPORT/PROCEDURE/ALTER_PROCEDURE
15-OCT-20 11:54:32.582: Processing object type SCHEMA_EXPORT/TABLE/COMMENT
15-OCT-20 11:54:32.630: Processing object type SCHEMA_EXPORT/TABLE/TABLE
15-OCT-20 11:54:32.972: Processing object type SCHEMA_EXPORT/TABLE/INDEX/FUNCTIONAL_INDEX/INDEX
15-OCT-20 11:54:34.086: Processing object type SCHEMA_EXPORT/TABLE/CONSTRAINT/REF_CONSTRAINT
15-OCT-20 11:54:34.612: Processing object type SCHEMA_EXPORT/TABLE/CONSTRAINT/CONSTRAINT
15-OCT-20 11:54:34.678: Processing object type SCHEMA_EXPORT/VIEW/VIEW
15-OCT-20 11:54:34.783: Processing object type SCHEMA_EXPORT/TABLE/INDEX/INDEX
15-OCT-20 11:54:47.347: Processing object type SCHEMA_EXPORT/STATISTICS/MARKER
15-OCT-20 11:58:02.392: . . exported "WL"."T_TEST_RECORD"                        18.99 GB 66499480 rows
15-OCT-20 11:59:22.653: . . exported "WL"."T_TEST"                            30.47 GB 70834724 rows

观察作业状态

真正管理作业需要attach作业进去操作

查看作业

set line 300 pages 100
col owner_name for a20
col job_name for a30
col state for a20
select owner_name,job_name,state from dba_datapump_jobs;

OWNER_NAME	     JOB_NAME			    STATE
-------------------- ------------------------------ --------------------
SYSTEM		     SJJ_EXPORT_FULL		    EXECUTING

登录作业

expdp system attach=sjj_export_full

查看作业状态

status

Export> status 

Job: SJJ_EXPORT_FULL
  Operation: EXPORT                         
  Mode: FULL                           
  State: STOP PENDING                   
  Bytes Processed: 32,384,054,664
  Percent Done: 99
  Current Parallelism: 8
  Job Error Count: 0
  Job heartbeat: 6
  Dump File: /backup/dumpfile/full20220216_%u.dmp
  Dump File: /backup/dumpfile/full20220216_01.dmp
    bytes written: 15,032,143,872
  Dump File: /backup/dumpfile/full20220216_02.dmp
    bytes written: 3,542,888,448
  Dump File: /backup/dumpfile/full20220216_03.dmp
    bytes written: 3,009,998,848
  Dump File: /backup/dumpfile/full20220216_04.dmp
    bytes written: 2,373,156,864
  Dump File: /backup/dumpfile/full20220216_05.dmp
    bytes written: 3,188,301,824
  Dump File: /backup/dumpfile/full20220216_06.dmp
    bytes written: 948,051,968
  Dump File: /backup/dumpfile/full20220216_07.dmp
    bytes written: 37,437,628,416
  Dump File: /backup/dumpfile/full20220216_08.dmp
    bytes written: 2,978,820,096
  
Worker 1 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Object start time: Wednesday, 16 February, 2022 20:35:08
  Object status at: Wednesday, 16 February, 2022 22:03:31
  Process Name: DW00
  State: WORK WAITING                   
  
Worker 2 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Access method: direct_path
  Object start time: Wednesday, 16 February, 2022 20:33:35
  Object status at: Wednesday, 16 February, 2022 20:40:42
  Process Name: DW01
  State: WORK WAITING                   
  
Worker 3 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Access method: direct_path
  Object start time: Wednesday, 16 February, 2022 20:33:35
  Object status at: Wednesday, 16 February, 2022 21:16:26
  Process Name: DW02
  State: WORK WAITING                   
  
Worker 4 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Access method: direct_path
  Object start time: Wednesday, 16 February, 2022 20:33:38
  Object status at: Wednesday, 16 February, 2022 20:40:37
  Process Name: DW03
  State: WORK WAITING                   
  
Worker 5 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Object start time: Wednesday, 16 February, 2022 20:36:11
  Object status at: Wednesday, 16 February, 2022 20:38:13
  Process Name: DW04
  State: EXECUTING                      
  Object Schema: WL
  Object Name: T_TEST
  Object Type: DATABASE_EXPORT/SCHEMA/TABLE/TABLE_DATA
  Completed Objects: 4
  Total Objects: 939
  Completed Rows: 430,816
  Worker Parallelism: 1
  
Worker 6 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Access method: external_table
  Object start time: Wednesday, 16 February, 2022 20:33:35
  Object status at: Wednesday, 16 February, 2022 20:41:10
  Process Name: DW05
  State: WORK WAITING                   
  
Worker 7 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Access method: direct_path
  Object start time: Wednesday, 16 February, 2022 20:33:35
  Object status at: Wednesday, 16 February, 2022 20:40:31
  Process Name: DW06
  State: WORK WAITING                   
  
Worker 8 Status:
  Instance ID: 1
  Instance name: wldb1
  Host name: wldb1
  Access method: direct_path
  Object start time: Wednesday, 16 February, 2022 20:34:00
  Object status at: Wednesday, 16 February, 2022 20:40:31
  Process Name: DW07
  State: WORK WAITING                   

Export>

Note：若导出日志长时间没反应，可以时不时查看作业status，观察相关对象是否在变化，判断作业是否正常。

其他操作：
STOP_JOB，停止作业，可以继续启动，dba_datapump_jobs信息依然存在
START_JOB，继续启动停止的作业
KILL_JOB，强制终止作业，dba_datapump_jobs信息会清除

三、数据导入

导入跟导出原理差不多

数据的导入也是数据库的存储过程作业，所以需要用到目录对象，定义dump文件、日志文件等的路径信息；
导入任务发起，作业会数据库自动创建，作业会自动创建master表记录作业信息，作业状态也会记录在dba_datapump_jobs视图
数据导入源端为dump文件，目标端为数据库，也就是灵活将dump文件的相关对象写入到目标数据库，理解链接关系，跟源库是无关系的
当然需要用到imp_full_database角色权限，目录对象的执行读写权限

1、创建用户和表空间

create tablespace TEST datafile size 31G autoextend on;

create user TEST identified by "TEST" default tablespace TEST quota unlimited on TEST;

Note：关键是定义好表空间，用户可以不用创建
若不同表空间需要利用REMAP_TABLESPACE重新映射表空间
若需要导入不同的用户，可以利用REMAP_SCHEMA重新映射用户
当然表结构和数据也可以重新映射

2、创建目录对象

准备目录
mkdir /backup/dumpfile
chown oracle.oinstall /backup/dumpfile

创建
create directory mydir as '/backup/dumpfile';

3、导入数据