Oracle 一次生产分库,升级,迁移

分库目的:

1) 迁出BUSINESS/BUSINESS_APP两业务系统用户数据, 减轻中央库的业务压力和负载, 调整数据库架构, 以应对后续新业务系统上线造成对中央库的冲击;

2) 新库升级Oracle10.2.0.1到10.2.0.4, 数据库本身修复了很多bug, 增强了数据库的稳定性.

3) 调整定时任务, 把原先的定时任务由crontab/job方式改为oracle scheduler.

大概操作步骤如下:

--1) 提前安装Oracle10.2.0.4, 并部署Streams复制(schema复制)

BEGIN
DBMS_STREAMS_ADM.MAINTAIN_SCHEMAS (
      schema_names                   => 'BUSINESS,BUSINESS_APP',
      source_directory_object        => 'DIR_SOURCE',
      destination_directory_object   => 'DIR_DEST',
      source_database                => 'SOUR.NET',
      destination_database           => 'DEST.LK',
      perform_actions                => TRUE,
      script_name                    => NULL,
      script_directory_object        => NULL,
      capture_name                   => 'cap_erating',
      capture_queue_table            => 'tab_cap_que_erating',
      capture_queue_name             => 'cap_que_erating',
      capture_queue_user             => NULL,
      propagation_name               => 'prop_erating',
      apply_name                     => 'app_erating',
      apply_queue_table              => 'tab_app_que_erating',
      apply_queue_name               => 'app_que_erating',
      apply_queue_user               => NULL,
      dump_file_name                 => 'SOUR.dmp',
      log_file                       => 'SOUR.log',
      bi_directional                 => FALSE,
      include_ddl                    => TRUE,
      instantiation                  => DBMS_STREAMS_ADM.instantiation_schema
   );
END;
/

--2) 检查目标库Streams复制是否正常

SELECT   'capture' process_type,
         capture_name process_name,
         status,
         error_message
FROM   DBA_CAPTURE
UNION ALL
SELECT   'propagation' process_type,
         propagation_name process_name,
         status,
         error_message
FROM   dba_propagation
UNION ALL
SELECT   'apply' process_type,
         apply_name process_name,
         status,
         error_message
FROM   dba_apply;

SELECT   apply_name,
           local_transaction_id,
           source_commit_scn,
           error_creation_time,
           error_message
    FROM   dba_apply_error
ORDER BY   source_commit_scn DESC;

--3) 停止全部业务系统, 要求只能中断15分钟

--4) 取消源库定时任务(包括crontab/job)

--5) 核查源库是否还有连接

SELECT   username, status, COUNT ( * )
    FROM   v$session
   WHERE   username IN
                 ('BUSINESS', 'BUSINESS_APP')
GROUP BY   username, status;

--6) 核查目标库是否还有Streams复制事务

SELECT   streams_name,
           streams_type,
           cumulative_message_count,
           first_message_time,
           XIDUSN,
           XIDSLT,
           XIDSQN,
           last_message_time,
           total_message_count
    FROM   v$streams_transaction
ORDER BY   3 DESC;

--7) 从源库获得重建序列语句, Streams复制这点特别注意, 因为Streams本身不会去同步序列值

set pages 0
SELECT   'DROP SEQUENCE ' || sequence_owner || '.' || sequence_name || ';'
            stmt
FROM   dba_sequences
WHERE   sequence_owner IN
               ('BUSINESS', 'BUSINESS_APP')
UNION ALL
SELECT      'CREATE SEQUENCE '
         || sequence_owner
         || '.'
         || sequence_name
         || ' start with '
         || last_number
         || ' MAXVALUE '
         || max_value
         || ' MINVALUE '
         || min_value
     || DECODE (cycle_flag, 'N', ' NOCYCLE ', ' CYCLE ')
     || DECODE (cache_size, 0, ' NOCACHE ', ' CACHE '||cache_size)

|| DECODE (ORDER_FLAG, 'N', ' NOORDER ', ' ORDER ')

|| ';' stmt

FROM   dba_sequences
WHERE   sequence_owner IN
               ('BUSINESS', 'BUSINESS_APP');

--8) 目标库重建序列

--9) 源库和目标库刷新同义词, 把同义词指向新的dblink, 同义词脚本提前准备好
--源库
@E:用户迁移sour_synonyms.sql
--目标库
@E:用户迁移dest_synonyms.sql

--10) 源库和目标库重新编译失效对象
@?/rdbms/admin/utlrp.sql
exec uts.get_invalid;

--11) 各业务系统更改数据库连接指向, 并启动各业务系统

--12) 测试业务系统启动情况
SELECT   username, status, COUNT ( * )
    FROM   v$session
   WHERE   username IN
                 ('BUSINESS', 'BUSINESS_APP')
GROUP BY   username, status;

--12) 增加定时任务, 为方便管理统一改为scheduler

--13) 删除流配置
exec dbms_streams_adm.remove_streams_configuration;

--14) 其它收尾工作, 如各开发人员查询用户的授权等等.

--End--

Oracle 一次生产分库,升级,迁移

今天完成了一个负载较高的中央数据库的分库操作, 并实现了oracle的滚动升级(10.2.0.1->10.2.0.4), 业务中断仅15分钟. 平台: RHEL AS 4 + Oracle 10.2.0.1