OceanBase 业务数据库实践（二）── DB2 迁移

数据库技术闲谈 2021-08-15

520

IBM DB2 是美国IBM公司开发的一套关系型数据库管理系统，它主要的运行环境为UNIX（包括IBM自家的AIX）、Linux、IBM i（旧称OS/400）、z/OS，以及Windows服务器版本。其中 LINUX、UNIX 和Windows 平台又简称 LUW 开放平台。客户包含省农信、农商行以及一些中字头金融机构。

目前业界应该没有哪家国产数据库号称兼容 DB2，所以 “去 DB2” ，业务的改造是不可避免的。业界数据库能力最接近 DB2 的是 ORACLE 数据库，而 OceanBase 又同时兼容 ORACLE 和 MySQL。所以 DB2 迁移到 OceanBase ，最佳选择是选择 ORACLE 租户类型，其次是 MySQL 租户。

本文主要是总结 DB2 迁移到 OceanBase ORACLE 租户的经验。

表结构转换

首先要评估 DB2 的表结构在 OB(ORACLE) 上的兼容性。DB2 跟 OB(ORACLE) 的数据类型差异还是比较大，下面仅就常用的类型的典型差异举例说明。

自增列类型

DB2 的 SQL 语法，在 OB(ORACLE) 里都可以找到对应的实现，但不完全一样。

DB2 的自增列（IDENTITY
）语法 OB 3.2 版本开始支持。在那之前，需要应用使用 SEQUENCE 给列填充值。

DB2 示例：

CREATE TABLE t4 ( ID BIGINT NOT NULL GENERATED BY DEFAULT AS IDENTITY (MAXVALUE 9223372036854775807), C1 VARCHAR (50) NOT NULL DEFAULT '' )  ;

ORACLE 示例：

CREATE TABLE t4 (ID NUMBER GENERATED ALWAYS AS IDENTITY MINVALUE 1 MAXVALUE 9223372036854775807 NOT NULL, C1 VARCHAR2(50) );

二者写法 NOT NULL
的位置不同。

DB2 的 DATE / TIME 跟 OB(ORACLE) 的 DATE 差异问题

默认情况下（指 DB2 不开启 ORACLE 兼容模式），DB2 的 DATE 只包含日期，TIME 只包含时间。而 OB(ORACLE) 的 DATE 包含日期和时间。

DB2:

db2 => DROP TABLE t1
DB20000I  The SQL command completed successfully.
db2 => CREATE TABLE t1(id bigint NOT NULL PRIMARY KEY ,c1 date , c2 time, c3 timestamp)
DB20000I  The SQL command completed successfully.
db2 => INSERT INTO t1 values(1, sysdate,sysdate,current_timestamp )
DB20000I  The SQL command completed successfully.
db2 => SELECT * FROM t1

ID                   C1         C2       C3
-------------------- ---------- -------- --------------------------
                   1 08/14/2021 05:29:55 2021-08-14-05.29.55.343157

  1 record(s) selected.

db2 =>

OB(ORACLE):

obclient> DROP TABLE t1;
Query OK, 0 rows affected (0.02 sec)

obclient> CREATE TABLE t1(id number NOT NULL PRIMARY KEY ,c1 date , c3 timestamp);
Query OK, 0 rows affected (0.04 sec)

obclient> INSERT INTO t1 values(1, sysdate,current_timestamp);
Query OK, 1 row affected (0.01 sec)

obclient> SELECT * FROM t1;
+----+---------------------+----------------------------+
| ID | C1                  | C3                         |
+----+---------------------+----------------------------+
|  1 | 2021-08-14 13:28:15 | 2021-08-14 13:28:15.177911 |
+----+---------------------+----------------------------+
1 row in set (0.00 sec)

字符集问题

如果 DB2 字符集已经是 UTF-8 ，那可以直接转换为 OB(ORACLE) 的 UTF8MB4 。UTF8MB4 是 UTF8 的超集，多一些表情字符。

如果 DB2 字符集不是 UTF-8 ，也建议转换为 OB(ORACLE) 的 UTF8MB4，相关字符串字段长度需要在 OB(ORACLE) 里放大。如果是单向同步，这个没问题。如果还要考虑 OB(ORACLE) 数据回流到 DB2，可能会出现实际字符串数据超长问题。

查看 DB2 数据库字符集方法：

[db2inst1@9a9fd42bf525 ~]$ db2 get db cfg for testdb |grep -i code
 Database code page                                      = 1208
 Database code set                                       = utf-8
 Database country/region code                            = 1
 [db2inst1@a32408ac626d ~]$ db2 get db cfg for testdb|grep -i national
 National character string mapping       (NCHAR_MAPPING) = CHAR_CU32

查看 OB(ORACLE) 租户字符集方法

MySQL [(none)]> show variables like '%character%';
+--------------------------+-----------+
| VARIABLE_NAME            | VALUE     |
+--------------------------+-----------+
| character_set_client     | utf8mb4   |
| character_set_connection | utf8mb4   |
| character_set_filesystem | binary    |
| character_set_results    | utf8mb4   |
| character_set_system     | utf8mb4   |
| nls_characterset         | AL32UTF8  |
| nls_nchar_characterset   | AL16UTF16 |
| nls_numeric_characters   | .,        |
+--------------------------+-----------+
8 rows in set (0.003 sec)

OB(ORACLE) 的参数 nls_characterset
定义了数据库的字符集，通常是 AL32UTF8
，每个中文字符字长为 3 个字节。同时 nls_nchar_characterset
定义了 Unicode字符串使用的字符集，每个中文字符字长 2 个字节。

字符串长度问题

DB2 和 ORACLE 都有参数定义一个字符存储时对应多少字节。这个会影响一个列里最多能存储几个中文字符。

DB2 查看字长单位方法

[db2inst1@8d6dd8adcacf ~]$ db2 get db cfg for testdb |grep -i string
 Default string units                     (STRING_UNITS) = SYSTEM
 National character string mapping       (NCHAR_MAPPING) = CHAR_CU32

当字符集是 Unicode
时，

如果 STRING_UNITS
值为 SYSTEM
，则 CHAR, VARCHAR, and CLOB 类型的长度在不指定 CODEUNITS32
时，会默认为是 OCTETS
。
如果 STRING_UNITS
值为 CODEUNITS32
时，则默认长度是 CODEUNITS32
。

字符串长度单位说明：

OCTETS
: 使用字节作为字符的单位。
CODEUNITS16
：使用 UTF-16
编码作为字符的单位，长度为 2 个字节。
CODEUNITS32
：使用 UTF-32
编码作为字符的单位，长度为 4 个字节。

NCHAR_MAPPING
取值为 CHAR_CU32
时，NCHAR
、NVARCHAR
和 NCLOB
列字符默认单位是 CODEUNITS32
。

OB(ORACLE) 查看字长单位方法。

obclient> show global variables like '%nls_length_semantics%';
+----------------------+-------+
| VARIABLE_NAME        | VALUE |
+----------------------+-------+
| nls_length_semantics | BYTE  |
+----------------------+-------+
1 row in set (0.00 sec)

参数 nls_length_semantics
是在租户下设置，只影响当前租户。取值有两种：

BYTE
: 使用字节作为字符的单位。
CHAR
: 使用字符作为单位。每个字符多少个字节跟字符集设置有关。字符集是 UTF-8
的时候，每个中文字符 3 个字节。

下面的示例方便具体理解 DB2 跟 OB(ORACLE) 的字符长度差异。

DB2 字符长度示例

db2 => DROP TABLE t2;
DB20000I  The SQL command completed successfully.
db2 => CREATE TABLE t2(id bigint NOT NULL PRIMARY KEY, c1 char(50), c2 varchar(50), c3 nchar(50), c4 nvarchar(50));
DB20000I  The SQL command completed successfully.
db2 => INSERT INTO t2 values(1,'中','中','中','中');
DB20000I  The SQL command completed successfully.
db2 => INSERT INTO t2 values(2,'中国','中国','中国','中国');
DB20000I  The SQL command completed successfully.
db2 => SELECT id, c1, length(c1) c1_length,lengthb(c1) c1_lengthb,c2,length(c2) c2_length,lengthb(c2) c2_lengthb, c3, length(c3) c3_length, lengthb(c3) c3_lengthb,c4, length(c4) c4_length,lengthb(c4) c4_lengthb FROM t2;

ID                   C1                                                 C1_LENGTH   C1_LENGTHB  C2                                                 C2_LENGTH   C2_LENGTHB  C3                                                                                                   C3_LENGTH   C3_LENGTHB  C4                                                                                                   C4_LENGTH   C4_LENGTHB
-------------------- -------------------------------------------------- ----------- ----------- -------------------------------------------------- ----------- ----------- ---------------------------------------------------------------------------------------------------- ----------- ----------- ---------------------------------------------------------------------------------------------------- ----------- -----------
                   1 中                                                         50          50 中                                                          6           6 中                                                                                                           50          53 中                                                                                                            3           6
                   2 中国                                                      50          50 中国                                                      12          12 中国                                                                                                        50          56 中国                                                                                                         6          12

  2 record(s) selected.

ORACLE 字符长度示例

obclient> DROP TABLE t2;
Query OK, 0 rows affected (0.02 sec)

obclient> CREATE TABLE t2(id number NOT NULL PRIMARY KEY, c1 char(50), c2 varchar2(50), c3 nchar(50), c4 nvarchar2(50));
Query OK, 0 rows affected (0.05 sec)

obclient> INSERT INTO t2 values(1,'中','中','中','中');
Query OK, 1 row affected (0.00 sec)

obclient> INSERT INTO t2 values(2,'中国','中国','中国','中国');
Query OK, 1 row affected (0.00 sec)

obclient> SELECT id, c1, length(c1) c1_length,lengthb(c1) c1_lengthb,c2,length(c2) c2_length,lengthb(c2) c2_lengthb, c3, length(c3) c3_length, lengthb(c3) c3_lengthb,c4, length(c4) c4_length,lengthb(c4) c4_lengthb FROM t2;
+----+----------------------------------------------------+-----------+------------+--------+-----------+------------+--------------------------------------------------------+-----------+------------+--------+-----------+------------+
| ID | C1                                                 | C1_LENGTH | C1_LENGTHB | C2     | C2_LENGTH | C2_LENGTHB | C3                                                     | C3_LENGTH | C3_LENGTHB | C4     | C4_LENGTH | C4_LENGTHB |
+----+----------------------------------------------------+-----------+------------+--------+-----------+------------+--------------------------------------------------------+-----------+------------+--------+-----------+------------+
|  1 | 中                                                 |        48 |         50 | 中     |         1 |          3 | 中                                                     |        50 |        100 |  中     |         1 |          2 |
|  2 | 中国                                               |        46 |         50 | 中国   |         2 |          6 | 中国                                                   |        50 |        100 |  中国   |         2 |          4 |
+----+----------------------------------------------------+-----------+------------+--------+-----------+------------+--------------------------------------------------------+-----------+------------+--------+-----------+------------+
2 rows in set (0.01 sec)

非空与空值问题

在 DB2 里，空值跟 NULL
是两个值，并不相等。但是在 OB(ORACLE) 里，空值就是 NULL
。
针对是否是 NULL
只能用 IS NULL
或者 IS NOT NULL
判断，而不能用等值符号判断。在 DB2 里针对是否是空值，是用等值符号判断。
所以，DB2 的应用如果定义了很多 NOT NULL
列，但实际又允许写入空值，这个应用迁移到 OB(ORACLE) 是会报错的。解决办法只能是拿掉 OB(ORACLE) 列的 NOT NULL
属性。

DB2 示例

db2 => DROP TABLE t3;
DB20000I  The SQL command completed successfully.
db2 => CREATE TABLE t3(id bigint NOT NULL PRIMARY KEY, c1 varchar(50) NOT NULL , c2 varchar(50) NULL );
DB20000I  The SQL command completed successfully.
db2 => INSERT INTO t3 values(1, '', '');
DB20000I  The SQL command completed successfully.
db2 => INSERT INTO t3 values(2, ' ', '');
DB20000I  The SQL command completed successfully.
db2 => SELECT id, c1, hex(c1), c2, hex(c2) FROM t3 ;

ID                   C1                                                 3                                                                                                    C2                                                 5
-------------------- -------------------------------------------------- ---------------------------------------------------------------------------------------------------- -------------------------------------------------- ----------------------------------------------------------------------------------------------------
                   1
                   2                                                    20

  2 record(s) selected.

db2 => SELECT id, c1, hex(c1), c2, hex(c2) FROM t3 WHERE c1 IS NULL ;

ID                   C1                                                 3                                                                                                    C2                                                 5
-------------------- -------------------------------------------------- ---------------------------------------------------------------------------------------------------- -------------------------------------------------- ----------------------------------------------------------------------------------------------------

  0 record(s) selected.

db2 => SELECT id, c1, hex(c1), c2, hex(c2) FROM t3 WHERE c1 = '';

ID                   C1                                                 3                                                                                                    C2                                                 5
-------------------- -------------------------------------------------- ---------------------------------------------------------------------------------------------------- -------------------------------------------------- ----------------------------------------------------------------------------------------------------
                   1
                   2                                                    20

  2 record(s) selected.

db2 =>

ORACLE 示例

obclient> DROP TABLE t3;
Query OK, 0 rows affected (0.02 sec)

obclient> CREATE TABLE t3(id number NOT NULL PRIMARY KEY, c1 varchar2(50) NOT NULL , c2 varchar2(50) NULL );
Query OK, 0 rows affected (0.05 sec)

obclient> INSERT INTO t3 values(1, '', '');
ORA-01400: cannot insert NULL into '(C1)'
obclient> INSERT INTO t3 values(2, ' ', '');
Query OK, 1 row affected (0.00 sec)

obclient> SELECT id, c1, dump(c1), c2, dump(c2,16) FROM t3 ;
+----+----+------------------+------+-------------+
| ID | C1 | DUMP(C1)         | C2   | DUMP(C2,16) |
+----+----+------------------+------+-------------+
|  2 |    | Typ=22 Len=1: 32 | NULL | NULL        |
+----+----+------------------+------+-------------+
1 row in set (0.01 sec)

obclient> SELECT id, c1, dump(c1), c2, dump(c2,16) FROM t3 WHERE c1 IS NULL ;
Empty set (0.00 sec)

obclient> SELECT id, c1, dump(c1), c2, dump(c2,16) FROM t3 WHERE c1 = '';
Empty set (0.00 sec)

obclient> SELECT id, c1, dump(c1), c2, dump(c2,16) FROM t3 WHERE c1 = ' ';
+----+----+------------------+------+-------------+
| ID | C1 | DUMP(C1)         | C2   | DUMP(C2,16) |
+----+----+------------------+------+-------------+
|  2 |    | Typ=22 Len=1: 32 | NULL | NULL        |
+----+----+------------------+------+-------------+
1 row in set (0.01 sec)

obclient>

范围分区表分区边界问题

DB2 的范围分区表语法跟 OB(ORACLE) 语法在指定分区的范围时语法完全不一样。除此之外，边界值也有不同逻辑。

DB2 的 RANGE 分区在定义分区边界时，可以通过在起始值和结束值后面带上选项 INCLUSIVE
或 EXCLUSIVE
，以表示是否包含边界值。不指定的时候就是默认为 INCLUSIVE
（包含边界值）。OB(ORACLE) 没有这个语法，默认分区范围就是不包含边界值，且只能指定上限，不能指定下限。

所以，DB2 的分区表分区范围不要使用 INCLUSIVE
选项，要明确指定 EXCLUSIVE
选项；否则，转换为 OB(ORACLE) 的时候要调整分区范围的起始值和结束值。

DB2 示例

[db2inst1@8d6dd8adcacf ~]$ db2 "DROP TABLE t6;"
DB20000I  The SQL command completed successfully.
[db2inst1@a32408ac626d ~]$ db2 "CREATE TABLE t6(
> id bigint NOT NULL,
> c1 varchar(50),
> c2 timestamp NOT NULL ,
> PRIMARY KEY (id)
> ) PARTITION BY RANGE(c2)
> ( PARTITION P1 STARTING '1/1/2020' ENDING '1/31/2020' INCLUSIVE ,
> PARTITION P2 STARTING '2/1/2020' ENDING '2/28/2020' INCLUSIVE ,
> PARTITION P3 STARTING '3/1/2020' ENDING '3/31/2020' INCLUSIVE ,
> PARTITION P4 STARTING '4/1/2020' ENDING '4/30/2020' INCLUSIVE ,
> PARTITION P5 STARTING '5/1/2020' ENDING '5/31/2020' INCLUSIVE
> );"
DB20000I  The SQL command completed successfully.
[db2inst1@a32408ac626d ~]$ db2 "INSERT INTO t6(id,c1,c2) values(1,'A','1/1/2020');"
DB20000I  The SQL command completed successfully.
[db2inst1@a32408ac626d ~]$ db2 "INSERT INTO t6(id,c1,c2) values(2,'B','5/31/2020');"
DB20000I  The SQL command completed successfully.
[db2inst1@a32408ac626d ~]$ db2 "INSERT INTO t6(id,c1,c2) values(3,'C','12/31/2019');"
DB21034E  The command was processed as an SQL statement because it was not a
valid Command Line Processor command.  During SQL processing it returned:
SQL0327N  The row cannot be inserted into table "TPCC.T6" because it is
outside the bounds of the defined data partition ranges.  SQLSTATE=22525
[db2inst1@a32408ac626d ~]$
[db2inst1@a32408ac626d ~]$ db2 "SELECT * FROM t6;"

ID                   C1                                                 C2
-------------------- -------------------------------------------------- --------------------------
                   1 A                                                  2020-01-01-00.00.00.000000
                   2 B                                                  2020-05-31-00.00.00.000000

  2 record(s) selected.

[db2inst1@a32408ac626d ~]$

ORACLE 用法

obclient> CREATE TABLE t6(
    -> id number NOT NULL,
    -> c1 varchar(50),
    -> c2 timestamp NOT NULL ,
    -> PRIMARY KEY (id,c2)
    -> ) PARTITION BY RANGE(c2)
    -> ( PARTITION P1 VALUES LESS THAN ( to_date('2020/02/01','YYYY/MM/DD')),
    -> PARTITION P2 VALUES LESS THAN ( to_date( '2020/03/01','YYYY/MM/DD')),
    -> PARTITION P3 VALUES LESS THAN ( to_date( '2020/04/01','YYYY/MM/DD')),
    -> PARTITION P4 VALUES LESS THAN ( to_date( '2020/05/01','YYYY/MM/DD')),
    -> PARTITION P5 VALUES LESS THAN ( to_date( '2020/06/01','YYYY/MM/DD'))
    -> );
Query OK, 0 rows affected (0.09 sec)

obclient> INSERT INTO t6(id,c1,c2) values(1,'A',to_date('2020/01/01','YYYY/MM/DD'));
Query OK, 1 row affected (0.01 sec)

obclient> INSERT INTO t6(id,c1,c2) values(2,'B',to_date('2020/03/01','YYYY/MM/DD'));
Query OK, 1 row affected (0.00 sec)

obclient> INSERT INTO t6(id,c1,c2) values(3,'C',to_date('2020/06/01','YYYY/MM/DD'));
ORA-14400: inserted partition key does not map to any partition
obclient>

SQL 语法转换

除了数据类型问题外，有些常用 SQL 写法也许要微调。

`rownumber`
函数

DB2 里有 rownumber
和 row_number
函数，作用一样，但是 OB(ORACLE) 只有 row_number
函数。如果业务使用的是 rownumber
，那么要调整一下 SQL。

DB2 应用通常把 rownumber
和 FETCH FIRST N ROWS
结合使用来进行分页，在 OB(ORACLE) 3.1 版本之前，需要转换为 OB(ORACLE) 的分页方法。

DB2 示例

CREATE TABLE t5 AS (SELECT * FROM SYSIBM.TABLES ) WITH DATA ;

SELECT * 
FROM (
	SELECT row_number() OVER (ORDER BY ORDER OF t5) rn , * FROM t5
	ORDER BY table_schema, table_type 
	FETCH FIRST 100 ROWS ONLY 
	WITH ur 
) t 
WHERE rn >=90;

业务在写这类 SQL 的时候可能会不写最里面的 ORDER BY
语句，这样性能会快很多。但是结果集的顺序就是不确定。需要确认是否符合业务需求。

OB(ORACLE) 示例

CREATE TABLE t5 AS SELECT * FROM all_objects;
EXPLAIN EXTENDED_NOADDR 
SELECT * FROM (
	SELECT rownum rn, owner, object_name, object_type, created, last_ddl_time
	FROM (
		SELECT * 
		FROM t5
		ORDER BY owner,object_name
	) WHERE rownum < =100
) WHERE rn >=90
;

========================================
|ID|OPERATOR      |NAME |EST. ROWS|COST|
----------------------------------------
|0 |COUNT         |     |11       |1143|
|1 | SUBPLAN SCAN |VIEW1|11       |1143|
|2 |  LIMIT       |     |11       |1143|
|3 |   TOP-N SORT |     |100      |1141|
|4 |    TABLE SCAN|T5   |729      |891 |
========================================

Outputs & filters: 
-------------------------------------
  0 - output([rownum() + ?], [VIEW1.T5B.OWNER], [VIEW1.T5B.OBJECT_NAME], [VIEW1.T5B.SUBOBJECT_NAME], [VIEW1.T5B.OBJECT_ID], [VIEW1.T5B.DATA_OBJECT_ID], [VIEW1.T5B.OBJECT_TYPE], [VIEW1.T5B.CREATED], [VIEW1.T5B.LAST_DDL_TIME], [VIEW1.T5B.TIMESTAMP], [VIEW1.T5B.STATUS], [VIEW1.T5B.TEMPORARY], [VIEW1.T5B.GENERATED], [VIEW1.T5B.SECONDARY], [VIEW1.T5B.NAMESPACE], [VIEW1.T5B.EDITION_NAME]), filter(nil)
  1 - output([VIEW1.T5B.OWNER], [VIEW1.T5B.OBJECT_NAME], [VIEW1.T5B.SUBOBJECT_NAME], [VIEW1.T5B.OBJECT_ID], [VIEW1.T5B.DATA_OBJECT_ID], [VIEW1.T5B.OBJECT_TYPE], [VIEW1.T5B.CREATED], [VIEW1.T5B.LAST_DDL_TIME], [VIEW1.T5B.TIMESTAMP], [VIEW1.T5B.STATUS], [VIEW1.T5B.TEMPORARY], [VIEW1.T5B.GENERATED], [VIEW1.T5B.SECONDARY], [VIEW1.T5B.NAMESPACE], [VIEW1.T5B.EDITION_NAME]), filter(nil), 
      access([VIEW1.T5B.OWNER], [VIEW1.T5B.OBJECT_NAME], [VIEW1.T5B.SUBOBJECT_NAME], [VIEW1.T5B.OBJECT_ID], [VIEW1.T5B.DATA_OBJECT_ID], [VIEW1.T5B.OBJECT_TYPE], [VIEW1.T5B.CREATED], [VIEW1.T5B.LAST_DDL_TIME], [VIEW1.T5B.TIMESTAMP], [VIEW1.T5B.STATUS], [VIEW1.T5B.TEMPORARY], [VIEW1.T5B.GENERATED], [VIEW1.T5B.SECONDARY], [VIEW1.T5B.NAMESPACE], [VIEW1.T5B.EDITION_NAME])
  2 - output([T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), filter(nil), limit(?), offset(?)
  3 - output([T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), filter(nil), sort_keys([T5.OWNER, ASC], [T5.OBJECT_NAME, ASC]), topn(cast(cast(?, NUMBER(-1, -1)) + cast(?, NUMBER(-1, -1)), BIGINT(-1, 0)))
  4 - output([T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), filter(nil), 
      access([T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), partitions(p0), 
      is_index_back=false, 
      range_key([T5.__pk_increment]), range(MIN ; MAX)always true

注意，OB(ORACLE) 并没有 ORDER BY ORDER OF t
这种用法。也不支持 with ur
这种脏读语法。

标准的 ORACLE 分页框架用法就是两层嵌套循环，最里层是先对结果集排序。这点跟 DB2 的写法不同。当表很大并且查询条件不是很好发挥索引性能的时候，这个分页查询性能往往会不好。主要是最里面的那层排序耗时。

特殊的业务场景下，如果要查询的字段很少，能通过索引覆盖的话，就可以提升性能。或者将这个查询改写为自关联的表连接,以提升分页的性能。

分页示例优化（假设 t5 表很大）：

EXPLAIN EXTENDED_NOADDR 
SELECT l.rn, r.* FROM (
	SELECT rownum rn,rid 
	FROM (
		SELECT rowid rid
		FROM t5
		ORDER BY owner,object_name
	) 	WHERE rownum <= 100
) l JOIN t5 r ON l.rid=r.rowid  
WHERE rn >=90
ORDER BY l.rn
;

=====================================================
|ID|OPERATOR          |NAME          |EST. ROWS|COST|
-----------------------------------------------------
|0 |SORT              |              |358      |2645|
|1 | MERGE JOIN       |              |358      |1877|
|2 |  TABLE SCAN      |R(rowid_index)|729      |1002|
|3 |  SORT            |              |50       |775 |
|4 |   SUBPLAN SCAN   |L             |50       |727 |
|5 |    COUNT         |              |100      |723 |
|6 |     SUBPLAN SCAN |VIEW1         |100      |722 |
|7 |      LIMIT       |              |100      |720 |
|8 |       TOP-N SORT |              |100      |719 |
|9 |        TABLE SCAN|T5            |729      |570 |
=====================================================

Outputs & filters: 
-------------------------------------
  0 - output([L.RN], [R.OWNER], [R.OBJECT_NAME], [R.SUBOBJECT_NAME], [R.OBJECT_ID], [R.DATA_OBJECT_ID], [R.OBJECT_TYPE], [R.CREATED], [R.LAST_DDL_TIME], [R.TIMESTAMP], [R.STATUS], [R.TEMPORARY], [R.GENERATED], [R.SECONDARY], [R.NAMESPACE], [R.EDITION_NAME]), filter(nil), sort_keys([L.RN, ASC])
  1 - output([L.RN], [R.OWNER], [R.OBJECT_NAME], [R.SUBOBJECT_NAME], [R.OBJECT_ID], [R.DATA_OBJECT_ID], [R.OBJECT_TYPE], [R.CREATED], [R.LAST_DDL_TIME], [R.TIMESTAMP], [R.STATUS], [R.TEMPORARY], [R.GENERATED], [R.SECONDARY], [R.NAMESPACE], [R.EDITION_NAME]), filter(nil), 
      equal_conds([L.RID = R.ROWID]), other_conds(nil)
  2 - output([R.ROWID], [R.OWNER], [R.OBJECT_NAME], [R.SUBOBJECT_NAME], [R.OBJECT_ID], [R.DATA_OBJECT_ID], [R.OBJECT_TYPE], [R.CREATED], [R.LAST_DDL_TIME], [R.TIMESTAMP], [R.STATUS], [R.TEMPORARY], [R.GENERATED], [R.SECONDARY], [R.NAMESPACE], [R.EDITION_NAME]), filter(nil), 
      access([R.__pk_increment], [R.ROWID], [R.OWNER], [R.OBJECT_NAME], [R.SUBOBJECT_NAME], [R.OBJECT_ID], [R.DATA_OBJECT_ID], [R.OBJECT_TYPE], [R.CREATED], [R.LAST_DDL_TIME], [R.TIMESTAMP], [R.STATUS], [R.TEMPORARY], [R.GENERATED], [R.SECONDARY], [R.NAMESPACE], [R.EDITION_NAME]), partitions(p0), 
      is_index_back=false, 
      range_key([R.ROWID]), range(MIN ; MAX)always true
  3 - output([L.RN], [L.RID]), filter(nil), sort_keys([L.RID, ASC])
  4 - output([L.RID], [L.RN]), filter([L.RN >= 90]), 
      access([L.RID], [L.RN])
  5 - output([rownum()], [VIEW1..RID]), filter(nil)
  6 - output([VIEW1..RID]), filter(nil), 
      access([VIEW1..RID])
  7 - output([T5.ROWID]), filter(nil), limit(?), offset(nil)
  8 - output([T5.ROWID]), filter(nil), sort_keys([T5.OWNER, ASC], [T5.OBJECT_NAME, ASC]), topn(?)
  9 - output([T5.ROWID], [T5.OWNER], [T5.OBJECT_NAME]), filter(nil), 
      access([T5.__pk_increment], [T5.ROWID], [T5.OWNER], [T5.OBJECT_NAME]), partitions(p0), 
      is_index_back=false, 
      range_key([T5.__pk_increment]), range(MIN ; MAX)always true

如果 DB2 的 row_number
函数不仅仅是分页，还有分组功能，那就转换为 ORACLE 对应的 row_number
函数用法。

`FETCH FIRST N ROWS`

在 OB 3.1 版本之后，也支持 FETCH FIRST N ROWS
。3.1 版本之前，就按上面分页方法。

explain extended_noaddr 
SELECT *
FROM (
	SELECT row_number() OVER (ORDER BY owner,object_name) rn , t5.* FROM t5 ORDER BY owner, object_name FETCH FIRST 100 ROWS ONLY
) t5b 
WHERE rn >=90;

==========================================
|ID|OPERATOR         |NAME|EST. ROWS|COST|
------------------------------------------
|0 |SUBPLAN SCAN     |T5B |50       |2749|
|1 | LIMIT           |    |100      |2746|
|2 |  WINDOW FUNCTION|    |100      |2744|
|3 |   SORT          |    |729      |2357|
|4 |    TABLE SCAN   |T5  |729      |891 |
==========================================

Outputs & filters: 
-------------------------------------
  0 - output([T5B.RN], [T5B.OWNER], [T5B.OBJECT_NAME], [T5B.SUBOBJECT_NAME], [T5B.OBJECT_ID], [T5B.DATA_OBJECT_ID], [T5B.OBJECT_TYPE], [T5B.CREATED], [T5B.LAST_DDL_TIME], [T5B.TIMESTAMP], [T5B.STATUS], [T5B.TEMPORARY], [T5B.GENERATED], [T5B.SECONDARY], [T5B.NAMESPACE], [T5B.EDITION_NAME]), filter([T5B.RN >= 90]), 
      access([T5B.RN], [T5B.OWNER], [T5B.OBJECT_NAME], [T5B.SUBOBJECT_NAME], [T5B.OBJECT_ID], [T5B.DATA_OBJECT_ID], [T5B.OBJECT_TYPE], [T5B.CREATED], [T5B.LAST_DDL_TIME], [T5B.TIMESTAMP], [T5B.STATUS], [T5B.TEMPORARY], [T5B.GENERATED], [T5B.SECONDARY], [T5B.NAMESPACE], [T5B.EDITION_NAME])
  1 - output([T_WIN_FUN_ROW_NUMBER()], [T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), filter(nil), limit(?), offset(nil)
  2 - output([T_WIN_FUN_ROW_NUMBER()], [T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), filter(nil), 
      win_expr(T_WIN_FUN_ROW_NUMBER()), partition_by(nil), order_by([T5.OWNER, ASC], [T5.OBJECT_NAME, ASC]), window_type(RANGE), upper(UNBOUNDED PRECEDING), lower(UNBOUNDED FOLLOWING)
  3 - output([T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), filter(nil), sort_keys([T5.OWNER, ASC], [T5.OBJECT_NAME, ASC])
  4 - output([T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), filter(nil), 
      access([T5.OWNER], [T5.OBJECT_NAME], [T5.SUBOBJECT_NAME], [T5.OBJECT_ID], [T5.DATA_OBJECT_ID], [T5.OBJECT_TYPE], [T5.CREATED], [T5.LAST_DDL_TIME], [T5.TIMESTAMP], [T5.STATUS], [T5.TEMPORARY], [T5.GENERATED], [T5.SECONDARY], [T5.NAMESPACE], [T5.EDITION_NAME]), partitions(p0), 
      is_index_back=false, 
      range_key([T5.__pk_increment]), range(MIN ; MAX)always true

列名重复问题

DB2 的查询结果集可以接受有重复的列名（投影列），但是 OB(ORACLE) 不行。

DB2 示例：

db2 => SELECT * FROM (SELECT id,c1,c1 FROM t3);

ID                   C1                                                 C1
-------------------- -------------------------------------------------- --------------------------------------------------
                   1
                   2

  2 record(s) selected.

db2 =>

ORACLE 示例：

obclient> SELECT * FROM (SELECT id,c1,c1  FROM t3);
ORA-00918: column 'C1' in field list ambiguously defined
obclient>

表别名问题

DB2 表别名语法用 AS
关键字，是可选的。但是 OB(ORACLE) 表别名不支持 AS
关键字。

DB2 示例：

db2 => select id , c1 as c1_new from t4  as t4_new ;

ID                   C1_NEW
-------------------- --------------------------------------------------
                   1 a

  1 record(s) selected.

db2 => select id , c1 c1_new from t4  t4_new ;

ID                   C1_NEW
-------------------- --------------------------------------------------
                   1 a

  1 record(s) selected.

OB(ORACLE) 示例：

obclient> select id , c1 as c1_new from t4  as t4_new ;
ORA-00900: You have an error in your SQL syntax; check the manual that corresponds to your OceanBase version for the right syntax to use near 't4_new' at line 1
obclient> select id , c1 as c1_new from t4  t4_new ;
+----+--------+
| ID | C1_NEW |
+----+--------+
|  2 | a      |
+----+--------+
1 row in set (0.00 sec)
obclient>

DB2 的 ORACLE 兼容模式

为了减少 DB2 和 ORACLE 的不兼容逻辑，或者说 IBM 当初为了减少 ORALCE 应用迁移到 DB2 上的修改成本，DB2 可以开启 ORACLE 兼容模式。这个需要在创建数据库的时候就指定，后期不能修改。适合以 OB(ORACLE) 为主，DB2 为异构备的方案。

DB2 数据库开启 ORACLE 兼容模式方法：

[db2inst1@08935f5bff4f ~]$ db2set -all
[i] DB2COMM=TCPIP
[g] DB2SYSTEM=08935f5bff4f
[db2inst1@08935f5bff4f ~]$ db2set DB2_COMPATIBILITY_VECTOR=ORA
[db2inst1@08935f5bff4f ~]$ db2set -all
[i] DB2_COMPATIBILITY_VECTOR=ORA
[i] DB2COMM=TCPIP
[g] DB2SYSTEM=08935f5bff4f
[db2inst1@08935f5bff4f ~]$ db2stop
01/28/2021 03:03:23 0 0 SQL1064N DB2STOP processing was successful.
SQL1064N DB2STOP processing was successful.
[db2inst1@08935f5bff4f ~]$ db2start
01/28/2021 03:05:13 0 0 SQL1063N DB2START processing was successful.
SQL1063N DB2START processing was successful.
[db2inst1@08935f5bff4f ~]$
[db2inst1@59f37c64eddd ~]$ db2 create db db2ora using codeset utf-8 territory US
DB20000I The CREATE DATABASE command completed successfully.
[db2inst1@59f37c64eddd ~]$

DB2 建表示例

[db2inst1@a32408ac626d ~]$ db2 "DROP TABLE T10;"
DB20000I  The SQL command completed successfully.
[db2inst1@a32408ac626d ~]$ db2 "CREATE TABLE T10 (
> ID BIGINT NOT NULL GENERATED BY DEFAULT AS IDENTITY (MAXVALUE 9223372036854775807),
> C1 NUMBER NOT NULL,
> C2 VARCHAR2(50) NOT NULL,
> C3 NVARCHAR2(50) NOT NULL,
> C4 DATE NOT NULL DEFAULT SYSDATE,
> C5 TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP
> );"
DB20000I  The SQL command completed successfully.
[db2inst1@a32408ac626d ~]$
[db2inst1@a32408ac626d ~]$ db2 "INSERT INTO t10(c1,c2,c3) values(1,'a','A');"
DB20000I  The SQL command completed successfully.
[db2inst1@a32408ac626d ~]$
[db2inst1@a32408ac626d ~]$ db2 "INSERT INTO t10(c1,c2,c3) values(2,'b','');"
DB21034E  The command was processed as an SQL statement because it was not a
valid Command Line Processor command.  During SQL processing it returned:
SQL0407N  Assignment of a NULL value to a NOT NULL column "TBSPACEID=2,
TABLEID=523, COLNO=3" is not allowed.  SQLSTATE=23502

可见，当开启 ORACLE 兼容模式后，支持 ORACLE 的列类型，并且 NULL
跟空值也等同对待了。

查看一下表结构,可见在 DB2 内部，表的列还是以 DB2 的原生类型存储。

[db2inst1@a32408ac626d ~]$ db2look -d testdb -e -t T10
-- No userid was specified, db2look tries to use Environment variable USER
-- USER is: DB2INST1
-- The db2look utility will consider only the specified tables
-- Creating DDL for table(s)
-- This CLP file was created using DB2LOOK Version "11.5"
-- Timestamp: Sat Aug 14 09:52:21 2021
-- Database Name: TESTDB
-- Database Manager Version: DB2/LINUXX8664 Version 11.5.6.0
-- Database Codepage: 1208
-- Database Collating Sequence is: IDENTITY
-- Alternate collating sequence(alt_collate): null
-- varchar2 compatibility(varchar2_compat): ON

-- Binding package automatically ...
-- Bind is successful

CONNECT TO TESTDB;

------------------------------------------------
-- DDL Statements for Table "TPCC"."T10"
------------------------------------------------

CREATE TABLE "TPCC"."T10"  (
                  "ID" BIGINT NOT NULL GENERATED BY DEFAULT AS IDENTITY (
                    START WITH +1
                    INCREMENT BY +1
                    MINVALUE +1
                    MAXVALUE +9223372036854775807
                    NO CYCLE
                    CACHE 20
                    NO ORDER ) ,
                  "C1" DECFLOAT(16) NOT NULL ,
                  "C2" VARCHAR(50 OCTETS) NOT NULL ,
                  "C3" VARCHAR(50 CODEUNITS32) NOT NULL ,
                  "C4" TIMESTAMP(0) NOT NULL WITH DEFAULT "SYSIBM"."TIMESTAMP"(CURRENT TIMESTAMP) ,
                  "C5" TIMESTAMP NOT NULL WITH DEFAULT CURRENT TIMESTAMP )
                 IN "USERSPACE1"
                 ORGANIZE BY ROW;

ALTER TABLE "TPCC"."T10" ALTER COLUMN "ID" RESTART WITH 21;

COMMIT WORK;

CONNECT RESET;

TERMINATE;

[db2inst1@a32408ac626d ~]$

注意，如果如果涉及到国家字符集，则按DB2的 Unicode 规则保存，每个汉字是 4 个字节。

DB2 和 OB(ORACLE) 的异构数据同步

DB2 和 ORACLE 语法的客观差异决定 DB2 到 OB(ORACLE) 的同步或者并轨运行不会是一件轻松的事情。实际项目有的是业务自己同步，有的是靠产品同步。

如果是产品做异构同步，业务需要做一些妥协。比如说遵守特定的数据库表结构设计规范、SQL 规范等等。传统的数据库同步产品 OGG 、DSG 等做异构同步也是同样的经验。

就 DB2 和 OB(ORACLE)之间的实时同步，可以使用 OB 官方数据同步产品（OMS）或者 OB 合作伙伴产品（DSG）。如果是离线同步，可以使用 Alibaba 开源的 DATAX 产品（ DATAX 的 OB 插件源码也已经随着 OB 社区版一起开源）。

OMS 的 DB2 同步功能包括：

DB2 和 OB(ORACLE) 彼此的全量表结构同步。
DB2 和 OB(ORACLE) 彼此的全量数据同步。
DB2 和 OB(ORACLE) 彼此的增量数据同步。
DB2 和 OB(ORACLE) 彼此的增量表结构同步。

注意，这里说的是具备双向同步的能力，但是只激活单向的同步链路。

非表结构对象在技术上存在一些客观的技术困难，需要人为分析再手动同步：

序列的同步。目标端再激活之前，序列的值需要推高。在 OB(ORACLE) 里目前推高序列的方法是重建序列。
同义词的同步。同义词链接的对象可能是其他用户或SCHEMA 下的对象，这有个依赖问题，需要分析后手动同步。
存储过程的同步。DB2 和 OB(ORACLE) 存储过程语法差异非常大，很多用法都要翻译为兼容性的实现。目前还需要手动做。

其他参考

单纯的看 DB2 到 OB(ORACLE) 的异构同步，是一件困难麻烦多收益少的事情。还需要其他角度来看这个事情。

首先是可以降低对 IBM 小机和存储等高端硬件的依赖，相应的数据库运维技术门槛也降低。稳定性方面 OB 通过多副本（3 副本或 5 副本）以及容灾集群来提高整体的可靠性。

第二是数据库架构从集中式转向分布式，业务的架构也相应的转向分布式架构，两条腿走路跑的快。虽然目前看复杂的 SQL，单条语句性能 OB 可能不如小机上的 DB2 跑的快，但是 OB 的并发处理能力强。当并发上去后（如50、100等等），复杂 SQL 在 OB 上的平均延时水平很可能比 DB2 低很多，吞吐能力会更高。

第三是成熟的开发商（ISV）的应用本身都适配过 DB2、ORACLE 和 MySQL。所以在核心业务表结构设计上对特定数据库依赖很少，尽可能的取各个数据库能力的交集。这样的业务做 DB2 到 OB(ORACLE) 的迁移，困难还会少很多。

附录

DB2 DOCKER 环境搭建

下载 db2 镜像

docker pull ibmcom/db2

docker images |grep db2
REPOSITORY                                            TAG                     IMAGE ID            CREATED             SIZE
ibmcom/db2                                            latest                  a6a5ee354fb1        7 weeks ago         2.95GB

起 db2 容器（兼容ORACLE）

# docker volume create db2-11.5 # docker run -itd --name db2_11.5 --privileged=true -p 50000:50000 \ -e LICENSE=accept \ -e DB2INST1_PASSWORD=db2inst1 \

-e DBNAME=testdb \

-e DBPORT=50000 \

-e TSPORT=55000 \

-e ENABLE_ORACLE_COMPATIBILITY=true \ -v db2-11.5:/database ibmcom/db2 -- 等 5 分钟，db2 容器初始化完毕 # docker logs db2_11.5 -- 进入 db2 容器 # docker exec -it db2-11.5 bash # su - db2inst1 -- 连接数据库，查看表信息 $ db2 connect to testdb $ db2 set current schema tpcc $ db2 list tables for schema tpcc $ db2 describe table t10 -- 查看 db2 归档日志设置，增量同步需要开启归档 $ db2 get db cfg for testdb |grep -i log -- 查看/修改 db2 默认监听端口 $ db2 get dbm cfg |grep -i tcp/ip TCP/IP Service name (SVCENAME) = db2c_db2inst1 $ cat /etc/services |grep db2c_db2inst1 db2c_db2inst1 50000/tcp db2c_db2inst1_ssl 50001/tcp --修改 /etc/services 端口后，重启 db2 $ db2stop $ db2start

DATAX FOR OB 同步任务文件示例

{
    "job": {
        "setting": {
            "speed": {
                "channel": 8,
            },
            "errorLimit": {
                "record": 10000
            }
        },
        "content": [
            {
                "reader": {
                    "name": "db2reader",
                    "parameter": {
                        "username": "db2inst1",
                        "password": "******",
                        "splitPK":"ID",
                        "column": ["id", "c1", "c2"] ,
                        "connection": [
                            {
                                "jdbcUrl": [
                                    "jdbc:db2://127.0.0.1:50000/testdb:currentSchema=TPCC;"
                                ],
                                "table": [
                                    "datax_test"
                                ]
                            }
                        ]
                    }
                },
                "writer": {
                    "name": "oceanbasev10writer",
                    "parameter": {
                        "writeMode": "insert",
                        "username": "tpcc",
                        "password": "******",
                        "column": ["id", "c1", "c2"] ,
                        "connection": [
                            {
                                "jdbcUrl": "||_dsc_ob10_dsc_||obdemo:oboracle||_dsc_ob10_dsc_||jdbc:oceanbase://127.0.0.1:2883/tpcc",
                                "table": [
                                    "datax_test"
                                ]
                            }
                        ]
                    }
                }
            }
        ]
    }
}

运行方法：

cd /home/admin/datax3 && bin/datax.py job/datax_test.json

更多阅读

OceanBase 企业版下载使用指引
从ORACLE/MySQL到OceanBase：数据导出&导入
OceanBase 索引列的NULL属性测试分析
闲话 OB 部署架构实践
OceanBase 性能诊断（一）

db2 oceanbase

文章转载自数据库技术闲谈，如果涉嫌侵权，请发送邮件至：contact@modb.pro进行举报，并提供相关证据，一经查实，墨天轮将立刻删除相关内容。