前言

博主所在单位目前使用Oracle GoldenGate将各个业务生产库汇聚到一起做数仓实时ODS平台，源端库可能涉及Oracle、Mysql、达梦、Guassdb库。
最近遇到了这个错误：OGG-03533：character 'c2 a0' at offset 0 that is not available。
导致报错的原因是源端库有业务人员手动用excel往数据库导入数据，源端库字符集：AMERICAN_AMERICA.AL32UTF8 ，目标库ODS字符集：AMERICAN_AMERICA.ZHS16GBK。
业务人员导入的数据中包含了全角空格及不间断空格，从utf8的数据库往gbk的数据库同步过程中，字符转码失败。
因为是金融领域生产环境，此同步进程中包含其他生产表，所以恢复要格外小心，否则可能会导致数据实时同步异常，数据错乱。
下面是我记录的本地测试解决方案。

一、重现报错

源端字符集：AMERICAN_AMERICA.AL32UTF8
ODS字符集：AMERICAN_AMERICA.ZHS16GBK

源端查询数据

2    SQL> select * from dept;
3    
4    DEPTNO DNAME    LOC
5    ‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
6    1 东东    a
7    10 ACCOUNTING    NEW YORK
8    20 RESEARCH    DALLAS
9    30 SALES    CHICAGO
10    40 OPERATIONS    BOSTON

ods端查询数据：

12    SQL> select deptno,dname,loc from dept;
13    
14    DEPTNO DNAME    LOC
15    ‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐
    ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
16    10 ACCOUNTING    NEW YORK
17    20 RESEARCH    DALLAS
18    30 SALES    CHICAGO
19    40 OPERATIONS    BOSTON
20    1 东东    a

源端执行插入操作：

22    
23    SQL> insert into dept values(60,'东
    123'||chr(to_number('C2A0','xxxx'))||'东A','a');
24    
25    1 row created.
26    
27    SQL> commit;
28    
29    Commit complete.

源端切换归档：

31    SQL> alter system switch logfile;
32    
33    System altered.

源端再次查询数据：

35    SQL> select * from dept;
36    
37    DEPTNO DNAME    LOC
38    ‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
39    1 东东    a
40    60 东123 东A    a
41    10 ACCOUNTING    NEW YORK
42    20 RESEARCH    DALLAS
43    30 SALES    CHICAGO
44    40 OPERATIONS    BOSTON
45    
46    6 rows selected.

ODS端查询进程状态：

49    GGSCI (ogg) 24> info all
50    
51    Program    Status    Group    Lag at Chkpt    Time Since Chkpt
52    
53    MANAGER    RUNNING
54    REPLICAT    ABENDED    R_SC    00:00:00    00:00:50

应用进程报错信息如下：

2018‐09‐20 22:39:36    INFO    OGG‐06510    Using the following key columns for target table SCOTT.DEPT_AUDIT: DEPTNO.


2018‐09‐20 22:39:36    INFO    OGG‐03010    Performing implicit conversion of column data from character set UTF‐8 to zhs16gbk.

Source Context :
SourceModule    : [ggapp.csconv]
SourceID    :
[/scratch/aime/adestore/views/aime_adc4150378/oggcore/OpenSys/src/ggli b/ggapp/csconv.cpp]
SourceFunction    : [conversion_error_handler]SourceLine    : [820]
ThreadBacktrace    : [18] elements
:
[/u01/app/oracle/ogg12/libgglog.so(CMessageContext::AddThreadContext()
+0x1b) [0x7f8e943ebf2b]]
:
[/u01/app/oracle/ogg12/libgglog.so(CMessageFactory::CreateMessage(CSou rceContext*, unsigned int, ...)+0x134) [0x7f8e943e5f84]]
:
[/u01/app/oracle/ogg12/libgglog.so(_MSG_ERR_COLUMN_CHARSET_CONVERSION_ CSTOCS2(CSourceContext*, char const*, unsigned long, ggs::gglib::ggapp::CDBObjName<(DBObjTy
pe)13> const&, ggs::gglib::ggapp::CDBObjName<(DBObjType)13> const&, char const*, char const*, CMessageFactory::MessageDisposition)+0x89) [0x7f8e943b4b63]]
:
[/u01/app/oracle/ogg12/replicat(conversion_error_handler(ULibCharSet, ULibCharSet, ULibError, ULibBadCharDetail const&, col_def const*, col_def const*, bool, ext
r_ptr_def*, bool&, bool&)+0x184) [0x634b66]]
:
[/u01/app/oracle/ogg12/replicat(CSCONV_convert(short, col_def const*, ULibCharSet, short, col_def*, ULibCharSet, char*, char*, col_def*, bool, extr_ptr_def*)+0x3
52) [0x635580]]
:
[/u01/app/oracle/ogg12/replicat(map_columns_int(ggs::gglib::gglcr::Com monLCR const&, ggs::gglib::gglcr::CommonLCR&, char*, map_col_def*, short*, short*, short, s
hort, extr_ptr_def*, bool, bool, bool)+0x1d33) [0x63f1e3]]
:
[/u01/app/oracle/ogg12/replicat(map_columns(ggs::gglib::gglcr::CommonL CR const&, ggs::gglib::gglcr::CommonLCR&, map_col_def*,  short*, short*, short, short, short
*, int, extr_ptr_def*)+0xfe) [0x63f70e]]
:
[/u01/app/oracle/ogg12/replicat(replicate_io(ggs::gglib::ggdatasource:
:DataSource*, file_def*, ggs::gglib::gglcr::CommonLCR const*, ggs::gglib::gglcr::CommonLCR*
, extr_ptr_def*, int*, int)+0x488) [0x61b858]]
:
[/u01/app/oracle/ogg12/replicat(process_record(ggs::gglib::ggdatasource::DataSource*, file_def*&, ggs::gglib::gglcr::CommonLCR const*, ggs::gglib::gglcr::CommonL
CR*, short&, extr_ptr_def*&, extr_ptr_def*&, int&, int&, int&, bool)+0x328) [0x623178]]
:
[/u01/app/oracle/ogg12/replicat(ggs::er::ReplicatContext::processRepli catLoop(ggs::Heartbeat::MapGeneratorParams&)+0x1261) [0x5da591]]
:
[/u01/app/oracle/ogg12/replicat(ggs::er::ReplicatContext::run(ggs::Hea rtbeat::MapGeneratorParams&)+0x29) [0x5ce579]]
: [/u01/app/oracle/ogg12/replicat()
[0x617936]]
:
[/u01/app/oracle/ogg12/replicat(ggs::gglib::MultiThreading::MainThread
::ExecMain()+0x60) [0x6d15c0]]
:
[/u01/app/oracle/ogg12/replicat(ggs::gglib::MultiThreading::Thread::Ru nThread(ggs::gglib::MultiThreading::Thread::ThreadArgs*)+0x14d) [0x6d25dd]]
:
[/u01/app/oracle/ogg12/replicat(ggs::gglib::MultiThreading::MainThread
::Run(int, char**)+0xb1) [0x6d26c1]]
: [/u01/app/oracle/ogg12/replicat(main+0x3b)
[0x61ab9b]]
: [/lib64/libc.so.6( libc_start_main+0xfd)
[0x3dc9c1ed1d]]
: [/u01/app/oracle/ogg12/replicat()
[0x54eb69]]

2018‐09‐20 22:44:28    ERROR    OGG‐03533    Conversion from character set UTF‐8 of source column DNAME to character set zhs16gbk of target column DNAME failed because the source column contains
a character 'c2 a0' at offset 0 that is not available in the target character set.

二、处理过程

在应用进程加入参数：

REPLACEBADCHAR ESCAPE

REPLACEBADCHAR 参 数 官 方 介 绍 如 下

REPLACEBADCHAR Valid For

Extract and Replicat

Description

Use the REPLACEBADCHAR parameter to control the response of the
process when a valid code point does not exist for either the source
or target character set when mapping character-type columns. By
default, the check for invalid code points is only performed when the
source and target databases have different character sets, and the
default response is to abend. You can use the FORCECHECK option to
force the process to check for invalid code points when the source and
target databases have the same character set. REPLACEBADCHAR applies
globally.

Default

ABORT Syntax

REPLACEBADCHAR {ABORT | SKIP | ESCAPE | SUBSTITUTE string | NULL |
SPACE} [FORCECHECK] [NOWARNING] ABORT The process abends on an invalid
code point. This is the default.

SKIP The process skips the record that has the invalid code point. Use
this option with caution, because skipping a record can cause data
discrepancies on the target.

ESCAPE The process replaces the data value with an escaped version of
the data value. Depending on the character set of the source database,
the value is output as one of the following:

If the source data is not UTF-16 (NCHAR/NVARCHAR), the output is
hexadecimal (\xXX).

If the source data is UTF-16, the output is Unicode (\uXXXX).

SUBSTITUTE string The process replaces the data with a specified
string, either Unicode notation or up to four characters. By default
the default substitution character of the target character set is used
for replacement.

NULL The process replaces an invalid character with the value of NULL
if the target column is nullable or, otherwise, assigns a white space
(U+0020).

SPACE The process replaces an invalid character with a white space
(U+0020).

FORCECHECK The process checks for invalid code points when the source
and target databases have identical character sets. This overrides the
default, where the validation is skipped when the source and target
character sets are identical.

NOWARNING The process suppresses warning messages related to
conversion and validation errors.

Examples

Example 1 The following example replaces invalid code points with
the value of NULL.

REPLACEBADCHAR NULL Example 2 Because ESCAPE is specified, Oracle
GoldenGate will replace the Euro symbol in a source NCHAR column with
the escaped version of u20AC, because the target is ISO-8859-1, which
does not support the Euro code point.

REPLACEBADCHAR ESCAPE Example 3 The following substitutes a control
character for invalid characters.

REPLACEBADCHAR SUBSTITUTE \u001A

重启进程：

1    GGSCI (ogg) 25> edit params r_Sc    
2        
3    GGSCI (ogg) 26> start r_sc    
4        
5    Sending START request to MANAGER ...    
6    REPLICAT R_SC starting    
7        
8        
9    GGSCI (ogg) 27> info all    
10        
11    Program    Status    Group    Lag at Chkpt    Time Since Chkpt
12        
13    MANAGER    RUNNING    
14    REPLICAT    RUNNING    R_SC    00:00:00    00:00:01

进程状态正常，再去ODS端查询数据：

1    SQL> select deptno,dname,loc from dept;
2    
3    DEPTNO DNAME    LOC
4    ‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐
    ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
5    10 ACCOUNTING    NEW YORK
6    20 RESEARCH    DALLAS
7    30 SALES    CHICAGO
8    40 OPERATIONS    BOSTON
9    1 东东    a
10    60 东123\xC2\xA0东A    a
11    
12    

发现之前报错的utf8编码为 c2a0的字符已在ODS端的表中替换成了源端utf8的字符编码。而其他字符还是正常的，可以在报错后不改数据库字符集的前提下将特殊字符转换成相应的uft8编码，再通过人工写函数或则存过映射替换修改这类字符，或则可以考虑将不间断空格改为GBK中的普通空格，使用如下配置：

REPLACEBADCHAR SPACE

ODS端使用这个参数后再次在源端插入相同的字符串看下：

源端插入数据：

2    SQL> insert into dept values(61,'东
    123'||chr(to_number('C2A0','xxxx'))||'东A','a');
3    
4    1 row created.
5    
6    SQL> commit;
7    
8    Commit complete.
9    
10    SQL> alter system switch logfile;
11    
12    System altered.
13    SQL> select * from dept;
14    
15    DEPTNO DNAME    LOC
16    ‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
17    1 东东    a
18    60 东123 东A    a
19    61 东123 东A    a
20    10 ACCOUNTING    NEW YORK
21    20 RESEARCH    DALLAS
22    30 SALES    CHICAGO
23    40 OPERATIONS    BOSTON

ODS端查询：

25    SQL> select deptno,dname,loc from dept;
26    
27    DEPTNO DNAME    LOC
28    ‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ ‐‐‐‐‐‐‐‐
    ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
29    10 ACCOUNTING    NEW YORK
30    20 RESEARCH    DALLAS
31    30 SALES    CHICAGO
32    40 OPERATIONS    BOSTON
33    1 东东    a
34    60 东123\xC2\xA0东A    a
35    61 东123    东A    a

发现utf8的不间断空格c2a0已经被替换成GBK的普通空格（U+0020）了。问题解决，在生产环境做了配置，实时同步恢复正常！

总结

本文主要介绍的是通过GoldenGate实时同步两个字符集不同的数据库时，如果遇到字符不兼容的问题时，应该怎么处理的解决方案，生产环境需谨慎，最后测试完无异常后在生产环境做了修改。