MySQL基于ROW格式的数据恢复

大家都知道MySQL Binlog 有三种格式，分别是Statement、Row、Mixd。Statement记录了用户执行的原始SQL，而Row则是记录了行的修改情况，在MySQL 5.6以上的版本默认是Mixd格式，但为了保证复制数据的完整性，建议生产环境都使用Row格式，就前面所说的Row记录的是行数据的修改情况，而不是原始SQL。那么线上或者测试环境误操删除或者更新几条数据后，又想恢复，那怎么办呢？下面演示基于Binlog格式为Row的误操后数据恢复，那么怎么把Binlog解析出来生成反向的原始SQL呢？下面我们一起来学习。

下面我们使用 binlog-rollback.pl 对数据进行恢复演示。（这脚本的作者不知道是谁，Github上也没找到这个脚本，所以无法标明出处），脚本是用Perl语言写的，非常好用的一个脚本，当然你也可以用Shell或者Python脚本来实现，下面是脚本的代码：

#!/usr/lib/perl -w

use strict;

use warnings;

use Class::Struct;

use Getopt::Long qw(:config no_ignore_case);                    # GetOption

# register handler system signals

use sigtrap 'handler', \&sig_int, 'normal-signals';

# catch signal

sub sig_int(){

    my ($signals) = @_;

    print STDERR "# Caught SIG$signals.\n";

    exit ;

}

my %opt;

my $srcfile;

my $host = '127.0.0.1';

my $port = ;

my ($user,$pwd);

my ($MYSQL, $MYSQLBINLOG, $ROLLBACK_DML);

my $outfile = '/dev/null';

my (%do_dbs,%do_tbs);

# tbname=>tbcol, tbcol: @n=>colname,type

my %tbcol_pos;

my $SPLITER_COL = ',';

my $SQLTYPE_IST = 'INSERT';

my $SQLTYPE_UPD = 'UPDATE';

my $SQLTYPE_DEL = 'DELETE';

my $SQLAREA_WHERE = 'WHERE';

my $SQLAREA_SET = 'SET';

my $PRE_FUNCT = '========================== ';

# =========================================================

# 基于row模式的binlog，生成DML(insert/update/delete)的rollback语句

# 通过mysqlbinlog -v 解析binlog生成可读的sql文件

# 提取需要处理的有效sql

#     "### "开头的行.如果输入的start-position位于某个event group中间，则会导致"无法识别event"错误

#

# 将INSERT/UPDATE/DELETE 的sql反转,并且1个完整sql只能占1行

#     INSERT: INSERT INTO => DELETE FROM, SET => WHERE

#     UPDATE: WHERE => SET, SET => WHERE

#     DELETE: DELETE FROM => INSERT INTO, WHERE => SET

# 用列名替换位置@{1,2,3}

#     通过desc table获得列顺序及对应的列名

#     特殊列类型value做特别处理

# 逆序

#

# 注意:

#     表结构与现在的表结构必须相同[谨记]

#     由于row模式是幂等的，并且恢复是一次性，所以只提取sql，不提取BEGIN/COMMIT

#     只能对INSERT/UPDATE/DELETE进行处理

# ========================================================

sub main{

    # get input option

    &get_options();

    #

    &init_tbcol();

    #

    &do_binlog_rollback();

}

&main();

# ----------------------------------------------------------------------------------------

# Func : get options and set option flag

# ----------------------------------------------------------------------------------------

sub get_options{

    #Get options info

    GetOptions(\%opt,

        'help',                    # OUT : print help info

        'f|srcfile=s',            # IN  : binlog file

        'o|outfile=s',            # out : output sql file

        'h|host=s',                # IN  :  host

        'u|user=s',             # IN  :  user

        'p|password=s',         # IN  :  password

        'P|port=i',                # IN  :  port

        'start-datetime=s',        # IN  :  start datetime

        'stop-datetime=s',        # IN  :  stop datetime

        'start-position=i',        # IN  :  start position

        'stop-position=i',        # IN  :  stop position

        'd|database=s',            # IN  :  database, split comma

        'T|table=s',            # IN  :  table, split comma

        'i|ignore',                # IN  :  ignore binlog check ddl and so on

        'debug',                # IN  :  print debug information

      ) or print_usage();

    if (!scalar(%opt)) {

        &print_usage();

    }

    # Handle for options

    if ($opt{'f'}){

        $srcfile = $opt{'f'};

    }else{

        &merror("please input binlog file");

    }

    $opt{'h'} and $host = $opt{'h'};

    $opt{'u'} and $user = $opt{'u'};

    $opt{'p'} and $pwd = $opt{'p'};

    $opt{'P'} and $port = $opt{'P'};

    if ($opt{'o'}) {

        $outfile = $opt{'o'};

        # 清空 outfile

        `echo '' > $outfile`;

    }

    #

    $MYSQL = qq{mysql -h$host -u$user -p'$pwd' -P$port};

    &mdebug("get_options::MYSQL\n\t$MYSQL");

    # 提取binlog,不需要显示列定义信息，用-v，而不用-vv

    $MYSQLBINLOG = qq{mysqlbinlog -v};

    $MYSQLBINLOG .= " --start-position=".$opt{'start-position'} if $opt{'start-position'};

    $MYSQLBINLOG .= " --stop-position=".$opt{'stop-position'} if $opt{'stop-postion'};

    $MYSQLBINLOG .= " --start-datetime='".$opt{'start-datetime'}."'" if $opt{'start-datetime'};

    $MYSQLBINLOG .= " --stop-datetime='$opt{'stop-datetime'}'" if $opt{'stop-datetime'};

    $MYSQLBINLOG .= " $srcfile";

    &mdebug("get_options::MYSQLBINLOG\n\t$MYSQLBINLOG");

    # 检查binlog中是否含有 ddl sql: CREATE|ALTER|DROP|RENAME

    &check_binlog() unless ($opt{'i'});

    # 不使用mysqlbinlog过滤，USE dbname;方式可能会漏掉某些sql，所以不在mysqlbinlog过滤

    # 指定数据库

    if ($opt{'d'}){

        my @dbs = split(/,/,$opt{'d'});

        foreach my $db (@dbs){

            $do_dbs{$db}=;

        }

    }

    # 指定表

    if ($opt{'T'}){

        my @tbs = split(/,/,$opt{'T'});

        foreach my $tb (@tbs){

            $do_tbs{$tb}=;

        }

    }

    # 提取有效DML SQL

    $ROLLBACK_DML = $MYSQLBINLOG." | grep '^### '";

    # 去掉注释: '### ' -> ''

    # 删除首尾空格

    $ROLLBACK_DML .= " | sed 's/###\\s*//g;s/\\s*\$//g'";

    &mdebug("rollback dml\n\t$ROLLBACK_DML");

    # 检查内容是否为空

    my $cmd = "$ROLLBACK_DML | wc -l";

    &mdebug("check contain dml sql\n\t$cmd");

    my $size = `$cmd`;

    chomp($size);

    unless ($size >){

        &merror("binlog DML is empty:$ROLLBACK_DML");

    };

}    

# ----------------------------------------------------------------------------------------

# Func :  check binlog contain DDL

# ----------------------------------------------------------------------------------------

sub check_binlog{

    &mdebug("$PRE_FUNCT check_binlog");

    my $cmd = "$MYSQLBINLOG ";

    $cmd .= " | grep -E -i '^(CREATE|ALTER|DROP|RENAME)' ";

    &mdebug("check binlog has DDL cmd\n\t$cmd");

    my $ddlcnt = `$cmd`;

    chomp($ddlcnt);

    my $ddlnum = `$cmd | wc -l`;

    chomp($ddlnum);

    my $res = ;

    if ($ddlnum>){

        # 在ddl sql前面加上前缀<DDL>

        $ddlcnt = `echo '$ddlcnt' | sed 's/^/<DDL>/g'`;

        &merror("binlog contain $ddlnum DDL:$MYSQLBINLOG. ddl sql:\n$ddlcnt");

    }

    return $res;

}

# ----------------------------------------------------------------------------------------

# Func : init all table column order

#        if input --database --table params, only get set table column order

# ----------------------------------------------------------------------------------------

sub init_tbcol{

    &mdebug("$PRE_FUNCT init_tbcol");

    # 提取DML语句

    my $cmd .= "$ROLLBACK_DML | grep -E '^(INSERT|UPDATE|DELETE)'";

    # 提取表名，并去重

    #$cmd .= " | awk '{if (\$1 ~ \"^UPDATE\") {print \$2}else {print \$3}}' | uniq ";

    $cmd .= " | awk '{if (\$1 ~ \"^UPDATE\") {print \$2}else {print \$3}}' | sort | uniq ";

    &mdebug("get table name cmd\n\t$cmd");

    open ALLTABLE, "$cmd | " or die "can't open file:$cmd\n";

    while (my $tbname = <ALLTABLE>){

        chomp($tbname);

        #if (exists $tbcol_pos{$tbname}){

        #    next;

        #}

        &init_one_tbcol($tbname) unless (&ignore_tb($tbname));

    }

    close ALLTABLE or die "can't close file:$cmd\n";

    # init tb col

    foreach my $tb (keys %tbcol_pos){

        &mdebug("tbname->$tb");

        my %colpos = %{$tbcol_pos{$tb}};

        foreach my $pos (keys %colpos){

            my $col = $colpos{$pos};

            my ($cname,$ctype) = split(/$SPLITER_COL/, $col);

            &mdebug("\tpos->$pos,cname->$cname,ctype->$ctype");

        }

    }

};

# ----------------------------------------------------------------------------------------

# Func : init one table column order

# ----------------------------------------------------------------------------------------

sub init_one_tbcol{

    my $tbname = shift;

    &mdebug("$PRE_FUNCT init_one_tbcol");

    # 获取表结构及列顺序

    my $cmd = $MYSQL." --skip-column-names --silent -e 'desc $tbname'";

    # 提取列名，并拼接

    $cmd .= " | awk -F\'\\t\' \'{print NR\"$SPLITER_COL`\"\$1\"`$SPLITER_COL\"\$2}'";

    &mdebug("get table column infor cmd\n\t$cmd");

    open TBCOL,"$cmd | " or die "can't open desc $tbname;";

    my %colpos;

    while (my $line = <TBCOL>){

        chomp($line);

        my ($pos,$col,$coltype) = split(/$SPLITER_COL/,$line);

        &mdebug("linesss=$line\n\t\tpos=$pos\n\t\tcol=$col\n\t\ttype=$coltype");

        $colpos{$pos} = $col.$SPLITER_COL.$coltype;

    }

    close TBCOL or die "can't colse desc $tbname";

    $tbcol_pos{$tbname} = \%colpos;

}

# ----------------------------------------------------------------------------------------

# Func :  rollback sql:    INSERT/UPDATE/DELETE

# ----------------------------------------------------------------------------------------

sub do_binlog_rollback{

    my $binlogfile = "$ROLLBACK_DML ";

    &mdebug("$PRE_FUNCT do_binlog_rollback");

    # INSERT|UPDATE|DELETE

    my $sqltype;

    # WHERE|SET

    my $sqlarea;

    my ($tbname, $sqlstr) = ('', '');

    my ($notignore, $isareabegin) = (,);

    # output sql file

    open SQLFILE, ">> $outfile" or die "Can't open sql file:$outfile";

    # binlog file

    open BINLOG, "$binlogfile |" or die "Can't open file: $binlogfile";

    while (my $line = <BINLOG>){

        chomp($line);

        if ($line =~ /^(INSERT|UPDATE|DELETE)/){

            # export sql

            if ($sqlstr ne ''){

                $sqlstr .= ";\n";

                print SQLFILE $sqlstr;

                &mdebug("export sql\n\t".$sqlstr);

                $sqlstr = '';

            }

            if ($line =~ /^INSERT/){

                $sqltype = $SQLTYPE_IST;

                $tbname = `echo '$line' | awk '{print \$3}'`;

                chomp($tbname);

                $sqlstr = qq{DELETE FROM $tbname};

            }elsif ($line =~ /^UPDATE/){

                $sqltype = $SQLTYPE_UPD;

                $tbname = `echo '$line' | awk '{print \$2}'`;

                chomp($tbname);

                $sqlstr = qq{UPDATE $tbname};

            }elsif ($line =~ /^DELETE/){

                $sqltype = $SQLTYPE_DEL;

                $tbname = `echo '$line' | awk '{print \$3}'`;

                chomp($tbname);

                $sqlstr = qq{INSERT INTO $tbname};

            }

            # check ignore table

            if(&ignore_tb($tbname)){

                $notignore = ;

                &mdebug("<BINLOG>#IGNORE#:line:".$line);

                $sqlstr = '';

            }else{

                $notignore = ;

                &mdebug("<BINLOG>#DO#:line:".$line);

            }

        }else {

            if($notignore){

                &merror("can't get tbname") unless (defined($tbname));

                if ($line =~ /^WHERE/){

                    $sqlarea = $SQLAREA_WHERE;

                    $sqlstr .= qq{ SET};

                    $isareabegin = ;

                }elsif ($line =~ /^SET/){

                    $sqlarea = $SQLAREA_SET;

                    $sqlstr .= qq{ WHERE};

                    $isareabegin = ;

                }elsif ($line =~ /^\@/){

                    $sqlstr .= &deal_col_value($tbname, $sqltype, $sqlarea, $isareabegin, $line);

                    $isareabegin = ;

                }else{

                    &mdebug("::unknown sql:".$line);

                }

            }

        }

    }

    # export last sql

    if ($sqlstr ne ''){

        $sqlstr .= ";\n";

        print SQLFILE $sqlstr;

        &mdebug("export sql\n\t".$sqlstr);

    }

    close BINLOG or die "Can't close binlog file: $binlogfile";

    close SQLFILE or die "Can't close out sql file: $outfile";

    # 逆序

    # 1!G: 只有第一行不执行G, 将hold space中的内容append回到pattern space

    # h: 将pattern space 拷贝到hold space

    # $!d: 除最后一行都删除

    my $invert = "sed -i '1!G;h;\$!d' $outfile";

    my $res = `$invert`;

    &mdebug("inverter order sqlfile :$invert");

}

# ----------------------------------------------------------------------------------------

# Func :  transfer column pos to name

#    deal column value

#

#  &deal_col_value($tbname, $sqltype, $sqlarea, $isareabegin, $line);

# ----------------------------------------------------------------------------------------

sub deal_col_value($$$$$){

    my ($tbname, $sqltype, $sqlarea, $isareabegin, $line) = @_;

    &mdebug("$PRE_FUNCT deal_col_value");

    &mdebug("input:tbname->$tbname,type->$sqltype,area->$sqlarea,areabegin->$isareabegin,line->$line");

    my @vals = split(/=/, $line);

    my $pos = substr($vals[],);

    my $valstartpos = length($pos)+;

    my $val = substr($line,$valstartpos);

    my %tbcol = %{$tbcol_pos{$tbname}};

    my ($cname,$ctype) = split(/$SPLITER_COL/,$tbcol{$pos});

    &merror("can't get $tbname column $cname type") unless (defined($cname) || defined($ctype));

    &mdebug("column infor:cname->$cname,type->$ctype");

    # join str

    my $joinstr;

    if ($isareabegin){

        $joinstr = ' ';

    }else{

        # WHERE 被替换为 SET, 使用 ,  连接

        if ($sqlarea eq $SQLAREA_WHERE){

            $joinstr = ', ';

        # SET 被替换为 WHERE 使用 AND 连接

        }elsif ($sqlarea eq $SQLAREA_SET){

            $joinstr = ' AND ';

        }else{

            &merror("!!!!!!The scripts error");

        }

    }

    #

    my $newline = $joinstr;

    # NULL value

    if (($val eq 'NULL') && ($sqlarea eq $SQLAREA_SET)){

        $newline .= qq{ $cname IS NULL};

    }else{

        # timestamp: record seconds

        if ($ctype eq 'timestamp'){

            $newline .= qq{$cname=from_unixtime($val)};

        # datetime: @n=yyyy-mm-dd hh::ii::ss

        }elsif ($ctype eq 'datetime'){

            $newline .= qq{$cname='$val'};

        }else{

            $newline .= qq{$cname=$val};

        }

    }

    &mdebug("\told>$line\n\tnew>$newline");

    return $newline;

}

# ----------------------------------------------------------------------------------------

# Func :  check is ignore table

# params: IN table full name #  format:`dbname`.`tbname`

# RETURN：

#        0 not ignore

#        1 ignore

# ----------------------------------------------------------------------------------------

sub ignore_tb($){

    my $fullname = shift;

    # 删除`

    $fullname =~ s/`//g;

    my ($dbname,$tbname) = split(/\./,$fullname);

    my $res = ;

    # 指定了数据库

    if ($opt{'d'}){

        # 与指定库相同

        if ($do_dbs{$dbname}){

            # 指定表

            if ($opt{'T'}){

                # 与指定表不同

                unless ($do_tbs{$tbname}){

                    $res = ;

                }

            }

        # 与指定库不同

        }else{

            $res = ;

        }

    }

    #&mdebug("Table check ignore:$fullname->$res");

    return $res;

}

# ----------------------------------------------------------------------------------------

# Func :  print debug msg

# ----------------------------------------------------------------------------------------

sub mdebug{

    my (@msg) = @_;

    print "<DEBUG>@msg\n" if ($opt{'debug'});

}

# ----------------------------------------------------------------------------------------

# Func :  print error msg and exit

# ----------------------------------------------------------------------------------------

sub merror{

    my (@msg) = @_;

    print "<Error>:@msg\n";

    &print_usage();

    exit();

}

# ----------------------------------------------------------------------------------------

# Func :  print usage

# ----------------------------------------------------------------------------------------

sub print_usage{

    print <<EOF;

==========================================================================================

Command line options :

    --help                # OUT : print help info

    -f, --srcfile            # IN  : binlog file. [required]

    -o, --outfile            # OUT : output sql file. [required]

    -h, --host            # IN  : host. default '127.0.0.1'

    -u, --user            # IN  : user. [required]

    -p, --password            # IN  : password. [required]

    -P, --port            # IN  : port. default '3306'

    --start-datetime        # IN  : start datetime

    --stop-datetime            # IN  : stop datetime

    --start-position        # IN  : start position

    --stop-position            # IN  : stop position

    -d, --database            # IN  : database, split comma

    -T, --table            # IN  : table, split comma. [required] set -d

    -i, --ignore            # IN  : ignore binlog check contain DDL(CREATE|ALTER|DROP|RENAME)

    --debug                # IN  :  print debug information

Sample :

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd'

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' -i

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --debug

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -h '192.168.1.2' -u 'user' -p 'pwd' -P

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --start-position=

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --start-position= --stop-position=

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' -d 'db1,db2'

   shell> perl binlog-rollback.pl -f 'mysql-bin.0000*' -o '/tmp/t.sql' -u 'user' -p 'pwd' -d 'db1,db2' -T 'tb1,tb2'

==========================================================================================

EOF

    exit;

}

;

这脚本含有注释以及使用说明，所以使用起来还是比较简单的，如果你会Perl语言，相信也很容易看懂代码。binlog-rollback.pl的使用参数如下：

[root@localhost mysql3306]# perl binlog-rollback.pl

==========================================================================================

Command line options :

        --help                          # OUT : print help info

        -f, --srcfile                   # IN  : binlog file. [required]

        -o, --outfile                   # OUT : output sql file. [required]

        -h, --host                      # IN  : host. default '127.0.0.1'

        -u, --user                      # IN  : user. [required]

        -p, --password                  # IN  : password. [required]

        -P, --port                      # IN  : port. default ''

        --start-datetime                # IN  : start datetime

        --stop-datetime                 # IN  : stop datetime

        --start-position                # IN  : start position

        --stop-position                 # IN  : stop position

        -d, --database                  # IN  : database, split comma

        -T, --table                     # IN  : table, split comma. [required] set -d

        -i, --ignore                    # IN  : ignore binlog check contain DDL(CREATE|ALTER|DROP|RENAME)

        --debug                         # IN  :  print debug information

Sample :

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd'

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' -i

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --debug

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -h '192.168.1.2' -u 'user' -p 'pwd' -P

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --start-position=

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --start-position= --stop-position=

   shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' -d 'db1,db2'

   shell> perl binlog-rollback.pl -f 'mysql-bin.0000*' -o '/tmp/t.sql' -u 'user' -p 'pwd' -d 'db1,db2' -T 'tb1,tb2'

==========================================================================================

[root@localhost mysql3306]#

下面主要演示对一个表的增、删、修（Insert/Delete/Update）操作，基于Binlog为Row格式的反向解析。

细心看脚本的朋友都能看到这个脚本需要提供一个连接MySQL的用户，主要是为了获取表结构。下面我们测试一个普通用户并给予SELECT权限即可，默认是host是127.0.0.1，这个可以修改脚本，我这里按脚本默认的：

<Test>[(none)]> GRANT SELECT ON *.* TO 'recovery'@'127.0.0.1' identified by '';

Query OK, 0 rows affected (0.08 sec)

<Test>[(none)]> flush privileges;

Query OK, 0 rows affected (0.04 sec)

<Test>[(none)]>

往xuanzhi库的表tb1里插入2行数据，记得binlog格式要为ROW：

<Test>[xuanzhi]> show global variables like 'binlog_format';

+---------------+-------+

| Variable_name | Value |

+---------------+-------+

| binlog_format | ROW   |

+---------------+-------+

1 row in set (0.00 sec)

<Test>[xuanzhi]> insert into xuanzhi.tb1 select 1,'aa';

Query OK, 1 row affected (0.01 sec)

Records: 1  Duplicates: 0  Warnings: 0

<Test>[xuanzhi]> insert into xuanzhi.tb1 select 2,'cc';

Query OK, 1 row affected (0.01 sec)

Records: 1  Duplicates: 0  Warnings: 0

<Test>[xuanzhi]> select * from xuanzhi.tb1;

+------+------+

| id   | name |

+------+------+

|    1 | aa   |

|    2 | cc   |

+------+------+

2 rows in set (0.00 sec)

<Test>[xuanzhi]>

为了看到运行脚本在不指定库看到的效果，我这里再往test库的user表插入两行数据：

<Test>[xuanzhi]> insert into test.user select 1,'user1',20;

Query OK, 1 row affected (0.03 sec)

Records: 1  Duplicates: 0  Warnings: 0

<Test>[xuanzhi]> insert into test.user select 2,'user2',30;

Query OK, 1 row affected (0.01 sec)

Records: 1  Duplicates: 0  Warnings: 0

<Test>[xuanzhi]>

查看此时的此时处于那个binlog：

<Test>[xuanzhi]> show master status;

+----------------------+----------+--------------+------------------+-------------------+

| File                 | Position | Binlog_Do_DB | Binlog_Ignore_DB | Executed_Gtid_Set |

+----------------------+----------+--------------+------------------+-------------------+

| localhost-bin.000023 |      936 |              |                  |                   |

+----------------------+----------+--------------+------------------+-------------------+

1 row in set (0.00 sec)

<Test>[xuanzhi]>

下面运行脚本 binlog-rollback.pl ，不指定任何库和表的情况下，这时表把binlog里所有DML操作都生成反向的SQL（最新的DML会生成在输入文件的最前面）：

[root@localhost mysql3306]# perl binlog-rollback.pl -f 'localhost-bin.000023'  -o '/data/t.sql' -u 'recovery' -p ''

mysql: [Warning] Using a password on the command line interface can be insecure.

mysql: [Warning] Using a password on the command line interface can be insecure.

[root@localhost mysql3306]#

我们查看输出的文件：/data/t.sql

[root@localhost mysql3306]# cat /data/t.sql

DELETE FROM `test`.`user` WHERE `id`= AND `name`='user2' AND `age`=;

DELETE FROM `test`.`user` WHERE `id`= AND `name`='user1' AND `age`=;

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='bb';

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='aa';

可以看到，INSERT操作的反向操作就是DELETE，这里把所有库的DML操作都查出来了，在后面会演示找单个库或者表所产生的反向SQL。

下面模拟运维人员、开发人员或者DBA误操删除数据，分别在不同的库删除一条记录：

<Test>[xuanzhi]> delete from xuanzhi.tb1 where id=2;

Query OK, 1 row affected (0.06 sec)

<Test>[xuanzhi]> delete from test.user where id=1;

Query OK, 1 row affected (0.00 sec)

<Test>[xuanzhi]>

这个时候发现自己删除错了，需要恢复，刚好这些数据不在最新的备份里，正常的恢复方法有两种：

一、是基于最新的完整备份+binlog进行数据恢复了，这时需要把备份导回去，还要找出Binlog DELETE前的pos位置，再进行binlog恢复，恢复完后再把记录恢复到误操的环境上。如果表很大，这时间要很久。
二、因为Binlog格式为ROW时，记录了行的修改，所以DELETE是可以看到所有列的值的，把binlog解析出来，找到被DELETE的记录，通过各种处理再恢复回去，但binlog不能基于一个库或表级别的解析，只能整个binlog解析再进行操作。

以上的方法都比较消耗时间，当然使用binlog-rollback.pl脚本有点类似第二种方法，但是binlog-rollback.pl可以指定库或表进行反向解析，还可以指定POS点，效率相当更高一些。

下面我们运行 binlog-rollback.pl 脚本，生成删除数据语句的反向SQL：

[root@localhost mysql3306]# perl binlog-rollback.pl -f 'localhost-bin.000023'  -o '/data/t.sql' -u 'recovery' -p ''

mysql: [Warning] Using a password on the command line interface can be insecure.

mysql: [Warning] Using a password on the command line interface can be insecure.

[root@localhost mysql3306]#

再次查看输出文件：

[root@localhost mysql3306]# cat /data/t.sql

INSERT INTO `test`.`user` SET `id`=, `name`='user1', `age`=;

INSERT INTO `xuanzhi`.`tb1` SET `id`=, `name`='bb';

DELETE FROM `test`.`user` WHERE `id`= AND `name`='user2' AND `age`=;

DELETE FROM `test`.`user` WHERE `id`= AND `name`='user1' AND `age`=;

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='bb';

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='aa';

[root@localhost mysql3306]#

刚刚DELETE的2条记录已经生成了反向INSERT语句，这样恢复就简单多啦：

INSERT INTO `test`.`user` SET `id`=, `name`='user1', `age`=;

INSERT INTO `xuanzhi`.`tb1` SET `id`=, `name`='bb';

下面我们模拟修改数据的时候，误修改了，如下：

<Test>[xuanzhi]> select * from xuanzhi.tb1;

+------+------+

| id   | name |

+------+------+

|    1 | aa   |

+------+------+

1 row in set (0.00 sec)

<Test>[xuanzhi]> select * from test.user;

+------+-------+------+

| id   | name  | age  |

+------+-------+------+

|    2 | user2 |   30 |

+------+-------+------+

1 row in set (0.00 sec)

<Test>[xuanzhi]> update  xuanzhi.tb1 set name = 'MySQL' where id=1;

Query OK, 1 row affected (0.00 sec)

Rows matched: 1  Changed: 1  Warnings: 0

<Test>[xuanzhi]> update test.user set age = 20 where id = 2;

Query OK, 1 row affected (0.01 sec)

Rows matched: 1  Changed: 1  Warnings: 0

<Test>[xuanzhi]>

这个时候发现修改错数据了，需要还原，同样可以使用脚本binlog-rollback.pl 进行对所在Binlog的DML生成反向的SQL，进行恢复：

[root@localhost mysql3306]# perl binlog-rollback.pl -f 'localhost-bin.000023'  -o '/data/t.sql' -u 'recovery' -p ''

mysql: [Warning] Using a password on the command line interface can be insecure.

mysql: [Warning] Using a password on the command line interface can be insecure.

[root@localhost mysql3306]#

再查看输出文件：

[root@localhost mysql3306]# cat /data/t.sql

UPDATE `test`.`user` SET `id`=, `name`='user2', `age`= WHERE `id`= AND `name`='user2' AND `age`=;

UPDATE `xuanzhi`.`tb1` SET `id`=, `name`='aa' WHERE `id`= AND `name`='MySQL';

INSERT INTO `test`.`user` SET `id`=, `name`='user1', `age`=;

INSERT INTO `xuanzhi`.`tb1` SET `id`=, `name`='bb';

DELETE FROM `test`.`user` WHERE `id`= AND `name`='user2' AND `age`=;

DELETE FROM `test`.`user` WHERE `id`= AND `name`='user1' AND `age`=;

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='bb';

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='aa';

[root@localhost mysql3306]#

可以看到生成了反向的UPDATE语句：

UPDATE `test`.`user` SET `id`=, `name`='user2', `age`= WHERE `id`= AND `name`='user2' AND `age`=;

UPDATE `xuanzhi`.`tb1` SET `id`=, `name`='aa' WHERE `id`= AND `name`='MySQL';

下面进行指定库的反向解析，参数为（-d）

[root@localhost mysql3306]# perl binlog-rollback.pl -f 'localhost-bin.000023'  -o '/data/t.sql' -u 'recovery' -p '' -d 'xuanzhi'

mysql: [Warning] Using a password on the command line interface can be insecure.

[root@localhost mysql3306]# cat /data/t.sql

UPDATE `xuanzhi`.`tb1` SET `id`=, `name`='aa' WHERE `id`= AND `name`='MySQL';

INSERT INTO `xuanzhi`.`tb1` SET `id`=, `name`='bb';

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='bb';

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='aa';

[root@localhost mysql3306]#

可以看到输入的文件只含xuanzhi库的所有DML的反向SQL。

下面进行指定库下某个表的反向解析，参数为：-T （为了看到效果在xuanzhi库下的tb2表删除一些记录）：

<Test>[xuanzhi]> select * from tb2;

+------+------+

| id   | name |

+------+------+

|    1 | aa   |

|    2 | bb   |

|    3 | cc   |

+------+------+

3 rows in set (0.04 sec)

<Test>[xuanzhi]> delete from xuanzhi.tb2 where id <2;

Query OK, 1 row affected (0.02 sec)

<Test>[xuanzhi]>

这个时候应该如果只指定xuanzhi库,那么tb1和tb2的DML操作的反向操作都会记录下来：

[root@localhost mysql3306]# perl binlog-rollback.pl -f 'localhost-bin.000023'  -o '/data/t.sql' -u 'recovery' -p '' -d 'xuanzhi'

mysql: [Warning] Using a password on the command line interface can be insecure.

mysql: [Warning] Using a password on the command line interface can be insecure.

[root@localhost mysql3306]# cat /data/t.sql

INSERT INTO `xuanzhi`.`tb2` SET `id`=, `name`='aa';

UPDATE `xuanzhi`.`tb1` SET `id`=, `name`='aa' WHERE `id`= AND `name`='MySQL';

INSERT INTO `xuanzhi`.`tb1` SET `id`=, `name`='bb';

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='bb';

DELETE FROM `xuanzhi`.`tb1` WHERE `id`= AND `name`='aa';

[root@localhost mysql3306]#

指定单个表tb2：

[root@localhost mysql3306]# perl binlog-rollback.pl -f 'localhost-bin.000023'  -o '/data/t.sql' -u 'recovery' -p '' -d 'xuanzhi' -T 'tb2'

mysql: [Warning] Using a password on the command line interface can be insecure.

[root@localhost mysql3306]# cat /data/t.sql

INSERT INTO `xuanzhi`.`tb2` SET `id`=, `name`='aa';

[root@localhost mysql3306]#

因为上面删除了一条tb2的数据，所有这个文件就对应生成一条tb2的INSERT记录

下面进行POS点生成反向SQL：（--start-position= --stop-position=）

# at

#  :: server id   end_log_pos  CRC32 0xb67ef6ba   Query   thread_id=    exec_time=     error_code=

SET TIMESTAMP=/*!*/;

BEGIN

/*!*/;

# at

#  :: server id   end_log_pos  CRC32 0x219a127c   Table_map: `test`.`user` mapped to number

# at

#  :: server id   end_log_pos  CRC32 0xf5e0d39e   Update_rows: table id  flags: STMT_END_F

BINLOG '

K+TdVhPqlBMAMwAAAJMGAAAAAEoAAAAAAAEABHRlc3QABHVzZXIAAwP+AwL+Hgd8Epoh

K+TdVh/qlBMAQgAAANUGAAAAAEoAAAAAAAEAAgAD///4AgAAAAV1c2VyMh4AAAD4AgAAAAV1c2Vy

MhQAAACe0+D1

'/*!*/;

### UPDATE `test`.`user`

### WHERE

###   @=

###   @='user2'

###   @=

### SET

###   @=

###   @='user2'

###   @=

# at

#  :: server id   end_log_pos  CRC32 0x1e62cb77   Xid =

COMMIT/*!*/;

# at

#  :: server id   end_log_pos  CRC32 0x04dfe1f0   Query   thread_id=    exec_time=     error_code=

SET TIMESTAMP=/*!*/;

BEGIN

/*!*/;

# at

#  :: server id   end_log_pos  CRC32 0x897ae6bd   Table_map: `xuanzhi`.`tb2` mapped to number

# at

#  :: server id   end_log_pos  CRC32 0xea61aff0   Delete_rows: table id  flags: STMT_END_F

BINLOG '

QOfdVhPqlBMANAAAAHMHAAAAAEYAAAAAAAEAB3h1YW56aGkAA3RiMgACA/4C/goDveZ6iQ==

QOfdViDqlBMAKwAAAJ4HAAAAAEYAAAAAAAEAAgAC//wBAAAAAmFh8K9h6g==

'/*!*/;

### DELETE FROM `xuanzhi`.`tb2`

### WHERE

###   @=

###   @='aa'

# at

#  :: server id   end_log_pos  CRC32 0x49e1ce9c   Xid =

COMMIT/*!*/;

从上面的binlog可以看到开始的--start-position=1557 结束的--stop-position=1981，这一段binlog里做了UPDATE `test`.`user` ... 和 DELETE FROM `xuanzhi`.`tb2` ... 的操作，那么用binlog-rollback.pl应该会生成一个UPDATE和一个INSERT语句

[root@localhost mysql3306]# perl binlog-rollback.pl -f 'localhost-bin.000023'  -o '/data/t.sql' -u 'recovery' -p '123456' --start-position=1557  --stop-position=1981

mysql: [Warning] Using a password on the command line interface can be insecure.

mysql: [Warning] Using a password on the command line interface can be insecure.

[root@localhost mysql3306]# cat /data/t.sql

INSERT INTO `xuanzhi`.`tb2` SET `id`=1, `name`='aa';

UPDATE `test`.`user` SET `id`=2, `name`='user2', `age`=30 WHERE `id`=2 AND `name`='user2' AND `age`=20;

[root@localhost mysql3306]#

更多的测试，就看同学们了，有测试不当的地方请告诉我，大家一起学习。

总结：一、感谢那些有分享精神的大神们，让我们学到了更多的东西，但开源的脚本需要多测试。

二、误操的情况，时有发生，所以我们要做好备份，做好一些数据恢复的测试。

三、该脚本在处理比较在的binlog时，会经常出现些小问题

作者：陆炫志

出处：xuanzhi的博客 http://www.cnblogs.com/xuanzhi201111

您的支持是对博主最大的鼓励，感谢您的认真阅读。本文版权归作者所有，欢迎转载，但请保留该声明。

MySQL基于ROW格式的数据恢复的更多相关文章

浅析MySQL基于ROW格式的二进制日志
上文分析的二进制日志实际上是基于STATEMENT格式的,下面我们来看看基于ROW格式的二进制日志,毕竟,两者对应的binlog事件类型也不一样,同时,很多童鞋反映基于ROW格式的二进制日志无法查到原 ...
my42_Mysql基于ROW格式的主从同步
模拟主从update事务,从库跳过部分update事务后,再次开始同步的现象主库 mysql> select * from dbamngdb.isNodeOK; +----+--------- ...
MySQL Binlog Mixed模式记录成Row格式
背景: 一个简单的主从结构,主的binlog format是Mixed模式,在执行一条简单的导入语句时,通过mysqlbinlog导出发现记录的Binlog全部变成了Row的格式(明明设置的是Mixe ...
my15_ mysql binlog格式从mixed修改为row格式
由于主库繁忙,就在从库上修改binlog格式 1. 从库切日志mysql> flush logs;Query OK, 0 rows affected (0.00 sec) mysql> f ...
基于GTID环境的数据恢复
下面说一个线上环境的数据恢复案例,线上环境误操作的事情,随时都有可能发生的,这时候运维DBA或者DBA会想,没人会傻到把数据库干掉的吧?又或者没有会闲得蛋痛删除了几条数据又想恢复这么无聊吧?适适这样的 ...
Mysql基于GTID主从复制
Mysql5.6基于GTID全局事务的复制什么是GTID? GTID(Global Transaction Identifiers)是全局事务标识当使用GTIDS时,在主上提交的每一个事务都会 ...
MySQL二进制日志格式对复制的影响
复制的分类基于SQL语句的复制 - SBR 主库二进制日志格式使用STATEMENT 在MySQL 5.1之前仅存在SBR模式, 又称之为逻辑复制. 主库记录CUD操作的SQL语句, 从库会读取并重 ...
MySQL基于binlog主从复制
MySQL复制介绍默认情况下复制是异步进行的,从库也不需要一直连接到主库来同步数据 MySQL复制的数据粒度可以是主实例上所有的数据库,也可以是指定的一个或多个数据库 ,也可以是一个数据库里的指定 ...
Mysql基于GTID复制模式-运维小结 (完整篇)
先来看mysql5.6主从同步操作时遇到的一个报错:mysql> change master to master_host='192.168.10.59',master_user='repli' ...

随机推荐

java.lang.RuntimeException: Cannot create a secure XMLInputFactory 异常处理
背景:本人在近日的webservice CXF接口的编写调试过程中,发现一个很奇怪的现象--我在本地对接口进行往返请求时完全没有问题,不管是main方法直接调用还是业务场景里调用都无问题,在其他同 ...
JS 如何将 HTML 页面导出为多页 PDF
参考链接:https://blog.csdn.net/pwc1996/article/details/70141383
[转载]Tensorflow 的reduce_sum()函数的axis，keep_dim这些参数到底是什么意思？
转载链接:https://www.zhihu.com/question/51325408/answer/125426642来源:知乎这个问题无外乎有三个难点: 什么是sum 什么是reduce 什么 ...
新年第一个目标一张表盘串讲所有canves的知识点
我们的目标首先是canves的坐标系统,基于浏览器的左上角为原点,x,y轴为正方向的坐标系统. 首先初始化,打标签 <canvas id="canvas" height=& ...
Linux查看本机IP：curl cip.cc
curl http://members.3322.org/dyndns/getip curl ip.6655.com/ip.aspx curl ifconfig.me curl icanhazip.c ...
基于React 的前端UI开发框架及与Electron 的结合 https://cxjs.io/
1.cxjs 基于React 的前端UI开发框架 https://cxjs.io/ coreu http://coreui.io/ 2.antd-admin ...
ASP.NET WEBAPI 使用Swagger生成API文档
一.安装新建一个没有身份验证的mvc项目 - SwaggerMvc5Demo,然后添加一个名为Remote(自定义)且包含基础读写(不想手写)的ApiController 开源地址:https: ...
ubuntu html5开发工具brackets
Brackets 是一款使用 HTML,CSS,JavaScript 创建的开源的针对 Web 开发的编辑器.实时预览,快速编辑,跨平台,可扩展,开源,让 Brackets 成为一款非常优秀的编辑器. ...
Razor视图基本语法
 @for (int i = 0; i < 10; i++) { <baobao>good</baobao> } < ...
java子类数组的引用转换成超类数组的引用
public class Person { } public class Student extends Person{ private String name; public Student(Str ...

MySQL基于ROW格式的数据恢复

MySQL基于ROW格式的数据恢复的更多相关文章

随机推荐

热门专题