小编给大家分享一下Mysql在ROW模式下如何通过binlog提取sql语句,希望大家阅读完这篇文章之后都有所收获,下面让我们一起去探讨吧! Linux基于row模式的binlo
小编给大家分享一下Mysql在ROW模式下如何通过binlog提取sql语句,希望大家阅读完这篇文章之后都有所收获,下面让我们一起去探讨吧!
Linux
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
mysql> select * from yoon;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 1 | HANK | YOON | 2006-02-15 04:34:33 |
| 2 | HANK | YOON | 2006-02-15 04:34:33 |
| 3 | HANK | YOON | 2006-02-15 04:34:33 |
| 4 | HANK | YOON | 2006-02-15 04:34:33 |
| 5 | HANK | YOON | 2006-02-15 04:34:33 |
| 6 | HANK | YOON | 2006-02-15 04:34:33 |
| 7 | HANK | YOON | 2006-02-15 04:34:33 |
| 8 | HANK | YOON | 2006-02-15 04:34:33 |
| 9 | HANK | YOON | 2006-02-15 04:34:33 |
| 10 | HANK | YOON | 2006-02-15 04:34:33 |
| 11 | HANK | YOON | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
11 rows in set (0.00 sec)
mysql> delete from yoon;
Query OK, 11 rows affected (1.03 sec)
mysql> select * from yoon;
Empty set (0.00 sec)
命令之间的空格一定要注意,否则就会无法提取SQL语句:
[root@hank-yoon data]# perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/export/data/mysql/data/yoon.sql' -u 'root' -p 'yoon'
Warning: Using a passWord on the command line interface can be insecure.
[root@hank-yoon data]# ls
auto.cnf hank ibdata2 ib_logfile1 modify.pl mysql-bin.000001 perfORMance_schema test yoon.sql
binlog-rollback.pl ibdata1 ib_logfile0 ib_logfile2 mysql mysql-bin.index sakila yoon
[root@hank-yoon data]# cat yoon.sql
INSERT INTO `yoon`.`yoon` SET `actor_id`=11, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=10, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=9, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=8, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=7, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=6, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=5, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=4, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=3, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=2, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
INSERT INTO `yoon`.`yoon` SET `actor_id`=1, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
mysql> INSERT INTO `yoon`.`yoon` SET `actor_id`=11, `first_name`='HANK', `last_name`='YOON', `last_update`=from_unixtime(1139949273);
Query OK, 1 row affected (0.01 sec)
mysql> select * from yoon;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 11 | HANK | YOON | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
点击(此处)折叠或打开
#!/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 1;
}
my %opt;
my $srcfile;
my $host = '127.0.0.1';
my $port = 3306;
my ($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}=1;
}
}
# 指定表
if ($opt{'T'}){
my @tbs = split(/,/,$opt{'T'});
foreach my $tb (@tbs){
$do_tbs{$tb}=1;
}
}
# 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 >0){
&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 $res = 0;
if ($ddlnum>0){
# 在ddl sql前面加上前缀<DDL>
$ddlcnt = `echo '$ddlcnt' | sed 's/^//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) = (0,0);
# 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)){
$notignore = 0;
&mdebug("#IGNORE#:line:".$line);
$sqlstr = '';
}else{
$notignore = 1;
&mdebug("#DO#:line:".$line);
}
}else {
if($notignore){
&merror("can't get tbname") unless (defined($tbname));
if ($line =~ /^WHERE/){
$sqlarea = $SQLAREA_WHERE;
$sqlstr .= qq{ SET};
$isareabegin = 1;
}elsif ($line =~ /^SET/){
$sqlarea = $SQLAREA_SET;
$sqlstr .= qq{ WHERE};
$isareabegin = 1;
}elsif ($line =~ /^\@/){
$sqlstr .= &deal_col_value($tbname, $sqltype, $sqlarea, $isareabegin, $line);
$isareabegin = 0;
}else{
&mdebug("::unknown sql:".$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[0],1);
my $valstartpos = length($pos)+2;
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 $res = 0;
# 指定了数据库
if ($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 = 1;
}
}
# 与指定库不同
}else{
$res = 1;
}
}
#&mdebug("Table check ignore:$fullname->$res");
return $res;
}
# ----------------------------------------------------------------------------------------
# Func : print debug msg
# ----------------------------------------------------------------------------------------
sub mdebug{
my (@msg) = @_;
print "@msg\n" if ($opt{'debug'});
}
# ----------------------------------------------------------------------------------------
# Func : print error msg and exit
# ----------------------------------------------------------------------------------------
sub merror{
my (@msg) = @_;
print ":@msg\n";
&print_usage();
exit(1);
}
# ----------------------------------------------------------------------------------------
# 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 3307
shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --start-position=107
shell> perl binlog-rollback.pl -f 'mysql-bin.000001' -o '/tmp/t.sql' -u 'user' -p 'pwd' --start-position=107 --stop-position=10000
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;
}
1;
看完了这篇文章,相信你对“MySQL在ROW模式下如何通过binlog提取SQL语句”有了一定的了解,如果想了解更多相关知识,欢迎关注编程网数据库频道,感谢各位的阅读!
--结束END--
本文标题: MySQL在ROW模式下如何通过binlog提取SQL语句
本文链接: https://lsjlt.com/news/62355.html(转载时请注明来源链接)
有问题或投稿请发送至: 邮箱/279061341@qq.com QQ/279061341
2024-10-23
2024-10-22
2024-10-22
2024-10-22
2024-10-22
2024-10-22
2024-10-22
2024-10-22
2024-10-22
2024-10-22
回答
回答
回答
回答
回答
回答
回答
回答
回答
回答
0