【MySQL】MySQL回滚工具

1、mysqlbinlog把事务从binlog中导出

2、从导出的binlog中找到要回滚的事务，去掉第一个DML语句前和最后一个DML语句后与DML无关的binlog信息

3、在目录中新建一个table.cnf，把表结构以@1=columns这样的顺序一行写一列

4、update回滚支持选择条件列和回滚的数据列，把回滚时不需要的条件（列）写到not_used.set和not_used.where中

例如：

文件 table.cnf

@1=id
@2=column_a
@3=column_b
@4=time

文件not_used.set
##写到这个文件里面的是update回滚时不需要更新的列
##例如假设回滚不恢复 id 列，文件中应该如下

@1=

文件not_used.where
##写到这个文件里面的是update回滚时条件忽略的列
##例如假设回滚时不需要列 time 和 column_b 作为回滚条件，文件中应该如下，顺序不敏感

@=3
@=4

文件not_used.values
##写到这个文件里面的是delete回滚时不自动插入的列，例如自增列或者TIMESTAMP
##例如假设回滚时不需要列 time 和 id 作为回滚条件，文件中应该如下，顺序不敏感

@4=
@1=

有的表列比较多，写个脚本自己拼配置文件

mysql里面show create table，把结果写到table.txt

#!/bin/bash
awk '{print $1}' ./table.txt >./table.ini
n=`wc -l ./table.ini`
i=1
cat ./table.ini | while read columns_name
do
        echo ""@"$i"="$columns_name" >> ./table.cnf
        i=$[$i+1]
done
rm -rf ./table.txt ./table.ini

然后not_used.where、not_used.set、not_used.values也可以用table.cnf转换一下编辑

awk  -F '`' '{print $1}' table.cnf  > ./not_used.set

脚本：表名自己写吧

#!/bin/bash
table_name="$2"

### DELETE DML 2 rows in binlog
delete=2

### UPDATE DML 3 rows in binlog
update=3

### How many columns for this rollback table
table_columns=`wc -l ./table.cnf | awk '{print $1}'`

### Format binlog
echo -e "33[47;30m wait for change binlog format 33[0m"
#cat ./mysql-bin.txt | awk '{$1="";print>"./bin.log"}' 
echo -e "33[47;30m change binlog format OK  33[0m"

### Count for DML
dml_delete_count=`cat ./bin.log | grep DELETE | wc -l `
dml_update_count=`cat ./bin.log | grep UPDATE | wc -l `
echo -e "33[47;30m dml_delete_count $dml_delete_count 33[0m"
echo -e "33[47;30m dml_update_count $dml_update_count 33[0m"

### How many rows for one DML
dml_delete_row=`echo |awk '{print "'$delete'"+"'$table_columns'"}'`
dml_update_row=`echo |awk '{print "'$update'"+"'$table_columns'"+"'$table_columns'"}'`
dml_update_where_row_begin=3
dml_update_where_row_finish=`echo |awk '{print 2+"'$table_columns'"}'`
dml_update_set_row_begin=`echo |awk '{print 4+"'$table_columns'"}'`
dml_update_set_row_finish=$dml_update_row
echo -e "33[47;30m dml_delete_row $dml_delete_row 33[0m"
echo -e "33[47;30m dml_update_row $dml_update_row 33[0m"

fun_delete()
{
b=''
for((i=1;i<=${dml_delete_count};i++))
do
        sed -n '1,'$dml_delete_row'p' ./bin.log  > ./bin.tmp
        sed -i '1,'$delete'd' ./bin.tmp
        cat ./not_used.values | while read columns_values
        do
                sed -i '/'$columns_values'/d' ./bin.tmp
        done 
        data=`awk -F '=' '{$1="";print}' ./bin.tmp | awk '{print $1}' | tr "
" "," | sed 's/,$//' `
        
        cp ./table.cnf ./dml_columns.tmp
        cat ./not_used.values | while read columns_values
        do
                sed -i '/'$columns_values'/d' ./dml_columns.tmp
        done
        dml_columns=`awk -F '=' '{print $2}' ./dml_columns.tmp | tr "
" "," | sed 's/,$//'`

        echo "insert into $table_name($dml_columns) values ($data);" >> ./rollback.sql 
        sed -i '1,'$dml_delete_row'd' ./bin.log
        rm -rf ./bin.tmp ./sql.tmp
        h=`echo | awk '{print int("'$i'"/"'$dml_delete_count'"*"100%")}'`
        printf "progress:[$h%%]
"


done
rm -rf ./bin.log
echo -e "
"
echo done
}


fun_update()
{
if [ $dml_update_count -lt 5000 ]
then 
        file=1
else
        file=1000
fi

file_count=$[${dml_update_count}/${file}]
file_mod=$[${dml_update_count}%${file}]
file_dml_pos_begin=1
file_dml_pos_finish=$[${file_count}*${dml_update_row}]

for((f=1;f<=$[${file}+1];f++))
do
        sed -n ''$file_dml_pos_begin','$file_dml_pos_finish'p' ./bin.log > ./bin.log.$f

        rows_no_update_begin=1
        rows_no_update_finish=$dml_update_row

        for((i=1;i<=${dml_update_count};i++))
        do
                sed -n ''$rows_no_update_begin','$rows_no_update_finish'p' ./bin.log.$f  > ./bin.tmp
                sed -n ''$dml_update_set_row_begin','$dml_update_set_row_finish'p' ./bin.tmp > ./bin.where
                sed -n ''$dml_update_where_row_begin','$dml_update_where_row_finish'p' ./bin.tmp > ./bin.set
                ### data have been set,and this data make to search for new data in rollback SQL,choose columns
                cat ./not_used.where | while read columns_where
                do
                        sed -i '/'$columns_where'/d' ./bin.where
                done
                dml_where=`awk '{print $1}' ./bin.where | tr "
" "," | sed 's/,$//'`


                ### data will be update,all columns or part of them
                cat "./not_used.set" | while read columns_set
                do
                        sed -i '/'$columns_set'/d' ./bin.set
                done
                dml_set=`awk '{print $1}' ./bin.set | tr "
" "," | sed 's/,$//'`

                echo "update $table_name set $dml_set where $dml_where;" >> ./rollback.sql


                # delete big bin.log too slow
                # sed -i '1,'$dml_update_row'd' ./bin.log

                rows_no_update_begin=$[$[${dml_update_row}*${i}]+1]
                rows_no_update_finish=$[${dml_update_row}*$[${i}+1]]

                # change columns'name
                cat ./table.cnf | while read t_tmp
                do
                        t_1="`echo $t_tmp | awk -F '=' '{print $1}'`="
                        t_2="`echo $t_tmp | awk -F '=' '{print $2}'`="
                sed -i 's/'$t_1'/'$t_2'/g' ./rollback.sql
                done

        done
        

        file_dml_pos_begin=$[$[${file_count}*${f}*${dml_update_row}]+1]
        file_dml_pos_finish=$[${file_count}*$[${f}+1]*${dml_update_row}]


        rm -rf ./bin.log.$f

        h=`echo | awk '{print int("'$f'"/"'$dml_update_count'"*"100%")}'`
        printf "progress:[$h%%]
"
        echo -e "
"
done

echo done
}


case $1 in
        delete)echo -e "33[47;32m begin fun_delete 33[0m";sleep 2;fun_delete
        ;;
        update)echo -e "33[47;32m begin fun_update 33[0m";sleep 2;fun_update
        ;;
        *)echo -e "33[47;31m err input,please choose delete or update,quit 33[0m";exit 1
esac