基于binlog来分析mysql的行记录修改情况

昨晚写的脚本，因为个人python能力有限，本来想这不发这文，后来想想，没准会有哪位园友给出优化建议。

1 实现内容

有些情况下，可能会统计在某个时间段内，MySQL修改了多少数据量？发生了多少事务？主要是哪些表格发生变动？变动的数量是怎么样的？ 但是却不需要行记录的修改内容，只需要了解 行数据的 变动情况。

这些情况部分可以通过监控来大致了解，但是也可以基于binlog来全盘分析，binlog的格式是row模式。

在写flashback的时候，顺带把这个也写了个脚步，使用python编写，都差不多原理，只是这个简单些，介于个人python弱的不行，性能可能还有很大的提升空间，也希望园友能协助优化下。

先贴python脚步的分析结果图如下，分为4个部分：事务耗时情况、事务影响行数情况、DML行数情况以及操作最频繁表格情况。

2 脚本简单描述

脚本依赖的模块中，pymysql需要自行安装。

创建类queryanalyse，其中有5个函数定义：_get_db、create_tab、rowrecord、binlogdesc跟closeconn。

2.1 _get_db

该函数用来解析输入参数值，参数值一共有7个，都是必须填写的。分别为host，user，password，port，table name for transaction，table name for records，对应的简写如下：

ALL options need to assign:

-h : host, the database host，which database will store the results after analysis

-u : user, the db user

-p : password, the db user's password

-P : port, the db port

-f : file path, the binlog file

-tr : table name for record , the table name to store the row record

-tt : table name for transaction, the table name to store transactions

比如，执行脚本：python queryanalyse.py -h=127.0.0.1 -P=3310 -u=root -p=password -f=/tmp/stock_binlog.log -tt=flashback.tbtran -tr=flashback.tbrow，该函数负责处理各个选项的参数值情况，并存储。

2.2 create_tab

创建两个表格，分别用来存储 binlog file文件的分析结果。一个用来存储事务的执行开始时间跟结束时间，由选项 -tt来赋值表名；一个是用来存储每一行记录的修改情况，由选项 -tr来赋值表名。

事务表记录内容：事务的开始时间及事务的结束时间。

行记录表的内容：库名，表名，DML类型以及事务对应事务表的编号。

root@localhost:mysql3310.sock 14:42:29 [flashback]>show create table tbrow \G*************************** 1. row ***************************
 Table: tbrowCreate Table: CREATE TABLE `tbrow` (
 `auto_id` int(10) unsigned NOT NULL AUTO_INCREMENT,
 `sqltype` int(11) NOT NULL COMMENT '1 is insert,2 is update,3 is delete',
 `tran_num` int(11) NOT NULL COMMENT 'the transaction number',
 `dbname` varchar(50) NOT NULL,
 `tbname` varchar(50) NOT NULL, PRIMARY KEY (`auto_id`), KEY `sqltype` (`sqltype`), KEY `dbname` (`dbname`), KEY `tbname` (`tbname`)
) ENGINE=InnoDB AUTO_INCREMENT=295151 DEFAULT CHARSET=utf81 row in set (0.00 sec)
 
root@localhost:mysql3310.sock 14:42:31 [flashback]>SHOW CREATE TABLE TBTRAN \G*************************** 1. row ***************************
 Table: TBTRANCreate Table: CREATE TABLE `tbtran` (
 `auto_id` int(10) unsigned NOT NULL AUTO_INCREMENT,
 `begin_time` datetime NOT NULL,
 `end_time` datetime NOT NULL, PRIMARY KEY (`auto_id`)
) ENGINE=InnoDB AUTO_INCREMENT=6390 DEFAULT CHARSET=utf81 row in set (0.00 sec)

2.3 rowrecord

重点函数，分析binlog文件内容。这里有几个规律：

1. 每个事务的结束点，是以 'Xid = ' 来查找

1. 事务的开始时间，是事务内的第一个 'Table_map' 行里边的时间

2. 事务的结束时间，是以 'Xid = '所在行的 里边的时间

2. 每个行数据是属于哪个表格，是以 'Table_map'来查找

3. DML的类型是按照 行记录开头的情况是否为：'### INSERT INTO' 、'### UPDATE' 、'### DELETE FROM'

4. 注意，单个事务可以包含多个表格多种DML多行数据修改的情况。

2.4 binlogdesc

描述分析结果，简单4个SQL分析。

1. 分析修改行数据的 事务耗时情况

2. 分析修改行数据的 事务影响行数情况

3. 分析DML分布情况

4. 分析 最多DML操作的表格 ，取前十个分析

2.5 closeconn

关闭数据库连接。

3 使用说明

首先，确保python安装了pymysql模块，把python脚本拷贝到文件 queryanalyse.py。

然后，把要分析的binlog文件先用 mysqlbinlog 指令分析存储，具体binlog的文件说明，可以查看之前的博文：关于binary log那些事——认真码了好长一篇。mysqlbinlog的指令使用方法，可以详细查看文档：https://dev.mysql.com/doc/refman/5.7/en/mysqlbinlog.html 。

比较常用通过指定开始时间跟结束时间来分析 binlog文件。

mysqlbinlog --start-datetime='2017-04-23 00:00:03' --stop-datetime='2017-04-23 00:30:00' --base64-output=decode-rows -v /data/mysql/logs/mysql-bin.007335 > /tmp/binlog_test.log

分析后，可以把这个 binlog_test.log文件拷贝到其他空闲服务器执行分析，只需要有个空闲的DB来存储分析记录即可。

假设这个时候，拷贝 binlog_test.log到测试服务器上，测试服务器上的数据库可以用来存储分析内容，则可以执行python脚本了，注意要进入到python脚本的目录中，或者指定python脚本路径。

python queryanalyse.py -h=127.0.0.1 -P=3310 -u=root -p=password -f= /tmp/binlog_test.log -tt=flashback.tbtran -tr=flashback.tbrow

没了，就等待输出吧。

性能是硬伤，在虚拟机上测试，大概500M的binlog文件需要分析2-3min，有待提高！

4 python脚本

 1 import pymysql 2 from pymysql.cursors import DictCursor 3 import re 4 import os 5 import sys 6 import datetime 7 import time 8 import logging 9 import importlib 10 importlib.reload(logging) 11 logging.basicConfig(level=logging.DEBUG,format='%(asctime)s %(levelname)s %(message)s ') 12 13 14 usage=''' usage: python [script's path] [option] 15 ALL options need to assign: 16 17 -h : host, the database host，which database will store the results after analysis 
 18 -u : user, the db user 19 -p : password, the db user's password 20 -P : port, the db port 21 -f : file path, the binlog file 22 -tr : table name for record , the table name to store the row record 23 -tt : table name for transaction, the table name to store transactions 24 Example: python queryanalyse.py -h=127.0.0.1 -P=3310 -u=root -p=password -f=/tmp/stock_binlog.log -tt=flashback.tbtran -tr=flashback.tbrow 25 26 ''' 27 28 class queryanalyse: 29 def init(self): 30 #初始化 31 self.host='' 32 self.user='' 33 self.password='' 34 self.port='3306' 35 self.fpath='' 36 self.tbrow='' 37 self.tbtran='' 38 39 self._get_db() 40 logging.info('assign values to parameters is done:host={},user={},password=***,port={},fpath={},tb_for_record={},tb_for_tran={}'.format(self.host,self.user,self.port,self.fpath,self.tbrow,self.tbtran)) 41 42 self.mysqlconn = pymysql.connect(host=self.host, user=self.user, password=self.password, port=self.port,charset='utf8') 43 self.cur = self.mysqlconn.cursor(cursor=DictCursor) 44 logging.info('MySQL which userd to store binlog event connection is ok') 45 46 self.begin_time='' 47 self.end_time='' 48 self.db_name='' 49 self.tb_name='' 50 51 def _get_db(self): 52 #解析用户输入的选项参数值，这里对password的处理是明文输入，可以自行处理成是input格式， 53 #由于可以拷贝binlog文件到非线上环境分析，所以password这块，没有特殊处理 54 logging.info('begin to assign values to parameters') 55 if len(sys.argv) == 1: 56 print(usage) 57 sys.exit(1) 58 elif sys.argv[1] == '--help': 59 print(usage) 60 sys.exit() 61 elif len(sys.argv) > 2: 62 for i in sys.argv[1:]: 63 _argv = i.split('=') 64 if _argv[0] == '-h': 65 self.host = _argv[1] 66 elif _argv[0] == '-u': 67 self.user = _argv[1] 68 elif _argv[0] == '-P': 69 self.port = int(_argv[1]) 70 elif _argv[0] == '-f': 71 self.fpath = _argv[1] 72 elif _argv[0] == '-tr': 73 self.tbrow = _argv[1] 74 elif _argv[0] == '-tt': 75 self.tbtran = _argv[1] 76 elif _argv[0] == '-p': 77 self.password = _argv[1] 78 else: 79 print(usage) 80 81 def create_tab(self): 82 #创建两个表格：一个用户存储事务情况，一个用户存储每一行数据修改的情况 83 #注意，一个事务可以存储多行数据修改的情况 84 logging.info('creating table ...') 85 create_tb_sql ='''CREATE TABLE IF NOT EXISTS {} ( 86 `auto_id` int(10) unsigned NOT NULL AUTO_INCREMENT, 87 `begin_time` datetime NOT NULL, 88 `end_time` datetime NOT NULL, 89 PRIMARY KEY (`auto_id`) 90 ) ENGINE=InnoDB DEFAULT CHARSET=utf8; 91 CREATE TABLE IF NOT EXISTS {} ( 92 `auto_id` int(10) unsigned NOT NULL AUTO_INCREMENT, 93 `sqltype` int(11) NOT NULL COMMENT '1 is insert,2 is update,3 is delete', 94 `tran_num` int(11) NOT NULL COMMENT 'the transaction number', 95 `dbname` varchar(50) NOT NULL, 96 `tbname` varchar(50) NOT NULL, 97 PRIMARY KEY (`auto_id`), 98 KEY `sqltype` (`sqltype`), 99 KEY `dbname` (`dbname`),100 KEY `tbname` (`tbname`)101 ) ENGINE=InnoDB DEFAULT CHARSET=utf8;102 truncate table {};103 truncate table {};104 '''.format(self.tbtran,self.tbrow,self.tbtran,self.tbrow)105 106 self.cur.execute(create_tb_sql)107 logging.info('created table {} and {}'.format(self.tbrow,self.tbtran))108 109 def rowrecord(self):110 #处理每一行binlog111 #事务的结束采用 'Xid =' 来划分112 #分析结果，按照一个事务为单位存储提交一次到db113 try:114 tran_num=1 #事务数115 record_sql='' #行记录的insert sql116 tran_sql='' #事务的insert sql117 118 self.create_tab()119 120 with open(self.fpath,'r') as binlog_file:121 logging.info('begining to analyze the binlog file ,this may be take a long time !!!')122 logging.info('analyzing...')123 124 for bline in binlog_file:125 126 if bline.find('Table_map:') != -1:127 l = bline.index('server')128 n = bline.index('Table_map')129 begin_time = bline[:l:].rstrip(' ').replace('#', '20')130 131 if record_sql=='':132 self.begin_time = begin_time[0:4] + '-' + begin_time[4:6] + '-' + begin_time[6:]133 134 self.db_name = bline[n::].split(' ')[1].replace('`', '').split('.')[0]135 self.tb_name = bline[n::].split(' ')[1].replace('`', '').split('.')[1]136 bline=''137 138 elif bline.startswith('### INSERT INTO'):139 record_sql=record_sql+"insert into {}(sqltype,tran_num,dbname,tbname) VALUES (1,{},'{}','{}');".format(self.tbrow,tran_num,self.db_name,self.tb_name)140 141 elif bline.startswith('### UPDATE'):142 record_sql=record_sql+"insert into {}(sqltype,tran_num,dbname,tbname) VALUES (2,{},'{}','{}');".format(self.tbrow,tran_num,self.db_name,self.tb_name)143 144 elif bline.startswith('### DELETE FROM'):145 record_sql=record_sql+"insert into {}(sqltype,tran_num,dbname,tbname) VALUES (3,{},'{}','{}');".format(self.tbrow,tran_num,self.db_name,self.tb_name)146 147 elif bline.find('Xid =') != -1:148 149 l = bline.index('server')150 end_time = bline[:l:].rstrip(' ').replace('#', '20')151 self.end_time = end_time[0:4] + '-' + end_time[4:6] + '-' + end_time[6:]152 tran_sql=record_sql+"insert into {}(begin_time,end_time) VALUES ('{}','{}')".format(self.tbtran,self.begin_time,self.end_time)153 154 self.cur.execute(tran_sql)155 self.mysqlconn.commit()156 record_sql = ''157 tran_num += 1158 159 except Exception:160 return 'funtion rowrecord error'161 162 def binlogdesc(self):163 sql=''164 t_num=0165 r_num=0166 logging.info('Analysed result printing...\n')167 #分析总的事务数跟行修改数量168 sql="select 'tbtran' name,count(*) nums from {} union all select 'tbrow' name,count(*) nums from {};".format(self.tbtran,self.tbrow)169 self.cur.execute(sql)170 rows=self.cur.fetchall()171 for row in rows:172 if row['name']=='tbtran':173 t_num = row['nums']174 else:175 r_num = row['nums']176 print('This binlog file has {} transactions, {} rows are changed '.format(t_num,r_num))177 178 # 计算 最耗时 的单个事务179 # 分析每个事务的耗时情况,分为5个时间段来描述180 # 这里正常应该是 以毫秒来分析的，但是binlog中，只精确时间到second181 sql='''select 
182 count(case when cost_sec between 0 and 1 then 1 end ) cos_1,183 count(case when cost_sec between 1.1 and 5 then 1 end ) cos_5,184 count(case when cost_sec between 5.1 and 10 then 1 end ) cos_10,185 count(case when cost_sec between 10.1 and 30 then 1 end ) cos_30,186 count(case when cost_sec >30.1 then 1 end ) cos_more,187 max(cost_sec) cos_max188 from 
189 (190 select 
191 auto_id,timestampdiff(second,begin_time,end_time) cost_sec192 from {}193 ) a;'''.format(self.tbtran)194 self.cur.execute(sql)195 rows=self.cur.fetchall()196 197 for row in rows:198 print('The most cost time : {} '.format(row['cos_max']))199 print('The distribution map of each transaction costed time: ')200 print('Cost time between 0 and 1 second : {} , {}%'.format(row['cos_1'],int(row['cos_1']*100/t_num)))201 print('Cost time between 1.1 and 5 second : {} , {}%'.format(row['cos_5'], int(row['cos_5'] * 100 / t_num)))202 print('Cost time between 5.1 and 10 second : {} , {}%'.format(row['cos_10'], int(row['cos_10'] * 100 / t_num)))203 print('Cost time between 10.1 and 30 second : {} , {}%'.format(row['cos_30'], int(row['cos_30'] * 100 / t_num)))204 print('Cost time > 30.1 : {} , {}%\n'.format(row['cos_more'], int(row['cos_more'] * 100 / t_num)))205 206 # 计算 单个事务影响行数最多 的行数量207 # 分析每个事务 影响行数 情况,分为5个梯度来描述208 sql='''select 
209 count(case when nums between 0 and 10 then 1 end ) row_1,210 count(case when nums between 11 and 100 then 1 end ) row_2,211 count(case when nums between 101 and 1000 then 1 end ) row_3,212 count(case when nums between 1001 and 10000 then 1 end ) row_4,213 count(case when nums >10001 then 1 end ) row_5,214 max(nums) row_max215 from 
216 (217 select 
218 count(*) nums219 from {} group by tran_num220 ) a;'''.format(self.tbrow)221 self.cur.execute(sql)222 rows=self.cur.fetchall()223 224 for row in rows:225 print('The most changed rows for each row: {} '.format(row['row_max']))226 print('The distribution map of each transaction changed rows : ')227 print('Changed rows between 1 and 10 second : {} , {}%'.format(row['row_1'],int(row['row_1']*100/t_num)))228 print('Changed rows between 11 and 100 second : {} , {}%'.format(row['row_2'], int(row['row_2'] * 100 / t_num)))229 print('Changed rows between 101 and 1000 second : {} , {}%'.format(row['row_3'], int(row['row_3'] * 100 / t_num)))230 print('Changed rows between 1001 and 10000 second : {} , {}%'.format(row['row_4'], int(row['row_4'] * 100 / t_num)))231 print('Changed rows > 10001 : {} , {}%\n'.format(row['row_5'], int(row['row_5'] * 100 / t_num)))232 233 # 分析 各个行数 DML的类型情况234 # 描述 delete，insert，update的分布情况235 sql='select sqltype ,count(*) nums from {} group by sqltype ;'.format(self.tbrow)236 self.cur.execute(sql)237 rows=self.cur.fetchall()238 239 print('The distribution map of the {} changed rows : '.format(r_num))240 for row in rows:241 242 if row['sqltype']==1:243 print('INSERT rows :{} , {}% '.format(row['nums'],int(row['nums']*100/r_num)))244 if row['sqltype']==2:245 print('UPDATE rows :{} , {}% '.format(row['nums'],int(row['nums']*100/r_num)))246 if row['sqltype']==3:247 print('DELETE rows :{} , {}%\n '.format(row['nums'],int(row['nums']*100/r_num)))248 249 # 描述 影响行数 最多的表格250 # 可以分析是哪些表格频繁操作，这里显示前10个table name251 sql = '''select 
252 dbname,tbname ,253 count(*) ALL_rows,254 count(*)*100/{} per,255 count(case when sqltype=1 then 1 end) INSERT_rows,256 count(case when sqltype=2 then 1 end) UPDATE_rows,257 count(case when sqltype=3 then 1 end) DELETE_rows258 from {} 
259 group by dbname,tbname 
260 order by ALL_rows desc 
261 limit 10;'''.format(r_num,self.tbrow)262 self.cur.execute(sql)263 rows = self.cur.fetchall()264 265 print('The distribution map of the {} changed rows : '.format(r_num))266 print('tablename'.ljust(50),267 '|','changed_rows'.center(15),268 '|','percent'.center(10),269 '|','insert_rows'.center(18),270 '|','update_rows'.center(18),271 '|','delete_rows'.center(18)272 )273 print('-------------------------------------------------------------------------------------------------------------------------------------------------')274 for row in rows:275 print((row['dbname']+'.'+row['tbname']).ljust(50),276 '|',str(row['ALL_rows']).rjust(15),277 '|',(str(int(row['per']))+'%').rjust(10),278 '|',str(row['INSERT_rows']).rjust(10)+' , '+(str(int(row['INSERT_rows']*100/row['ALL_rows']))+'%').ljust(5),279 '|',str(row['UPDATE_rows']).rjust(10)+' , '+(str(int(row['UPDATE_rows']*100/row['ALL_rows']))+'%').ljust(5),280 '|',str(row['DELETE_rows']).rjust(10)+' , '+(str(int(row['DELETE_rows']*100/row['ALL_rows']))+'%').ljust(5),281 )282 print('\n')283 284 logging.info('Finished to analyse the binlog file !!!')285 286 def closeconn(self):287 self.cur.close()288 logging.info('release db connections\n')289 290 def main():291 p = queryanalyse()292 p.rowrecord()293 p.binlogdesc()294 p.closeconn()295 296 if name == "main":297 main()