mysql慢查询分析工具比较与实战

  • 2020 年 4 月 10 日
  • 笔记

00 前言

在进行mysql性能优化的时候,第一个想到的便是查看慢sql。

但是对于慢sql有没有什么好的工具进行分析呢?

推荐两个工具mysqldumpslow及pt-query-digest。

mysqlslowdump较为简单,常用命令:

#得到返回记录最多的20个sql  mysqldumpslow -s r -t 20 slowSQl.log    # 得到平均访问次数最多的20条sql  mysqldumpslow -s ar -t 20 slowSQl.log  

如果linux上没有安装mysqldumpslow,yum install安装下就行了。

本文主要说下pt-query-digest。

pt-query-digest可以非常清晰地将slowSQL分析出来,类似oracle的AWR报告。

    # Rank Query ID                        Response time   Calls R/Call V/M  # ==== =============================== =============== ===== ====== =====  #    1 0xABD1DCCCCD5AA5128E10C27B34... 1246.6948 41.7%   283 4.4053  0.04 UPDATE ziweidashi_deviceinfo  #    2 0x6914B81AAD1785E50708ABD113...  877.6900 29.3%   339 2.5891  0.09 SELECT birthDay_notify  #    3 0x44D9474C6D5C58DD07B5FEEA0D...  299.4193 10.0%    71 4.2172  0.05 SELECT tmall_product_orders  #    4 0xA9BE84CBE3DAA9B1CDD9B5A9EC...  127.0137  4.2%    46 2.7612  0.04 SELECT daily_user_action_log  #    5 0xCF0E12117C971C3013142E3717...  118.3138  4.0%    49 2.4146  0.05 SELECT tmall_user_take_coupon_record  #    6 0x94263184D24186330B13193534...   97.0805  3.2%    35 2.7737  0.56 SELECT tgg_users  #    7 0xC51165F1287A2ECDA221AC1F54...   52.5870  1.8%    22 2.3903  0.04 SELECT util_user_task_log  #    8 0xB8004D6D8A7A7967E04CD81E26...   43.7895  1.5%    16 2.7368  0.08 SELECT daily_user_action_log  #    9 0x910E19224F33DAA6391927B8E8...   41.3720  1.4%    15 2.7581  1.17 SELECT qifugong_tianbi_record  # MISC 0xMISC                            86.7871  2.9%    30 2.8929   0.0 <12 ITEMS>  

并且不只可以分析慢SQL日志,还可以分析binlog、general log。

此外,pt-query-digest是percona-toolkit工具包的其中一个工具。
这个工具包下还有很多实用的性能分析辅助工具。

01 安装排坑

1、下载

# 进入安装目录  cd /usr/local/src    # 下载percona-toolkit 工具包  wget percona.com/get/percona-toolkit.tar.gz    # 解压  tar zxf percona-toolkit.tar.gz    # 进入解压文件夹  cd /usr/local/src/percona-toolkit-3.1.0    # 安装perl模块,制定依赖路径  perl Makefile.PL PREFIX=/usr/local/percona-toolkit  

2、报错 prerequisite DBD::mysql 3 not found

报错如下,找不到DBD包

[root@iZ2zebthf35ejlps5v87ksZ percona-toolkit-3.1.0]# perl Makefile.PL PREFIX=/usr/local/percona-toolkit  Checking if your kit is complete...  Looks good  Warning: prerequisite DBD::mysql 3 not found.  Warning: prerequisite DBI 1.46 not found.  Writing Makefile for percona-toolkit  

百度问题,找到链接,https://blog.csdn.net/heizistudio/article/details/45724707?locationNum=8&fps=1

安装依赖包

yum install perl-DBD-MySQL  

然后重新执行命令

[root@iZ2zebthf35ejlps5v87ksZ percona-toolkit-3.1.0]# perl Makefile.PL PREFIX=/usr/local/percona-toolkit  Writing Makefile for percona-toolkit  

3、安装

make && make install  

安装后内容如下

……  Installing /usr/local/percona-toolkit/bin/pt-summary  Installing /usr/local/percona-toolkit/bin/pt-table-sync  Appending installation info to /usr/local/percona-toolkit/lib64/perl5/perllocal.pod  

4、使用

[root@iZ2zebthf35ejlps5v87ksZ bin]# pt-query-digest /usr/local/mysql/data/slow.log  -bash: pt-query-digest: command not found  

发现没找到pt-query-digest命令,是因为bash命令默认是从/usr/bin下找的;

如果rpm安装,会默认添加到/usr/bin下;

而我们现在是编译二进制安装到,并且默认是装到了/usr/local/percona-toolkit下,发现本文件夹下有个bin目录,pt工具都在其下。

-rwxrwxr-x 1 hc hc   41747 Sep 16  2019 pt-align  -rwxrwxr-x 1 hc hc  270675 Sep 16  2019 pt-archiver  -rwxrwxr-x 1 hc hc  170783 Sep 16  2019 pt-config-diff  -rwxrwxr-x 1 hc hc  167978 Sep 16  2019 pt-deadlock-logger  -rwxrwxr-x 1 hc hc  166450 Sep 16  2019 pt-diskstats  -rwxrwxr-x 1 hc hc  171099 Sep 16  2019 pt-duplicate-key-checker  -rwxrwxr-x 1 hc hc   50157 Sep 16  2019 pt-fifo-split  -rwxrwxr-x 1 hc hc  151809 Sep 16  2019 pt-find  -rwxrwxr-x 1 hc hc   67304 Sep 16  2019 pt-fingerprint  -rwxrwxr-x 1 hc hc  134955 Sep 16  2019 pt-fk-error-logger  -rwxrwxr-x 1 hc hc  223887 Sep 16  2019 pt-heartbeat  -rwxrwxr-x 1 hc hc  228213 Sep 16  2019 pt-index-usage  -rwxrwxr-x 1 hc hc   32405 Sep 16  2019 pt-ioprofile  -rwxrwxr-x 1 hc hc  256092 Sep 16  2019 pt-kill  -rwxrwxr-x 1 hc hc   21913 Sep 16  2019 pt-mext  -rwxrwxr-x 1 hc hc 8196032 Sep 16  2019 pt-mongodb-query-digest  -rwxrwxr-x 1 hc hc 8522944 Sep 16  2019 pt-mongodb-summary  -rwxrwxr-x 1 hc hc  108113 Sep 16  2019 pt-mysql-summary  -rwxrwxr-x 1 hc hc  426996 Sep 16  2019 pt-online-schema-change  -rwxrwxr-x 1 hc hc 4794784 Sep 16  2019 pt-pg-summary  -rwxrwxr-x 1 hc hc   24598 Sep 16  2019 pt-pmp  -rwxrwxr-x 1 hc hc  527607 Sep 16  2019 pt-query-digest  -rwxrwxr-x 1 hc hc 3624992 Sep 16  2019 pt-secure-collect  -rwxrwxr-x 1 hc hc   78242 Sep 16  2019 pt-show-grants  -rwxrwxr-x 1 hc hc   37784 Sep 16  2019 pt-sift  -rwxrwxr-x 1 hc hc  146952 Sep 16  2019 pt-slave-delay  -rwxrwxr-x 1 hc hc  131404 Sep 16  2019 pt-slave-find  -rwxrwxr-x 1 hc hc  184944 Sep 16  2019 pt-slave-restart  -rwxrwxr-x 1 hc hc   76226 Sep 16  2019 pt-stalk  -rwxrwxr-x 1 hc hc   90816 Sep 16  2019 pt-summary  -rwxrwxr-x 1 hc hc  459729 Sep 16  2019 pt-table-checksum  -rwxrwxr-x 1 hc hc  405119 Sep 16  2019 pt-table-sync  -rwxrwxr-x 1 hc hc  247743 Sep 16  2019 pt-table-usage  -rwxrwxr-x 1 hc hc  333011 Sep 16  2019 pt-upgrade  -rwxrwxr-x 1 hc hc  178415 Sep 16  2019 pt-variable-advisor  -rwxrwxr-x 1 hc hc  102545 Sep 16  2019 pt-visual-explain  

本次使用到主力工具,pt-query-digest,执行命令,进行慢日志分析

./pt-query-digest /usr/local/mysql/data/slow.log  

5、又报错 Can’t locate Digest/MD5.pm in @INC

[root@iZ2zebthf35ejlps5v87ksZ bin]# ./pt-query-digest /usr/local/mysql/data/slow.log  Can't locate Digest/MD5.pm in @INC (@INC contains: /usr/local/lib64/perl5 /usr/local/share/perl5 /usr/lib64/perl5/vendor_perl /usr/share/perl5/vendor_perl /usr/lib64/perl5 /usr/share/perl5 .) at ./pt-query-digest line 2470.  BEGIN failed--compilation aborted at ./pt-query-digest line 2470.  

安装perl-Digest-MD5工具

yum -y install perl-Digest-MD5  

终于可以运行了

./pt-query-digest /usr/local/mysql/data/slow.log  

6、无脑命令如下

yum -y install perl-DBD-MySQL  yum -y install perl-Digest-MD5    cd /usr/local/src  wget percona.com/get/percona-toolkit.tar.gz  tar zxf percona-toolkit.tar.gz  cd /usr/local/src/percona-toolkit-3.1.0  perl Makefile.PL PREFIX=/usr/local/percona-toolkit  make && make install  

二、分析实战

1、执行工具pt-query-digest

./pt-query-digest /usr/local/src/slowsqlExample/slow0312.log  

2、结果分析

找了一个慢sql,分析结果如下

[root@iZ2zebthf35ejlps5v87ksZ bin]# ./pt-query-digest /usr/local/src/slowsqlExample/slow0312.log    第一部分  该工具执行日志分析的用户时间,系统时间,物理内存占用大小,虚拟内存占用大小  # 360ms user time, 10ms system time, 22.56M rss, 187.09M vsz  工具执行时间  # Current date: Fri Mar 20 22:54:14 2020  运行分析工具的主机名  # Hostname: iZ2zebthf35ejlps5v87ksZ  被分析的文件名  # Files: /usr/local/src/slowsqlExample/slow0312.log  语句总数量,唯一的语句数量,QPS,并发数  # Overall: 906 total, 21 unique, 0.02 QPS, 0.07x concurrency _____________  日志记录的时间范围  # Time range: 2020-03-11 12:22:13 to 2020-03-12 00:16:57  # Attribute          total     min     max     avg     95%  stddev  median  # ============     ======= ======= ======= ======= ======= ======= =======  语句执行时间  # Exec time          2991s      2s     10s      3s      5s      1s      3s  锁占用时间  # Lock time          552ms    24us   371ms   609us   103us    12ms    57us  发送到客户端的行数  # Rows sent        167.53k       0  17.99k  189.35  487.09   1.22k       0  select语句扫描行数  # Rows examine     980.73M     238   1.96M   1.08M   1.95M 757.80k 753.18k  查询的字符数  # Query size       258.71k      17   1.77k  292.41  463.90  202.02  329.68      第二部分  # Profile  Rank:所有语句的排名,默认按查询时间降序排列,通过--order-by指定  Query ID:语句的ID,(去掉多余空格和文本字符,计算hash值)  Response:总的响应时间  time:该查询在本次分析中总的时间占比  calls:执行次数,即本次分析总共有多少条这种类型的查询语句  R/Call:平均每次执行的响应时间  V/M:响应时间Variance-to-mean的比率  Item:查询对象  # Rank Query ID                        Response time   Calls R/Call V/M  # ==== =============================== =============== ===== ====== =====  #    1 0xABD1DCCCCD5AA5128E10C27B34... 1246.6948 41.7%   283 4.4053  0.04 UPDATE ziweidashi_deviceinfo  #    2 0x6914B81AAD1785E50708ABD113...  877.6900 29.3%   339 2.5891  0.09 SELECT birthDay_notify  #    3 0x44D9474C6D5C58DD07B5FEEA0D...  299.4193 10.0%    71 4.2172  0.05 SELECT tmall_product_orders  #    4 0xA9BE84CBE3DAA9B1CDD9B5A9EC...  127.0137  4.2%    46 2.7612  0.04 SELECT daily_user_action_log  #    5 0xCF0E12117C971C3013142E3717...  118.3138  4.0%    49 2.4146  0.05 SELECT tmall_user_take_coupon_record  #    6 0x94263184D24186330B13193534...   97.0805  3.2%    35 2.7737  0.56 SELECT tgg_users  #    7 0xC51165F1287A2ECDA221AC1F54...   52.5870  1.8%    22 2.3903  0.04 SELECT util_user_task_log  #    8 0xB8004D6D8A7A7967E04CD81E26...   43.7895  1.5%    16 2.7368  0.08 SELECT daily_user_action_log  #    9 0x910E19224F33DAA6391927B8E8...   41.3720  1.4%    15 2.7581  1.17 SELECT qifugong_tianbi_record  # MISC 0xMISC                            86.7871  2.9%    30 2.8929   0.0 <12 ITEMS>      第三及后续部分,第一条查询语句 query id:0xABD1DCCCCD5AA5128E10C27B34BC04E7  # Query 1: 0.01 QPS, 0.03x concurrency, ID 0xABD1DCCCCD5AA5128E10C27B34BC04E7 at byte 355748  # Scores: V/M = 0.04  # Time range: 2020-03-11 12:24:03 to 2020-03-12 00:16:13  # Attribute    pct   total     min     max     avg     95%  stddev  median  # ============ === ======= ======= ======= ======= ======= ======= =======  # Count         31     283  # Exec time     41   1247s      4s      8s      4s      5s   437ms      4s  # Lock time     69   386ms    24us   371ms     1ms    93us    21ms    44us  # Rows sent      0       0       0       0       0       0       0       0  # Rows examine  18 180.00M 651.14k 651.45k 651.29k 650.62k       0 650.62k  # Query size    10  27.64k     100     100     100     100       0     100  # String:  数据库名  # Databases    taxen_ziweidashi  执行主机  # Hosts        118.190.93.166  执行用户  # Users        devAccount  查询时间占比  # Query_time distribution  #   1us  #  10us  # 100us  #   1ms  #  10ms  # 100ms  #    1s  ################################################################  #  10s+  # Tables  #    SHOW TABLE STATUS FROM `taxen_ziweidashi` LIKE 'ziweidashi_deviceinfo'G  #    SHOW CREATE TABLE `taxen_ziweidashi`.`ziweidashi_deviceinfo`G  UPDATE ziweidashi_deviceinfo           SET expired = 1          WHERE createTime   <=   1583942580685G  # Converted for EXPLAIN  # EXPLAIN /*!50100 PARTITIONS*/  select  expired = 1 from ziweidashi_deviceinfo where  createTime   <=   1583942580685G    # Query 2: 0.03 QPS, 0.07x concurrency, ID 0x6914B81AAD1785E50708ABD11319E02E at byte 13829  # Scores: V/M = 0.09  # Time range: 2020-03-11 12:22:13 to 16:05:47  # Attribute    pct   total     min     max     avg     95%  stddev  median  # ============ === ======= ======= ======= ======= ======= ======= =======  # Count         37     339  # Exec time     29    878s      2s      4s      3s      4s   472ms      2s  # Lock time      5    29ms    31us     4ms    86us    98us   229us    66us  # Rows sent      0      24       0       2    0.07       0    0.32       0  # Rows examine  67 665.20M   1.96M   1.96M   1.96M   1.96M       0   1.96M  # Query size    59 154.47k     462     467  466.60  463.90    2.07  463.90  # String:  # Hosts        10.66.186.115  # Users        root  # Query_time distribution  #   1us  #  10us  # 100us  #   1ms  #  10ms  # 100ms  #    1s  ################################################################  #  10s+  # Tables  #    SHOW TABLE STATUS LIKE 'birthDay_notify'G  #    SHOW CREATE TABLE `birthDay_notify`G  # EXPLAIN /*!50100 PARTITIONS*/  select birthdayno0_.id as id1_1_, birthdayno0_.index_card_show_date as index_ca2_1_, birthdayno0_.userId as userId3_1_, birthdayno0_.push_content as push_con4_1_, birthdayno0_.card_content as card_con5_1_, birthdayno0_.birthday_userId as birthday6_1_, birthdayno0_.birthday_contactId as birthday7_1_, birthdayno0_.need_push as need_pus8_1_ from birthDay_notify birthdayno0_ where birthdayno0_.userId=1304747 and birthdayno0_.index_card_show_date='2020-03-11 00:00:00'G    ……省略  

3、实例优化

找出这几条语句,对症下药,进行写法的修改、索引的设计,基本可以解决慢SQL问题。

例如query1的语句

    UPDATE ziweidashi_deviceinfo           SET expired = 1          WHERE createTime   <=   1583942580685  

分析后发现,这张表大部分数据的expired字段都是1,每次update都相当于全表查询、锁定了一次。

从逻辑上分析,是expired不等于1才修改的。

可以修改为

UPDATE ziweidashi_deviceinfo           SET expired = 1          WHERE createTime   <=   1583942580685          and expired != 1  

直接从平均的5秒执行时间降低到了0.04秒。

其他语句类似。

三、常用命令

1.分析慢查询文件

pt-query-digest  slow.log > slow_report.log  

2.分析最近12小时内的查询

pt-query-digest  --since=12h  slow.log > slow_report2.log  

3.分析指定时间范围内的查询

pt-query-digest slow.log --since '2017-01-07 09:30:00' --until '2017-01-07 10:00:00'> > slow_report3.log  

4、通过tcpdump抓取mysql的tcp协议数据,然后再分析

tcpdump -s 65535 -x -nn -q -tttt -i any -c 1000 port 3306 > mysql.tcp.txt    pt-query-digest --type tcpdump mysql.tcp.txt> slow_report9.log  

5、分析binlog

mysqlbinlog mysql-bin.000093 > mysql-bin000093.sql
pt-query-digest –type=binlog mysql-bin000093.sql > slow_report10.log

6、分析general log

pt-query-digest  --type=genlog  localhost.log > slow_report11.log  

四、参考资料

1、高性能mysql(第三版)

2、MySQL慢查询分析工具pt-query-digest详解 作者:枫叶工作室。

3、Warning: prerequisite DBD::mysql 3 not found 作者:ora600

4、使用lcov时遇到错误can’t locate Digest/MD5.pm in @INC (@INC contains: /usr/local/lib64/perl5 …的错误 作者:迷茫的叶

5、percona官网