关于MySQL的慢日志分析工具

天我们看看关于MySQL慢日志的阅读。
我们知道，如果我们的语句不够优化，那么首先MySQL的慢日志是进一步优化的离线证据，虽然里面有好多“伪慢语句”！
先不说怎么优化，如果你的日志有一条语句赌住了，那么会有不计其数的慢语句填充到MySQL的满日志里面。那么首先提炼出这些语句就非常头疼。
今天主要介绍两种工具：
1,mysqldumpslow。（咱们MySQL自带的简单而又实用的工具）

我们先来看下mysqldumpslow的结果。
[root@localhost ~]# mysqldumpslow -r localhost-slow.log

Reading mysql slow query log from localhost-slow.log
Count: 2 Time=7.00s (14s) Lock=0.00s (0s) Rows=0.0 (0), root[root]@localhost
select * from t_page_sample order by id desc limit N,N

Count: 1 Time=11.00s (11s) Lock=0.00s (0s) Rows=1.0 (1), root[root]@localhost
select count(*) from t_page_sample

Count: 1 Time=1418.00s (1418s) Lock=0.00s (0s) Rows=0.0 (0), root[root]@localhost
insert ignore into t_page_sample select ceil(rand()*N), ceil(rand()*N), date_sub(now(),interval floor(rand()*N) day), now() from t_page_sample

比如要查找排序的慢语句：
[root@localhost ~]# mysqldumpslow -r -g "order by " localhost-slow.log

Reading mysql slow query log from localhost-slow.log
Count: 2 Time=7.00s (14s) Lock=0.00s (0s) Rows=0.0 (0), root[root]@localhost
select * from t_page_sample order by id desc limit N,N

自带的mysqldumpslow简单实用，作为我个人的首选。关于具体的参数含义，可以参见它自身的HELP。

2,mk-query-digest。(网上著名的开源脚本家族Maatkit中一员)
手册地址：
下载方法：wget
完了赋给它可执行权限就OK。

mk-query-digest 功能太多，我今天只是试了下它对MySQL慢日志的分析功能。
以下是我觉得比较实用的功能。
1）分析慢日志并且把找出来的语句写到规定的表里。
[root@localhost ~]# ./mk-query-digest --limit 2 --select Query_time,Lock_time,Rows_sent,Rows_examined,ts --create-review-table --review D=t_girl,t=query_review localhost-slow.log

# 280ms user time, 80ms system time, 11.56M rss, 16.65M vsz
# Current date: Sat May 8 02:47:39 2010
# Files: localhost-slow.log
# Overall: 4 total, 3 unique, 0.01 QPS, 1.96x concurrency ________________
# total min max avg 95% stddev median
# Exec time 1443s 5s 1418s 361s 1357s 584s 684s
# Lock time 0 0 0 0 0 0 0
# Rows sent 1 0 1 0.25 0.99 0.43 0
# Rows exam 6.98M 1.69M 1.89M 1.74M 1.86M 76.62k 1.69M
# Time range 2010-05-08 00:28:42 to 2010-05-08 00:40:58

# Profile
# Rank Query ID Response time Calls R/Call Item
# ==== ================== ================ ===== ========= ===============
# 1 0x2A94F91D8C3B4B26 1418.0000 99.0% 1 1418.0000 INSERT SELECT t_page_sample
# 2 0x06754F1BD3C8D697 14.0000 1.0% 2 7.0000 SELECT t_page_sample

# Query 1: 0 QPS, 0x concurrency, ID 0x2A94F91D8C3B4B26 at byte 0 ________
# This item is included in the report because it matches --limit.
# pct total min max avg 95% stddev median
# Count 25 1
# Exec time 98 1418s 1418s 1418s 1418s 1418s 0 1418s
# Lock time 0 0 0 0 0 0 0 0
# Rows sent 0 0 0 0 0 0 0 0
# Rows exam 27 1.89M 1.89M 1.89M 1.89M 1.89M 0 1.89M
# Time range 2010-05-08 00:28:42 to 2010-05-08 00:28:42
# Query_time distribution
# 1us
# 10us
# 100us
# 1ms
# 10ms
# 100ms
# 1s
# 10s+ ################################################################
# Review information
# first_seen: 2010-05-08 00:28:42
# last_seen: 2010-05-08 00:28:42
# reviewed_by:
# reviewed_on:
# comments:
# Tables
# SHOW TABLE STATUS FROM `t_girl` LIKE 't_page_sample'\G
# SHOW CREATE TABLE `t_girl`.`t_page_sample`\G
insert ignore into t_page_sample select ceil(rand()*10000000), ceil(rand()*9), date_sub(now(),interval floor(rand()*20) day), now() from t_page_sample\G

# Query 2: 0.07 QPS, 0.47x concurrency, ID 0x06754F1BD3C8D697 at byte 1499
# This item is included in the report because it matches --limit.
# pct total min max avg 95% stddev median
# Count 50 2
# Exec time 0 14s 5s 9s 7s 9s 3s 7s
# Lock time 0 0 0 0 0 0 0 0
# Rows sent 0 0 0 0 0 0 0 0
# Rows exam 48 3.39M 1.69M 1.69M 1.69M 1.69M 0 1.69M
# Time range 2010-05-08 00:40:28 to 2010-05-08 00:40:58
# Query_time distribution
# 1us
# 10us
# 100us
# 1ms
# 10ms
# 100ms
# 1s ################################################################
# 10s+
# Review information
# first_seen: 2010-05-08 00:40:28
# last_seen: 2010-05-08 00:40:58
# reviewed_by:
# reviewed_on:
# comments:
# Tables
# SHOW TABLE STATUS FROM `t_girl` LIKE 't_page_sample'\G
# SHOW CREATE TABLE `t_girl`.`t_page_sample`\G
# EXPLAIN
select * from t_page_sample order by id desc limit 4400000,2\G

[root@localhost ~]#

因为慢日志里面会有写和读语句，所以当以后想要分析某类语句时，只需要简单的SELECT即可出来。
mysql> select * from query_review where fingerprint like 'select%'\G

*************************** 1. row ***************************
checksum: 465365117438580375
fingerprint: select * from t_page_sample order by id desc limit ?
sample: select * from t_page_sample order by id desc limit 4400000,2
first_seen: 2010-05-08 00:40:28
last_seen: 2010-05-08 00:40:58
reviewed_by: NULL
reviewed_on: NULL
comments: NULL
1 rows in set (0.00 sec)

2. 分析当前运行的SQL语句。
以前我都是自己写脚本，配合CRONTAB来定时抓取信息到固定的文件里以备分析。不过现在可以用它来打印出比较详细的报告来。
[root@localhost ~]# ./mk-query-digest --select Query_time,Lock_time,Rows_sent,Rows_examined --processlist h=localhost,u=root

# Caught SIGINT.

# 690ms user time, 2.4s system time, 11.27M rss, 16.41M vsz
# Current date: Sat May 8 03:17:39 2010
# Files: STDIN
# Overall: 1 total, 1 unique, 0 QPS, 0x concurrency ______________________
# total min max avg 95% stddev median
# Exec time 1273313855s 1273313855s 1273313855s 1273313855s 1273313855s 0 1273313855s
# Lock time 0 0 0 0 0 0 0

# Profile
# Rank Query ID Response time Calls R/Call Ite
# ==== ================== ====================== ===== =============== ===
# 1 0xB52E1970DE36E57F 1273313854.8595 100.0% 1 1273313854.8595 SELECT t_page_sample

# Query 1: 0 QPS, 0x concurrency, ID 0xB52E1970DE36E57F at byte 0 ________
# This item is included in the report because it matches --limit.
# pct total min max avg 95% stddev median
# Count 100 1
# Exec time 100 1273313855s 1273313855s 1273313855s 1273313855s 1273313855s 0 1273313855s
# Lock time 0 0 0 0 0 0 0 0
# Query_time distribution
# 1us
# 10us
# 100us
# 1ms
# 10ms
# 100ms
# 1s
# 10s+ ################################################################
# Tables
# SHOW TABLE STATUS FROM `t_girl` LIKE 't_page_sample'\G
# SHOW CREATE TABLE `t_girl`.`t_page_sample`\G
# EXPLAIN
select count(*) from t_page_sample\G
[root@localhost ~]#

当想停止截取当前语句时，按住CTRL+C就OK。