当我们需要优化MySQL查询时,第一想到的是开启慢日志,慢日志可以看到执行消耗超过一定时间的SQL语句和未使用索引的SQL。但如果我们想分析所有SQL查询的分布,即哪类SQL查询次数最多的时候,我们可以开启general log来统计。
开启general log
-
mysql> show variables like '%general%';
+------------------+-------------------------------------+
| Variable_name | Value |
+------------------+-------------------------------------+
| general_log | OFF |
| general_log_file | /usr/local/mysql/data/localhost.log |
+------------------+-------------------------------------+
-
mysql> set global general_log = "ON";
analysis-general-log.py脚本
-
#!/usr/bin/python
-
-
# sort and count mysql general log
-
# Author: Jason
-
# Url: www.centos.bz
-
# Email: admin#centos.bz
-
# Created: UTC 2015-02-15 17:51:53
-
-
import re
-
import sys
-
import os
-
-
if len(sys.argv) == 2:
-
logPath = sys.argv[1]
-
if not os.path.exists(logPath):
-
print ("file " + logPath + " does not exists.")
-
sys.exit(1)
-
else:
-
print ("Usage: " + sys.argv[0] + " logPath")
-
sys.exit(1)
-
-
logFo = open(logPath)
-
match = 0
-
-
for line in logFo:
-
line = re.sub(r"\n","",line)
-
if match == 0:
-
# match line begin with numbers
-
lineMatch = re.match(r"\s+[0-9]+\s+.*",line,flags=re.I)
-
if lineMatch:
-
lineTmp = lineMatch.group(0)
-
match = match + 1
-
continue
-
-
elif match == 1:
-
# match line begin with numbers
-
lineMatch = re.match(r"\s+[0-9]+\s+.*",line,flags=re.I)
-
if lineMatch:
-
# match only query
-
lineMatchQuery = re.match(r".*Query\s+(.*)",lineTmp,flags=re.I)
-
if lineMatchQuery:
-
lineTmp = lineMatchQuery.group(1)
-
# remove extra space
-
lineTmp = re.sub(r"\s+", " ",lineTmp)
-
# replace values (value) to values (x)
-
lineTmp = re.sub(r"values\s*\(.*?\)", "values (x)",lineTmp,flags=re.I)
-
# replace filed = 'value' to filed = 'x'
-
lineTmp = re.sub(r"(=|>|<|>=|<=)\s*('|\").*?\2","\\1 'x'",lineTmp)
-
# replace filed = value to filed = x
-
lineTmp = re.sub(r"(=|>|<|>=|<=)\s*[0-9]+","\\1 x",lineTmp)
-
# replace like 'value' to like 'x'
-
lineTmp = re.sub(r"like\s+('|\").*?\1","like 'x'",lineTmp,flags=re.I)
-
# replace in (value) to in (x)
-
lineTmp = re.sub(r"in\s+\(.*?\)","in (x)",lineTmp,flags=re.I)
-
# replace limit x,y to limit
-
lineTmp = re.sub(r"limit.*","limit",lineTmp,flags=re.I)
-
-
print (lineTmp)
-
-
match = 1
-
lineTmp = lineMatch.group(0)
-
else:
-
lineTmp += line
-
match = 1
-
-
logFo.close()
使用方法:
-
analysis-general-log.py general.log | sort | uniq -c | sort -nr
1032 SELECT * FROM wp_comments WHERE ( comment_approved = 'x' OR comment_approved = 'x' ) AND comment_post_ID = x ORDER BY comment_date_gmt DESC
653 SELECT post_id, meta_key, meta_value FROM wp_postmeta WHERE post_id in (x) ORDER BY meta_id ASC
527 SELECT FOUND_ROWS()
438 SELECT t.*, tt.* FROM wp_terms AS t INNER JOIN wp_term_taxonomy AS tt ON t.term_id = tt.term_id WHERE tt.taxonomy = 'x' AND t.term_id = x limit
341 SELECT option_value FROM wp_options WHERE option_name = 'x' limit
329 SELECT t.*, tt.*, tr.object_id FROM wp_terms AS t INNER JOIN wp_term_taxonomy AS tt ON tt.term_id = t.term_id INNER JOIN wp_term_relationships AS tr ON tr.term_taxonomy_id = tt.term_taxonomy_id WHERE tt.taxonomy in (x) AND tr.object_id in (x) ORDER BY t.name ASC
311 SELECT wp_posts.* FROM wp_posts WHERE 1= x AND wp_posts.ID in (x) AND wp_posts.post_type = 'x' AND ((wp_posts.post_status = 'x')) ORDER BY wp_posts.post_date DESC
219 SELECT wp_posts.* FROM wp_posts WHERE ID in (x)
218 SELECT tr.object_id FROM wp_term_relationships AS tr INNER JOIN wp_term_taxonomy AS tt ON tr.term_taxonomy_id = tt.term_taxonomy_id WHERE tt.taxonomy in (x) AND tt.term_id in (x) ORDER BY tr.object_id ASC
217 SELECT wp_posts.* FROM wp_posts WHERE 1= x AND wp_posts.ID in (x) AND wp_posts.post_type = 'x' AND ((wp_posts.post_status = 'x')) ORDER BY wp_posts.menu_order ASC
202 SELECT SQL_CALC_FOUND_ROWS wp_posts.ID FROM wp_posts WHERE 1= x AND wp_posts.post_type = 'x' AND (wp_posts.post_status = 'x') ORDER BY wp_posts.post_date DESC limit
118 SET NAMES utf8
115 SET SESSION sql_mode= 'x'
115 SELECT @@SESSION.sql_mode
112 SELECT option_name, option_value FROM wp_options WHERE autoload = 'x'
111 SELECT user_id, meta_key, meta_value FROM wp_usermeta WHERE user_id in (x) ORDER BY umeta_id ASC
108 SELECT YEAR(min(post_date_gmt)) AS firstdate, YEAR(max(post_date_gmt)) AS lastdate FROM wp_posts WHERE post_status = 'x'
108 SELECT t.*, tt.* FROM wp_terms AS t INNER JOIN wp_term_taxonomy AS tt ON t.term_id = tt.term_id WHERE tt.taxonomy in (x) AND tt.count > x ORDER BY tt.count DESC limit
107 SELECT t.*, tt.* FROM wp_terms AS t INNER JOIN wp_term_taxonomy AS tt ON t.term_id = tt.term_id WHERE tt.taxonomy in (x) AND t.term_id in (x) ORDER BY t.name ASC
107 SELECT * FROM wp_users WHERE ID = 'x'
106 SELECT SQL_CALC_FOUND_ROWS wp_posts.ID FROM wp_posts WHERE 1= x AND wp_posts.post_type = 'x' AND (wp_posts.post_status = 'x') AND post_date > 'x' ORDER BY wp_posts.post_date DESC limit
106 SELECT SQL_CALC_FOUND_ROWS wp_posts.ID FROM wp_posts WHERE 1= x AND wp_posts.post_type = 'x' AND (wp_posts.post_status = 'x') AND post_date > 'x' ORDER BY RAND() DESC limit
105 SELECT SQL_CALC_FOUND_ROWS wp_posts.ID FROM wp_posts WHERE 1= x AND wp_posts.post_type = 'x' AND (wp_posts.post_status = 'x') AND post_date > 'x' ORDER BY wp_posts.comment_count DESC limit
(责任编辑:最模板) |