需要記錄慢查詢,在配置文件中加入慢查詢的相關配置。
/etc/my.cnf
general_log=off
log_output=table
slow_query_log=1
long_query_time=0.3
general_log=off 會記錄所有的查詢,建議關閉(set global general_log = off)
log_output=table 將log記錄到表中
slow_query_log=1 開啓慢查詢記錄
long_query_time=0.3 大於0.3秒的查詢會被記錄
配置好以後,我的做法是使用python每週讀取數據庫中的記錄,郵件發出來以便分析。
sql如下:
SELECT max(user_host), min(start_time), max(start_time), count(*), sum( CASE WHEN TIME_TO_SEC(query_time) = 0 THEN 0.3 ELSE TIME_TO_SEC(query_time) END ), max( CASE WHEN TIME_TO_SEC(query_time) = 0 THEN 0.3 ELSE TIME_TO_SEC(query_time) END ), min( CASE WHEN TIME_TO_SEC(query_time) = 0 THEN 0.3 ELSE TIME_TO_SEC(query_time) END ), round( avg( CASE WHEN TIME_TO_SEC(query_time) = 0 THEN 0.3 ELSE TIME_TO_SEC(query_time) END ), 2 ), sql_text FROM slow_log WHERE sql_text NOT LIKE '%SQL_NO_CACHE%' GROUP BY LEFT (sql_text, 40) ORDER BY count(*) DESC LIMIT 20;
部分python代碼如下:
for i in range(len(conn)): log=p.get_table(conn[i],sql) query_log+='_'*120+'\r\n' query_log+="Slow_Log_%s:%s____%s\r\n\r\n" % (conn[i]["host"],conn[i]["port"],days) for n in range(len(log)): query_log+="%s %s %s %s %s " % (str(log[n][3]),str(log[n][4]),str(log[n][5]),str(log[n][6]),str(log[n][7])) query_log+="user_host :%s start_time :%s end_time :%s\r\n" % (str(log[n][0]),str(log[n][1]),str(log[n][2])) query_log+=' '*40+squeeze(str(log[n][8]))+'\r\n\r\n' mail.send_mail(mail_list,mail_title,query_log)
得到的郵件部分:
Slow_Log_x.x.x.x:3306____2016-05-26
1889 704.4 33.0 0.3 0.37 user_host :x[x] @ [x] start_time :2016-05-25 09:30:27 end_time :2016-05-26 09:16:05
select max(online_user),round(sum(online_user)/count(id)) from online where datetime >= '2016-05-25' and server_id = 27
這樣就可以看到每週比較慢的查詢,然後進行優化分析。