1、查看監控，發現整點時間有寫IO過高情況
2、iotop 分析確認io高峰是由mysql導致的
3、開啟general log，分析SQL
set global general_log = on ;
4、觀察binlog 與 general log 發現 文件增長量不大，懷疑不是有Insert 與 update 與 delete 導致的 寫IO過高
5、show full processlist ;發現有慢SQL
*************************** 6. row ***************************
Id: 337153
User: user_car_bill
Host: 192.168.3.10:63018
db: yoolifin
Command: Query
Time: 295
State: Creating sort index
Info: SELECT 1 AS `Number of Records`,
`自定義 SQL 查詢`.`實還金額` AS `實還金額`,
`自定義 SQL 查詢`.`應還客戶數` AS `應還客戶數`,
`自定義 SQL 查詢`.`應還金額` AS `應還金額`,
`自定義 SQL 查詢`.`累計發過代扣客戶數` AS `累計發過代扣客戶數`,
`自定義 SQL 查詢`.`累計客戶覆蓋率` AS `累計客戶覆蓋率`,
`自定義 SQL 查詢`.`累計還款收回率` AS `累計還款收回率`,
`自定義 SQL 查詢`.`還款日發過代扣客戶數` AS `還款日發過代扣客戶數`,
`自定義 SQL 查詢`.`還款日客戶覆蓋率` AS `還款日客戶覆蓋率`,
`自定義 SQL 查詢`.`還款日期` AS `還款日期`,
`自定義 SQL 查詢`.`還款日還款收回率` AS `還款日還款收回率`
FROM (
select a.dt 還款日期,a.c 應還客戶數,a.totle_repayment 應還金額,a.fact_repayment 實還金額, a.fact_repayment/a.totle_repayment 累計還款收回率, a.very_fact_repayment/a.totle_repayment 還款日還款收回率,t.c 還款日發過代扣客戶數,t.c/a.c 還款日客戶覆蓋率,tt.c 累計發過代扣客戶數,tt.c/a.c 累計客戶覆蓋率
FROM(
select date(payment_date) dt,count(DISTINCT APP_NO) c,sum(totle_repayment) totle_repayment,sum(fact_repayment) fact_repayment ,sum(if(payment_date=fact_payment_date,fact_repayment,0)) very_fact_repayment
from fin_repayment m
where
date(payment_date)>'2017-06-01' and date(payment_date)<=date(curdate() + INTERVAL 20 DAY) and DATE_FORMAT(payment_date,'%d') in ('05','20')
and m.`status` in ('1','2','5')
and ifnull(cut_chanl,0)<>3
group by date(payment_date)
) a
left join (
select date(n.SUBMIT_TIME) dt,count(DISTINCT APP_NO) c
from fin_back_info n
where date(n.SUBMIT_TIME)>'2017-06-01' and DATE_FORMAT(n.SUBMIT_TIME,'%d') in ('05','20') and n.PAYMENT_DATE=DATE(n.SUBMIT_TIME)
group by date(n.SUBMIT_TIME)
) t
on a.dt=t.dt
left join (
select date(n.PAYMENT_DATE) dt,count(DISTINCT APP_NO) c from fin_back_info n
where date(n.SUBMIT_TIME)>'2017-06-01'
group by date(n.PAYMENT_DATE)
) tt
on a.dt=tt.dt
) `自定義 SQL 查詢`
*************************** 6. row ***************************
6、分析慢SQL，產生大量的派生表，可能是由於內存不夠，寫入磁盤導致的IO過高
7、執行該SQL，觀看監控，產生IO高峰，確認為該SQL導致的IO瓶頸過高