需要優化的sql語句如下：
SELECT
a.SYS_ID,
a.CUST_TYPE,
a.CUST_ID,
a.CUST_NAME,
a.area_info,
a.CUST_RAGE,
a.INFO_AUTH,
a.CUST_STATE,
a.LIST_TOP,
a.COMMEND,
DATE_FORMAT(a.auth_date, '%Y-%m-%d') AS auth_date,
DATE_FORMAT(a.add_date, '%Y-%m-%d') AS add_date,
a.email,
DATE_FORMAT(a.oper_time, '%Y-%m-%d') AS oper_time,
a.COMPANY_ADDRESS,
s.CELL_PHONE AS CONTACT_PHONE,
a.ACCESS_STATUS,
s.CONTACT_NAME AS CONTACT_NAME,
b.MESSAGE_SWITCH,
b.USER_NAME,
b.user_Id AS user_Id,
b.QQ_NUM,
b.CELLPHONE,
b.USER_STATE,
b.ROLE_AUTH_ID,
(SELECT
r.role_name
FROM role r
WHERE r.ROLE_AUTH_ID = b.ROLE_AUTH_ID) AS role_name,
d.CONTACT_NAME AS customerServiceUserName,
f.CONTACT_NAME AS devCustomerServiceUser,
a.MAJOR_TRADE_ID,
a.AVAILABILITY_TRADE_ID,
b.PERSONAL_ID,
a.BUIS_REMARK,
CASE WHEN d.REGIONAL_MANAGER = 0 THEN d.MANAGE_AREA ELSE d.SALESMAN_MANAGER_AREA END AS SALESMAN_MANAGER_NAME
FROM spot_company a
LEFT JOIN (SELECT
rf.field_id_val,
rf.dept_id,
rf.ROLE_ID,
rf.user_Id
FROM rows_filter rf
WHERE rf.table_name = UPPER('spot_company')
AND rf.field_name = UPPER('CUST_ID')) rf
ON a.CUST_ID = rf.field_id_val
LEFT JOIN spot_user_info b
ON a.CUST_ID = b.CUST_ID
LEFT JOIN (
SELECT *
FROM spot_frequent_contacts
WHERE (IS_FREQUENT, CUST_ID) IN (SELECT
MAX(IS_FREQUENT),
CUST_ID
FROM spot_frequent_contacts
GROUP BY CUST_ID)
GROUP BY CUST_ID) s
ON s.CUST_ID = a.CUST_ID
LEFT JOIN spot_user_info d
ON a.CUSTOMER_SERVICE_USER = d.user_Id
LEFT JOIN spot_user_info f
ON a.DEV_CUSTOMER_SERVICE_USER = f.user_Id
WHERE a.CUST_ID != 100000000000000
AND b.USER_TYPE = '1'
AND a.CUST_STATE != 0
AND b.USER_ACCOUNT_TYPE = 1
AND b.USER_STATE = 1
AND a.CUST_STATE != 9
AND a.CUST_STATE != 3
ORDER BY a.oper_time DESC, a.CUST_ID DESC;
執行explain分析後得出
id select_type table type possible_keys key key_len ref rows Extra
------ ------------------ ---------------------- ------ ------------------------------------------------- ---------------- ------- --------------------------------- ------ ----------------------------------------------------
1 PRIMARY b ALL CUST_ID,IDX_USER_TYPE,user_state (NULL) (NULL) (NULL) 1963 Using where; Using temporary; Using filesort
1 PRIMARY a eq_ref PRIMARY,company_custid,IDX_CUST_STATE_CUST_SUPPLY PRIMARY 8 paohe.b.CUST_ID 1 Using where
1 PRIMARY <derived4> ALL (NULL) (NULL) (NULL) (NULL) 260 Using where; Using join buffer (Block Nested Loop)
1 PRIMARY d eq_ref PRIMARY,user_id PRIMARY 8 paohe.a.CUSTOMER_SERVICE_USER 1 Using where
1 PRIMARY f eq_ref PRIMARY,user_id PRIMARY 8 paohe.a.DEV_CUSTOMER_SERVICE_USER 1 Using where
1 PRIMARY <derived3> ALL (NULL) (NULL) (NULL) (NULL) 1187 Using where; Using join buffer (Block Nested Loop)
4 DERIVED spot_frequent_contacts index IDX_FC_CUST_ID IDX_FC_CUST_ID 62 (NULL) 260 Using where
5 SUBQUERY spot_frequent_contacts index IDX_FC_CUST_ID IDX_FC_CUST_ID 62 (NULL) 260 (NULL)
3 DERIVED rf ref PRIMARY PRIMARY 304 const,const 1187 Using where
2 DEPENDENT SUBQUERY r ref IDX_ROLE_AUTH_ID IDX_ROLE_AUTH_ID 48 paohe.b.role_auth_id 1 Using index condition
看到執行結果裏面除了外層的sql 類型（type）爲ALL之外，其他內容都執行了主鍵或索引，在索引優化方面是沒有什麼可以再優化了。
之後想看一下profile信息看看sql的執行時間，先執行 SET profiling=1讓mysql做一下記錄，用完之後記得將其重新設置爲0。
再執行一次上面的sql語句看看記錄時間。
Query_ID Duration Query
-------- ---------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
11 0.00007050 SET profiling_history_size = 15
12 0.00118750 SHOW STATUS
13 1.77348200 SELECT
a.SYS_ID,
a.CUST_TYPE,
a.CUST_ID,
a.CUST_NAME,
a.area_info,
a.CUST_RAGE,
a.INFO_AUTH,
a.CUST_STATE,
a.LIST_TOP,
a.COMMEND,
DATE_FORMAT(a.auth_date, '%Y-%m-%d') AS auth_date,
DATE_FORMAT(a.add_date, '%Y-%m-%d') AS add_date,
a.email,
DATE_FORMAT(a.oper_time, '%Y-%m-%
14 0.00112575 SHOW STATUS
15 0.00070225 select state, round(sum(duration),5) as `duration (summed) in sec` from information_schema.profiling where query_id = 13 group by state order by `duration (summed) in sec` desc
這裏主要是看13，整個sql執行用了1.7734秒，只是執行一條sql語句居然就用了差不多2秒，因此會覺得網頁很慢。
但是看了索引已經沒有優化的情況下還是執行這麼慢應該如何入手？
那就有可能是join的時候出現問題了，最有可能的就是連接的時候字段類型不一致導致，連接的時候先做類型轉換，之後再做連接遍歷。
排查了一下，如果將
LEFT JOIN (SELECT
rf.field_id_val,
rf.dept_id,
rf.ROLE_ID,
rf.user_Id
FROM rows_filter rf
WHERE rf.table_name = UPPER('spot_company')
AND rf.field_name = UPPER('CUST_ID')) rf
ON a.CUST_ID = rf.field_id_val
這一段sql語句屏蔽了之後性能就上來了，這段是行過濾權限查詢的sql。
從語句可以看到最後是用CUST_ID和field_id_val來進行關聯的，而CUST_ID是bigint類型，而field_id_val是varchar類型。由於在這個連接裏面field_id_val字段是屬於外連接部分，因此可以判斷優化的步驟是將field_id_val字段轉換成int類型後再進行關聯。
因此將上面的sql改成：
LEFT JOIN (SELECT
CAST(rf.field_id_val AS UNSIGNED integer) AS field_id_val,
rf.dept_id,
rf.ROLE_ID,
rf.user_Id
FROM rows_filter rf
WHERE rf.table_name = UPPER('spot_company')
AND rf.field_name = UPPER('CUST_ID')) rf
ON a.CUST_ID = rf.field_id_val
使用CAST方法在查詢出來的時候先將field_id_val字段轉成Integer類型再做關聯，之後我們再看看執行時間。
Query_ID Duration Query
-------- ---------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
17 0.00007125 SET profiling_history_size = 15
18 0.00117550 SHOW STATUS
19 0.31157250 SELECT
a.SYS_ID,
a.CUST_TYPE,
a.CUST_ID,
a.CUST_NAME,
a.area_info,
a.CUST_RAGE,
a.INFO_AUTH,
a.CUST_STATE,
a.LIST_TOP,
a.COMMEND,
DATE_FORMAT(a.auth_date, '%Y-%m-%d') AS auth_date,
DATE_FORMAT(a.add_date, '%Y-%m-%d') AS add_date,
a.email,
DATE_FORMAT(a.oper_time, '%Y-%m-%
20 0.00116000 SHOW STATUS
21 0.00071575 select state, round(sum(duration),5) as `duration (summed) in sec` from information_schema.profiling where query_id = 19 group by state order by `duration (summed) in sec` desc
看得出在修改之後19，執行時間縮短到了0.3115秒，性能提高了80%。