在ORACLE 中,我們常常推薦使用exists 來替代in,往往也能取得比較好的優(yōu)化效果。在ORACLE應用遷往MySQL的過程中，我們發(fā)現部分in 的子查詢語句帶到MYSQL中，其執(zhí)行效率變得非常低下，這很讓人覺得匪夷所思。于是，我分析了一波。
成都創(chuàng)新互聯，為您提供網站建設、成都網站制作、網站營銷推廣、網站開發(fā)設計，對服務戶外休閑椅等多個行業(yè)擁有豐富的網站建設及推廣經驗。成都創(chuàng)新互聯網站建設公司成立于2013年,提供專業(yè)網站制作報價服務,我們深知市場的競爭激烈，認真對待每位客戶，為客戶提供賞心悅目的作品。 與客戶共同發(fā)展進步，是我們永遠的責任！
對兩個表，分別是一大一小進行關聯查詢:
mysql> select count(*) from users; +----------+ | count(*) | +----------+ | 19 | +----------+ 1 row in set mysql> select count(*) from orders; +----------+ | count(*) | +----------+ | 86310 | +----------+ 1 row in set mysql>
開啟profile，發(fā)現無論是子查詢是大表還是小表 ，exists的語句總是比in執(zhí)行慢：
mysql> show profiles; +----------+------------+--------------------------------------------------------------------------------------------+ | Query_ID | Duration | Query | +----------+------------+--------------------------------------------------------------------------------------------+ | 1 | 1.08661625 | select count(1) from orders o where o.user_id in(select u.id from users u) | | 2 | 1.56956275 | select count(1) from orders o where exists (select 1 from users u where u.id = o.user_id) | | 3 | 0.81266425 | select count(1) from users u where u.id in(select o.user_id from orders o) | | 4 | 8.4164905 | select count(1) from users u where exists (select 1 from orders o where u.id = o.user_id) | +----------+------------+--------------------------------------------------------------------------------------------+ 4 rows in set
而查看exists語句的profile內容，發(fā)現其存在多個executing 和sending data過程，這是整個sql執(zhí)行的主要耗時過程：
mysql> show profile for query 2 ; +----------------------------+----------+ | Status | Duration | +----------------------------+----------+ | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.5E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 1E-6 | | Sending data | 1.3E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.7E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.1E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 1E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.1E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.2E-5 | | executing | 2E-6 | | Sending data | 1.5E-5 | | end | 3E-6 | | query end | 3E-6 | | waiting for handler commit | 1E-5 | | closing tables | 9E-6 | | freeing items | 0.000152 | | cleaning up | 1.7E-5 | +----------------------------+----------+ 100 rows in set
而在in 子查詢中，sending data這個過程只有一次，這也是整個sql執(zhí)行主要耗時地方：
mysql> show profile for query 1; +--------------------------------+----------+ | Status | Duration | +--------------------------------+----------+ | starting | 9.3E-5 | | Executing hook on transaction | 6E-6 | | starting | 8E-6 | | checking permissions | 5E-6 | | checking permissions | 4E-6 | | Opening tables | 0.004849 | | init | 1.8E-5 | | System lock | 1.4E-5 | | optimizing | 1.4E-5 | | statistics | 3.1E-5 | | preparing | 2.2E-5 | | executing | 3E-6 | | Sending data | 1.081273 | 《《《《《《《《《《《《《《《《《《《 | end | 1.3E-5 | | query end | 3E-6 | | waiting for handler commit | 1E-5 | | closing tables | 5.2E-5 | | freeing items | 0.000171 | | cleaning up | 2.9E-5 | +--------------------------------+----------+
關于sending data和executing 解析：
Sending data
The thread is reading and processing rows for a SELECT statement, and sending data to the client. Because operations occurring during this state tend to perform large amounts of disk access (reads), it is often the longest-running state over the lifetime of a given query.
說Sending data 是：線程正在為一個select語句讀取和處理行，并且發(fā)送數據到客戶端。因為這期間操作傾向于大量的磁盤訪問（讀?。?，所以這常是整個查詢周期中運行時間最長的階段。（這是MySQL 5.5的解釋，5.7的解釋完全一樣，但是5.7多了一個Sending to client狀態(tài)）
這樣就清楚了，Sending data 做了 讀取，處理（過濾，排序等。。）和發(fā)送 三件事情，接下來再看該狀態(tài)下的cpu 和 io 信息 以分析語句的瓶頸是 讀取還是處理 ，再做相應的優(yōu)化調整。。
executing
The thread has begun executing a statement.
在exists語句中不斷地executing和Sending data 應該是在不斷地掃描抓取數據進行匹配，那這應該與MYSQL 的算法有關，通過trace查看其實是在多次子查詢 join_execution：
重慶分公司
重慶分公司
{
"subselect_execution": {
"select#": 2,
"steps": [
{
"join_execution": {
"select#": 2,
"steps": [
] /* steps */
} /* join_execution */
}
] /* steps */
} /* subselect_execution */
},
想知道這是雞肋嗎，不知有沒有相關的解析？
