The palest ink is better than best memory——好記性不如爛筆頭。2012補記

一、關鍵詞

SQL優化、執行計劃、explain、using filesort、using temporary

二、摘要

開發中，為滿足一業務功能，使用mysql書寫sql時，一條sql往往有多種寫法，那麼我們就需要選擇執行效率比較高的sql。
因此要比較分析sql的執行過程，且同一條sql我們要比較選擇使用最優索引。
通過explain命令可以得到:

整體執行順序（子查詢）、查詢型別、使用什麼表（表 泛指）、資料訪問/讀取操作型別、使用到索引等。

先整體瞭解下SQL查詢的基本執行過程：

1）應用通過MySQL API把查詢命令傳送給MySQL伺服器，然後被解析

2）檢查許可權、MySQL optimizer進行優化，經過解析和優化後的查詢命令被編譯為CPU可執行的二進位制形式的查詢計劃（query
plan），並可以被快取

3）如果存在索引，那麼先掃描索引，如果資料被索引覆蓋，那麼不需要額外的查詢，如果不是，根據索引查詢和讀取對應的記錄

4）如果有關聯查詢，查詢次序是掃描第一張表找到滿足條件的記錄，按照第一張表和第二張表的關聯鍵值，掃描第二張表查詢滿足條件的記錄，按此順序迴圈

5）輸出查詢結果，並記錄binary logs

三、Explain返回資訊：

• select_type列：

• type列（依次從最差到最優）：

• Extra列常見情況（需要優化）：

四、事例（2012公司專案）：

案例1：

EXPLAIN
SELECT e1.*,0 AS 'hasCount' FROM eticketdetail e1 WHERE e1.brandUserId =5303 AND e1.status =1 AND e1.parentId =0 
UNION  
SELECT e2.*,0 AS 'hasCount' FROM eticketdetail e2 WHERE e2.status =1 AND  e2.id IN (SELECT e3.parentId FROM eticketdetail e3 WHERE  e3.parentId >0 AND e3.brandUserId = 5303)
ORDER BY createtime LIMIT 0,10

EXPLAIN
SELECT e1.*,0 AS 'hasCount' FROM eticketdetail e1 WHERE e1.brandUserId =5303 AND e1.status =1 AND e1.parentId =0 
UNION  
SELECT e2.*,0 AS 'hasCount' FROM eticketdetail e2 WHERE e2.status =1 AND  EXISTS (SELECT 1 FROM eticketdetail e3 WHERE  e3.parentId >0 AND e3.brandUserId = 5303)
ORDER BY createtime LIMIT 0,10

EXPLAIN
SELECT e1.*,0 AS 'hasCount' FROM eticketdetail e1 WHERE e1.brandUserId =5303 AND e1.status =1 AND e1.parentId =0 
UNION  
SELECT e2.*,0 AS 'hasCount' FROM eticketdetail e2 INNER JOIN (SELECT e3.parentId FROM eticketdetail e3 WHERE  e3.parentId >0 AND e3.brandUserId = 5303) tab ON e2.id = tab.parentid  WHERE e2.status =1 
ORDER BY createtime LIMIT 0,10

案例2：

第一種寫法：
SELECT 
  mp.*
FROM memberpromotion mp
  INNER JOIN memberpromotionlocation mpl
    ON mpl.MemberPromotionID = mp.ID
WHERE mp.status = 1
    AND mp.BrandUserID = 2402
AND mpl.shopId = 0

第二種寫法：
SELECT 
  mp.*
FROM (SELECT * FROM memberpromotion memb WHERE memb.BrandUserID = 2402) mp
  INNER JOIN memberpromotionlocation mpl
    ON mpl.MemberPromotionID = mp.ID
WHERE mp.status = 1
AND mpl.shopId = 0

案例3：

第一幅圖語句：
SELECT SQL_NO_CACHE
  e.*,
  (SELECT
     COUNT(id)
   FROM eticketdetail
   WHERE userId = e.userId
       AND orderId = e.orderId
       AND promotionId = e.promotionId
       AND promotionProductId = e.promotionProductId
       AND STATUS != 3
       AND STATUS = 0 ) AS 'hasCount'
FROM eticketdetail e
WHERE userId = 5302
    AND isGift = 0 AND isReal = 0 AND STATUS = 0
GROUP BY userId,orderId,promotionId,promotionProductId 
UNION 
SELECT e.*, 1    AS 'hasCount' FROM eticketdetail e
WHERE userId = 5302  AND isGift = 1 AND STATUS = 0
ORDER BY CreateTime DESC

第二幅圖語句：（UNION改為UNION ALL）
SELECT SQL_NO_CACHE
  e.*,
  (SELECT
     COUNT(id)
   FROM eticketdetail
   WHERE userId = e.userId
       AND orderId = e.orderId
       AND promotionId = e.promotionId
       AND promotionProductId = e.promotionProductId
       AND STATUS != 3
       AND STATUS = 0 ) AS 'hasCount'
FROM eticketdetail e
WHERE userId = 5302
    AND isGift = 0
    AND isReal = 0
    AND STATUS = 0
GROUP BY userId,orderId,promotionId,promotionProductId 
UNION ALL
SELECT e.*, 1    AS 'hasCount' FROM eticketdetail e
WHERE userId = 5302  AND isGift = 1 AND STATUS = 0
ORDER BY CreateTime DESC

第三幅圖： 增加userId索引後

第四幅圖：
SELECT 
  e.*,
  tab.t AS 'hasCount'
FROM eticketdetail e 
INNER JOIN (
    SELECT id,COUNT(id) AS t FROM eticketdetail WHERE userid=5302  AND isGift = 0 AND isReal = 0 AND STATUS != 3 AND STATUS = 0 GROUP BY orderId,promotionId,promotionProductId ORDER BY NULL
) tab ON tab.id = e.ID
WHERE userId = 5302
    AND isGift = 0
    AND isReal = 0
    AND STATUS = 0
GROUP BY orderId,promotionId,promotionProductId 
UNION ALL
SELECT e.*, 1    AS 'hasCount' FROM eticketdetail e
WHERE userId =5302  AND isGift = 1 AND STATUS = 0
ORDER BY CreateTime DESC

這玩意，只有動手多折騰比較幾下~

五、參考：