項目上線後,談一下感觸比較深的一點:查詢優化

前言

  開心一刻git

    兒子有道題不會作,喊我過去教他。我推了推一旁的老公:我頭疼,你去吧。老公不動,我:零花錢漲一千。話音剛落,老公就屁顛屁顛跑去兒子房間。進去不到幾分鐘,一聲怒吼伴隨着兒子的哭聲傳來的瞬間,老公從兒子房間出來,邊走邊說:「朽木不可雕也。」兒子從房間探出半個身子,一臉委屈:「爸爸也不會作,他說給我一塊錢,讓我明天早點去學校抄同窗的。還讓我不要告訴你,我不願,他就吼我。」sql

問題背景

  前段時間,被緊急調到一個新項目,支撐新項目的開發。跌跌撞撞之下,項目也正常上線了,期間收穫頗多,不管是業務上的,仍是業務以外的。業務上的就很少說了,不具通用性,意義不大,有一點業務以外的東東給個人感觸比較深,特記錄下來,與你們分享下 : 查詢優化數據庫

查詢優化

  完整示例工程:data-init,包括數據庫表的 ddl 和 dml,以及數據批量的生成微信

  相關表

    涉及的表很少,一共三張:額度表、記錄表 、 存款表ide

    額度表 t_custmor_credit函數

CREATE TABLE t_customer_credit (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主鍵',
  login_name VARCHAR(50) NOT NULL COMMENT '名稱',
    credit_type TINYINT(1) NOT NULL COMMENT '額度類型,1:自由資金,2:凍結資金,3:優惠',
    amount DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '額度值',
    create_by VARCHAR(50) NOT NULL COMMENT '建立者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '建立時間',
    update_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP COMMENT '建立時間',
    update_by VARCHAR(50) NOT NULL COMMENT '修改者',
  PRIMARY KEY (id)
);
View Code

      記錄每一個顧客的當前額度,額度一共分三種:自由資金、凍結資金和優惠,也就是說每一個顧客會有 3 條記錄來表示他的各個額度。表中數據以下性能

    額度記錄 t_custmor_credit_record優化

CREATE TABLE t_customer_credit_record (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主鍵',
  login_name VARCHAR(50) NOT NULL COMMENT '名稱',
    credit_type TINYINT(1) NOT NULL COMMENT '額度類型,參考t_custmor_credit的credit_type',
    bill_no VARCHAR(50) NOT NULL COMMENT '訂單號',
    amount_before DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '前額度值',
    amount_change DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '額度變化值',
    amount_after DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '後額度值',
    create_by VARCHAR(50) NOT NULL COMMENT '建立者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '建立時間',
    remark VARCHAR(500) NOT NULL DEFAULT '' COMMENT '備註',
  PRIMARY KEY (id)
);
View Code

      記錄顧客額度的每一次變化,只要有額度變化(不論是哪一個額度進行了變化),都會新增3條記錄,每一個類型的額度都會新增一條記錄。另外,該表只會有數據的插入,不會有數據的刪、改。表中數據以下spa

    存款表 t_custmor_deposit設計

CREATE TABLE t_customer_deposit (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主鍵',
  login_name VARCHAR(50) NOT NULL COMMENT '名稱',
    bill_no VARCHAR(50) NOT NULL COMMENT '訂單號',
    amount DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '存款金額',
    deposit_state TINYINT(1) NOT NULL COMMENT '存款狀態: 1成功,2失敗,3未知',
    channal TINYINT(2) NOT NULL COMMENT '存款渠道: 1:銀聯,2支付寶,3微信',
    create_by VARCHAR(50) NOT NULL COMMENT '建立者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '建立時間',
    remark VARCHAR(500) NOT NULL DEFAULT '' COMMENT '備註',
  PRIMARY KEY (id)
);
View Code

      記錄顧客的每一次存款,該表只會有數據的插入,不會有數據的刪、改。表中數據以下

  產品需求

    分頁展現以下列表(具體的過濾條件就不列了,咱們就當是初始狀態,還未輸入過濾條件)

     實現比較簡單,t_custmor_credit_record 左關聯 t_custmor_deposit 就好,可是咱們的額度記錄表與需求列表有些許的出入,須要作一下簡單的行轉列。

  需求實現

    咱們先來看看最初的SQL查詢,這多是不少人最容易想到的

SELECT MIN(tcd.channal) channal, MAX(tccr.id) mId,tccr.login_name,tccr.bill_no,tccr.create_time,
    IF(credit_type=1,amount_before,0) AS freeBefore,
    IF(credit_type=1,amount_change,0) AS freeChange,
    IF(credit_type=1,amount_after,0) AS freeAfter,
    IF(credit_type=2,amount_before,0) AS freezeBefore,
    IF(credit_type=2,amount_change,0) AS freezeChange,
    IF(credit_type=2,amount_after,0) AS freezeAfter,
    IF(credit_type=3,amount_before,0) AS promotionBefore,
    IF(credit_type=3,amount_change,0) AS promotionChange,
    IF(credit_type=3,amount_after,0) AS promotionAfter
FROM t_customer_credit_record tccr
LEFT JOIN t_customer_deposit tcd ON tccr.bill_no = tcd.bill_no
GROUP BY tccr.bill_no,tccr.login_name,tccr.create_time
ORDER BY mId desc
LIMIT 0, 10;
View Code

    數據量少的時候,也許能在咱們接受的時間內查出咱們須要的結果,一旦數據量多了,這個SQL就跑不動了;咱們先看下 60w 數據的狀況下,咱們只進行 t_custmor_credit_record 單表查詢

SELECT MAX(id) mId,login_name,bill_no,create_time,
    IF(credit_type=1,amount_before,0) AS freeBefore,
    IF(credit_type=1,amount_change,0) AS freeChange,
    IF(credit_type=1,amount_after,0) AS freeAfter,
    IF(credit_type=2,amount_before,0) AS freezeBefore,
    IF(credit_type=2,amount_change,0) AS freezeChange,
    IF(credit_type=2,amount_after,0) AS freezeAfter,
    IF(credit_type=3,amount_before,0) AS promotionBefore,
    IF(credit_type=3,amount_change,0) AS promotionChange,
    IF(credit_type=3,amount_after,0) AS promotionAfter
FROM t_customer_credit_record
GROUP BY bill_no,login_name,create_time
ORDER BY mId desc
LIMIT 0, 10;
View Code

    效果以下

    花了近 8 秒,這還只是單表,若是執行上面的聯表SQL,那時間又得增長很多(我試驗的結果是直接卡住了,看不到查詢結果);

  優化處理

    加索引

      查詢慢的時候,咱們最容易想到的優化方式每每就是加索引;上述SQL執行的時候,t_custmor_credit_record 和 t_custmor_deposit都沒有建索引(主鍵索引除外),那麼咱們就加索引唄。個人項目中加的是惟一索引,作了惟一約束,那我這裏也加惟一索引

ALTER TABLE t_customer_credit_record ADD UNIQUE uk_unique (bill_no,login_name,create_time,credit_type);
ALTER TABLE t_customer_deposit ADD UNIQUE uk_billno (bill_no);

      此時咱們看下SQL執行效果

      咱們發現,t_custmor_credit_record 單表查詢的效率幾乎沒變,將近 8 秒,但 t_custmor_credit_record 與 t_custmor_deposit 聯表的查詢卻在 11 秒內有結果了。加了索引爲何還這麼慢了? 難道沒走索引?

      咱們是否是發現了什麼? IF函數對聯表查詢是否走索引有影響,也對單表的查詢速度有影響。上圖中的 t_custmor_credit_record 單表查詢,有IF函數,查詢時間近 8 秒,沒有IF函數,查詢時間 2 秒左右;t_custmor_credit_record 與 t_custmor_deposit 聯表查,有IF函數,t_custmor_credit_record 走的是全表查,查詢時間近 11 秒,沒有IF函數,t_custmor_credit_record 走的是索引,查詢時間 3 秒不到。那麼咱們有沒有什麼辦法拿掉這個IF函數呢?

    使用 CASE...WHEN....THEN 代替 IF

SELECT MAX(id) mId,login_name,bill_no,create_time,
    CASE credit_type WHEN 1 THEN amount_before ELSE 0 END AS freeBefore,
    CASE credit_type WHEN 1 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 1 THEN amount_after ELSE 0 END AS freeAfter,
    CASE credit_type WHEN 2 THEN amount_before ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_after ELSE 0 END AS freeChange,
    CASE credit_type WHEN 3 THEN amount_before ELSE 0 END AS promotionBefore,
    CASE credit_type WHEN 3 THEN amount_change ELSE 0 END AS promotionChange,
    CASE credit_type WHEN 3 THEN amount_after ELSE 0 END AS promotionAfter
FROM t_customer_credit_record
GROUP BY bill_no,login_name,create_time
ORDER BY mId desc
LIMIT 0, 10;

SELECT MIN(tcd.channal) channal, MAX(tccr.id) mId,tccr.login_name,tccr.bill_no,tccr.create_time,
    CASE credit_type WHEN 1 THEN amount_before ELSE 0 END AS freeBefore,
    CASE credit_type WHEN 1 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 1 THEN amount_after ELSE 0 END AS freeAfter,
    CASE credit_type WHEN 2 THEN amount_before ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_after ELSE 0 END AS freeChange,
    CASE credit_type WHEN 3 THEN amount_before ELSE 0 END AS promotionBefore,
    CASE credit_type WHEN 3 THEN amount_change ELSE 0 END AS promotionChange,
    CASE credit_type WHEN 3 THEN amount_after ELSE 0 END AS promotionAfter
FROM t_customer_credit_record tccr
LEFT JOIN t_customer_deposit tcd ON tccr.bill_no = tcd.bill_no
GROUP BY tccr.bill_no,tccr.login_name,tccr.create_time
ORDER BY mId desc
LIMIT 0, 10;
View Code

      咱們能夠看到,執行時間與 IF 所差無幾,執行計劃也是與 IF 的一致,這也就反映出不是 IF的問題,應該是 GROUP BY 的問題。咱們用 GROUP BY 結合 IF(或 CASE...WHEN....THEN),就是爲了將 3 條額度記錄合併成一條、行轉列以後輸出咱們想要的結果,那有沒有不用 GROUP BY、又能實現咱們需求的方式了?

    自聯代替 GROUP BY

      咱們再仔細琢磨下這個需求,咋一看,確實須要行轉列,那麼就須要用到 GROUP BY,那麼效率也就低了,這彷佛是無解了? 真的非要行轉列嗎,假設咱們將額度記錄拆分紅 3 張表:一張表只存自由資金的額度變化、一張表只存凍結資金的額度變化、一張表只存優惠的額度變化,這樣是否是隻須要聯表查而不要用 GROUP BY 來進行行轉列了? 有小夥伴有可能會問:t_custmor_credit_record 表已經定了,數據都跑了很多了,再將其進行拆分,既要改表(同時還要遷移數據),還要改代碼,工程量會很大! 咱們換個角度來看 t_custmor_credit_record ,目前它是 3 中額度記錄的一個總和表,咱們能不能從它的身上作文章,變化出咱們想要的那 3 張表,而後進行聯表查詢呢? 確定能夠的,相似以下

-- 自由資金額度記錄表
SELECT * FROM t_customer_credit_record WHERE credit_type = 1;
-- 凍結資金額度記錄表
SELECT * FROM t_customer_credit_record WHERE credit_type = 2;
-- 優惠額度記錄表
SELECT * FROM t_customer_credit_record WHERE credit_type = 3;

      接下來的 SQL 怎麼寫,我想你們都知道了吧,自聯就好了,寫法有不少種,常見的寫法有以下 4 種

-- 不用group by,作法1, 我的比較推薦, 但此種方式不支持存款表的過濾條件
SELECT d.channal,a.amount_before AS freeBefore,a.amount_change AS freeChange, a.amount_after freeAfter,
    b.amount_before AS freezeBefore,b.amount_change AS freezeChange, b.amount_after freezeAfter,
    c.amount_before AS promotionBefore,c.amount_change AS promotionChange, c.amount_after promotionAfter
FROM (
    SELECT * FROM t_customer_credit_record WHERE credit_type = 1 ORDER BY id DESC LIMIT 0, 10
) a
LEFT JOIN t_customer_credit_record b ON a.bill_no = b.bill_no AND b.credit_type = 2
LEFT JOIN t_customer_credit_record c ON a.bill_no = c.bill_no AND c.credit_type = 3
LEFT JOIN t_customer_deposit d ON a.bill_no = d.bill_no;

-- 不用group by,作法2, 此種方式支持存款表的過濾條件
SELECT a.channal,a.amount_before AS freeBefore,a.amount_change AS freeChange, a.amount_after freeAfter,
    b.amount_before AS freezeBefore,b.amount_change AS freezeChange, b.amount_after freezeAfter,
    c.amount_before AS promotionBefore,c.amount_change AS promotionChange, c.amount_after promotionAfter
FROM (
    SELECT r.*,d.channal FROM t_customer_credit_record r LEFT JOIN t_customer_deposit d ON r.bill_no = d.bill_no
    WHERE r.credit_type = 1 ORDER BY r.id DESC LIMIT 0, 10
) a
LEFT JOIN t_customer_credit_record b ON a.bill_no = b.bill_no AND b.credit_type = 2
LEFT JOIN t_customer_credit_record c ON a.bill_no = c.bill_no AND c.credit_type = 3;

-- 不用group by,作法3, 這是最容易想到的方法
SELECT d.channal,a.amount_before AS freeBefore,a.amount_change AS freeChange, a.amount_after freeAfter,
    b.amount_before AS freezeBefore,b.amount_change AS freezeChange, b.amount_after freezeAfter,
    c.amount_before AS promotionBefore,c.amount_change AS promotionChange, c.amount_after promotionAfter
FROM t_customer_credit_record a
LEFT JOIN t_customer_credit_record b ON a.bill_no = b.bill_no
LEFT JOIN t_customer_credit_record c ON a.bill_no = c.bill_no
LEFT JOIN t_customer_deposit d ON a.bill_no = d.bill_no
WHERE a.credit_type = 1 AND b.credit_type = 2 AND c.credit_type = 3 
ORDER BY a.id DESC LIMIT 0, 10;

-- 不用group by,作法4
SELECT d.channal,a.amount_before AS freeBefore,a.amount_change AS freeChange, a.amount_after freeAfter,
    b.amount_before AS freezeBefore,b.amount_change AS freezeChange, b.amount_after freezeAfter,
    c.amount_before AS promotionBefore,c.amount_change AS promotionChange, c.amount_after promotionAfter
FROM t_customer_credit_record a
LEFT JOIN t_customer_credit_record b ON a.bill_no = b.bill_no AND b.credit_type = 2
LEFT JOIN t_customer_credit_record c ON a.bill_no = c.bill_no AND c.credit_type = 3
LEFT JOIN t_customer_deposit d ON a.bill_no = d.bill_no
WHERE a.credit_type = 1 
ORDER BY a.id DESC LIMIT 0, 10;
View Code

      執行結果以下

      就目前的數據量而言,4 種寫法的效率同樣,可是數據量再往上走,它們以前仍是有性能差異的,你們能夠仔細看看這 4 個 SQL 的執行計劃,它們之間仍是有區別的。最終個人項目中採用的是第一種寫法

    表從新設計

      咱們回過頭去看看 t_customer_credit 和 t_custmor_credit_record,是否真的有必要用 3 條記錄來存放顧客的 3 種額度,一條記錄將用戶的 3 種額度都記錄下來不是更好嗎? 以下所示

-- 自認爲更好的表設計
DROP TABLE IF EXISTS t_customer_credit_plus;
CREATE TABLE t_customer_credit_plus (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主鍵',
  login_name VARCHAR(50) NOT NULL COMMENT '登陸名',
    free_amount DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '自由資金額度',
    freeze_amount DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '凍結資金額度',
    promotion_amount DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '優惠資金額度',
    create_by VARCHAR(50) NOT NULL COMMENT '建立者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '建立時間',
    update_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP COMMENT '建立時間',
    update_by VARCHAR(50) NOT NULL COMMENT '修改者',
  PRIMARY KEY (id),
  UNIQUE KEY `uk_login_name` (`login_name`)
);
DROP TABLE IF EXISTS t_customer_credit_record_plus;
CREATE TABLE t_customer_credit_record_plus (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主鍵',
  login_name VARCHAR(50) NOT NULL COMMENT '登陸名',
    bill_no VARCHAR(50) NOT NULL COMMENT '訂單號',
    free_amount_before DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '自由資金前額度值',
    free_amount_change DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '自由資金前額度變化值',
    free_amount_after DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '自由資金先後額度值',
    freeze_amount_before DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '凍結資金前額度值',
    freeze_amount_change DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '凍結資金額度變化值',
    freeze_amount_after DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '凍結資金後額度值',
    promotion_amount_before DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '優惠前額度值',
    promotion_amount_change DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '優惠額度變化值',
    promotion_amount_after DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '優惠後額度值',
    create_by VARCHAR(50) NOT NULL COMMENT '建立者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '建立時間',
    remark VARCHAR(500) NOT NULL DEFAULT '' COMMENT '備註',
  PRIMARY KEY (id),
  UNIQUE KEY `uk_unique` (`bill_no`,`login_name`,`create_time`)
);
View Code

      我想不少人都會有相同的感受吧,可是從拓展性出發,分 3 條記錄的作法更好,爲何呢? 若是後續新增 1 種或多種另外的額度類型,上述的 1 條記錄的表設計就須要新增字段來適配了, 可是 3 條記錄的作法,只須要拓展credit_type的值就行了,表無需改動。各有利弊,如何選擇,須要團隊協商以後作出最好的選擇。

      最終項目中採用的仍是 3 條記錄存放 3 個額度的方式,沒有采用我說的;緣由是:你們都認爲效率影響不大,也容易理解,關鍵是拓展性很好,後續很方便就能加入新的額度類型。

    業務上的協調

      最後咱們再回到需求上來,這個 存款渠道 真的有必要顯示在額度記錄嗎?

      一、對公司來講,存款記錄越多,那確定是越好,但咱們從實際出發,存款記錄在額度記錄中佔的比例大嗎,這個相信你們也都能想象獲得,比例很是低,可能 100 條記錄中會有 1 條;

      二、原本就有單獨的存款頁面展現顧客的存款,去專門的存款記錄頁面看豈不是更直觀?

      最後和產品討論,還真把這一列給拿掉了,那麼咱們也就不須要關聯存款表來查了,SQL 更簡單,效率也更高了!

總結

  一、Oracle 的 DECODE 與 Mysql 中的 IF 都是 CASE 表達式的簡化函數,具備特殊性(只有 Oracle 才認識 DECODE、只有 Mysql 才認識 IF),而不是標準的 SQL,CASE 表達式是標準 SQL 所認可的功能;

  二、SQL 行轉列,每每是 GROUP BY 配合聚合函數(SUM、MAX、MIN等)來實現,固然也包括 IF 和 CASE...WHEN....THEN; 

  三、索引是提升查詢效率的最有效的、也是最經常使用的方式,咱們對查詢的優化都要往索引上靠,EXPLAIN 能夠查看SQL的執行計劃,咱們能夠從中獲取SQL優化的提示;

  四、必定要結合業務來寫出高效的SQL

    可能不少小夥伴會有這樣的不滿:上述的 3 個額度的例子有點特殊,不具有通用性,上述高效的SQL也只是在你(樓主)的項目中有效。你說的對,可是咱們要知道,技術自己就是用來服務業務的,脫離了業務,技術有什麼實際意義? 可是咱們回過頭去細看,我舉的例子真的就特殊到獨一無二? 我想還算比較通用吧,仍是能套用不少場景的。

  五、要勇於質疑需求

    雖然大多數時候產品都考慮比較周到,但也不能徹底保證他沒有不犯迷糊的時候吧。有疑惑就向產品問清楚,咱們實現的也更快、更準確。必定不要對疑問藏着掖着,以我親身經從來講,不少時候開發認爲的都是對的,若是藏着掖着,那你就準備返工改爲你以前認爲的那樣吧!

相關文章
相關標籤/搜索