mysql查詢優化--臨時表和文件排序（Using temporary; Using filesort問題解決）

先看一段sql：

<span style="font-size:18px;">SELECT
    *
FROM
    rank_user AS rankUser
LEFT JOIN rank_user_level AS userLevel ON rankUser.id = userLevel.user_id
LEFT JOIN rank_product AS product ON userLevel.new_level = product.level_id
LEFT JOIN rank_product_fee AS fee ON userLevel.fee_id = fee.fee_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
     rankUser.create_time DESC
LIMIT 10 OFFSET 0</span>

<span style="font-size:18px;">SELECT
	*
FROM
	rank_user AS rankUser
LEFT JOIN rank_user_level AS userLevel ON rankUser.id = userLevel.user_id
LEFT JOIN rank_product AS product ON userLevel.new_level = product.level_id
LEFT JOIN rank_product_fee AS fee ON userLevel.fee_id = fee.fee_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
	 rankUser.create_time DESC
LIMIT 10 OFFSET 0</span>

介紹一下這段sql的表的構成：一張主表：rank_user；兩張跟rank_user直接關聯（多張表通過同一字段最好是主鍵進行關聯）的表：rank_user_level ，rank_user_login_stat ；兩張跟rank_user非直接關聯的表：rank_product ，rank_product_fee 。這段sql看似簡單，但是執行時間卻很長，我們來看一下執行計划：

執行時間1.45s，可以看到，這段不僅僅掃描全表，而且使用了臨時表，進行了文件排序。

為了找到原因，我們把排序去掉看一下：

SELECT
    *
FROM
    rank_user AS rankUser
LEFT JOIN rank_user_level AS userLevel ON rankUser.id = userLevel.user_id
LEFT JOIN rank_product AS product ON userLevel.new_level = product.level_id
LEFT JOIN rank_product_fee AS fee ON userLevel.fee_id = fee.fee_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
-- ORDER BY
    -- rankUser.create_time DESC
LIMIT 10 OFFSET 0

SELECT
	*
FROM
	rank_user AS rankUser
LEFT JOIN rank_user_level AS userLevel ON rankUser.id = userLevel.user_id
LEFT JOIN rank_product AS product ON userLevel.new_level = product.level_id
LEFT JOIN rank_product_fee AS fee ON userLevel.fee_id = fee.fee_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
-- ORDER BY
	-- rankUser.create_time DESC
LIMIT 10 OFFSET 0

執行時間0.015s，掃描行數67452，果然是排序惹的禍。但是僅僅是排序惹的禍嗎？別忘了這里有兩張非直接關聯的表，這樣的查詢，如果有查詢條件或者排序分組的時候往往都需要創建臨時表（這個沒有辦法，想想也知道）。為了驗證這個觀點，我們把兩張非直接關聯的表去掉看一下：

SELECT
    *
FROM
    rank_user AS rankUser
LEFT JOIN rank_user_level AS userLevel ON rankUser.id = userLevel.user_id
-- LEFT JOIN rank_product AS product ON userLevel.new_level = product.level_id
-- LEFT JOIN rank_product_fee AS fee ON userLevel.fee_id = fee.fee_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
    rankUser.create_time DESC
LIMIT 10 OFFSET 0

SELECT
	*
FROM
	rank_user AS rankUser
LEFT JOIN rank_user_level AS userLevel ON rankUser.id = userLevel.user_id
-- LEFT JOIN rank_product AS product ON userLevel.new_level = product.level_id
-- LEFT JOIN rank_product_fee AS fee ON userLevel.fee_id = fee.fee_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
	rankUser.create_time DESC
LIMIT 10 OFFSET 0

執行時間0.003s,掃描行數10，屌爆了有木有，mysql多表直接關聯在沒有其他篩選條件的情況下，查詢速度大大提升，而且排序可以使用create_time這個索引，直接取到前十條。

到了這里，我想大家應該已經明白第一條sql查詢時間很長的原因了：多表非直接關聯的前提下還要排序。mysql查詢往往最需要優化的地方就是臨時表和文件排序了。這里總結一下教訓：

1.mysql查詢存在直接關聯和非直接關聯的問題，這兩種查詢效率差別很大；

2.mysql排序盡量使用索引；

3.mysql多表關聯left join其他表的時候，如果以其他表的字段作為查詢條件都會產生臨時表；

4.mysql在非直接關聯的基礎上進行排序會很慢，需要進行優化；

知道了問題，我們就好優化了，這里我給出了兩種方案：

第一種（子查詢，適合子查詢部分不作為查詢條件）：

SELECT
            rankUser.id, rankUser.qq, rankUser.phone, rankUser.regip, rankUser.channel, rankUser.create_time, rankUser.qudao_key, rankUser.qq_openid, rankUser.wechat_openid,
            userLevel.recommend_count,userLevel.end_time,userLevel.new_level,userLevel.`level`,userLevel.new_recommend_count,userLevel.`is_limited`,
            (case when userLevel.new_level > 1 then 1 else 0 end) is_official_user,
            (select product_name from rank_product where level_id = userLevel.new_level) product_name,
            (select period from rank_product_fee where fee_id = userLevel.fee_id) period,
            userLoginInfo.last_login, userLoginInfo.login_count, userLoginInfo.login_seconds
        FROM rank_user AS rankUser
        LEFT JOIN rank_user_level as userLevel on userLevel.user_id=rankUser.id
        LEFT JOIN rank_user_login_stat as userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
    rankUser.create_time DESC
LIMIT 10 OFFSET 0

SELECT
            rankUser.id, rankUser.qq, rankUser.phone, rankUser.regip, rankUser.channel, rankUser.create_time, rankUser.qudao_key, rankUser.qq_openid, rankUser.wechat_openid,
            userLevel.recommend_count,userLevel.end_time,userLevel.new_level,userLevel.`level`,userLevel.new_recommend_count,userLevel.`is_limited`,
            (case when userLevel.new_level > 1 then 1 else 0 end) is_official_user,
            (select product_name from rank_product where level_id = userLevel.new_level) product_name,
            (select period from rank_product_fee where fee_id = userLevel.fee_id) period,
            userLoginInfo.last_login, userLoginInfo.login_count, userLoginInfo.login_seconds
        FROM rank_user AS rankUser
        LEFT JOIN rank_user_level as userLevel on userLevel.user_id=rankUser.id
        LEFT JOIN rank_user_login_stat as userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
	rankUser.create_time DESC
LIMIT 10 OFFSET 0

第二種（非直接關聯轉變成直接關聯，這個要根據業務來定，我這里rank_product和rank_product_fee兩張表只是為了查詢rank_user_level表中的產品和產品費用信息，而rank_user_level是一張直接關聯的表，故這里可以先將這三張表進行合並，然后再和rank_user表進行聯合查詢）：

SELECT
    *
FROM
    rank_user AS rankUser
LEFT JOIN (
        select
        l.*,p.product_name,f.period
        from
        rank_user_level l,rank_product p,rank_product_fee f
        where
        l.new_level = p.level_id
        and l.fee_id = f.fee_id
) AS userLevel ON rankUser.id = userLevel.user_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
    rankUser.create_time DESC
LIMIT 10 OFFSET 0

SELECT
	*
FROM
	rank_user AS rankUser
LEFT JOIN (
		select 
		l.*,p.product_name,f.period 
		from 
		rank_user_level l,rank_product p,rank_product_fee f
		where 
		l.new_level = p.level_id 
		and l.fee_id = f.fee_id
) AS userLevel ON rankUser.id = userLevel.user_id
LEFT JOIN rank_user_login_stat AS userLoginInfo ON rankUser.id = userLoginInfo.user_id
ORDER BY
	rankUser.create_time DESC
LIMIT 10 OFFSET 0