背景：目前公司的报表计算需要20分钟，导致需要这么久的原因是，一次计算操作，执行了10万次+的update操作；看代码，有两个地方是这样写的，循环执行update，每个循环达到了4.6万次。

 

一，开始着手优化，第一个地方计算的是平均值，之前是根据report_date，trade_code这两个字段分组循环计算每一个平均值，现在我改成了一条SQL来执行；速度显著提高。

    SET @rownum := 0;
    INSERT INTO zy_ticket_industry_history_0914
    SELECT
    @rownum := @rownum +1 AS 'id',
    zb.report_date,zi.trade_code,
    (SELECT COUNT(trade_code) AS 'total'  FROM zy_ticket_info zin WHERE zin.trade_code = zi.trade_code GROUP BY zin.trade_code) AS company_count,
    CONVERT(MAX(zp.total_operating_income),DECIMAL(20,2)) AS 'business_income_top',
    (SELECT zzi.zq_jc AS 'business_income_top_name' FROM zy_ticket_profit zzp ,zy_ticket_info zzi WHERE zzp.zq_code=zzi.zq_code AND zi.trade_code=zzi.trade_code AND zzi.enable='Y' ORDER BY zzp.total_operating_income DESC LIMIT 1) AS 'business_income_top_name',
    CONVERT(AVG(zb.total_assets),DECIMAL(20,2)) AS 'total_assets_avg' ,
    CONVERT(AVG(zb.total_current_assets),DECIMAL(20,2)) AS 'current_assets_avg',
    CONVERT(AVG( zb.total_liabilit )/ AVG( zb.total_assets ),DECIMAL(20,2)) AS 'leverage_ratio_avg' ,
    CONVERT(AVG( zb.total_current_assets )/ AVG( zb.total_current_liabilit ),DECIMAL(20,2)) AS 'current_ratio_avg' ,
    CONVERT((AVG(zb.total_current_assets)-AVG(zb.net_inventory))/AVG(zb.total_current_assets ),DECIMAL(20,2)) AS 'quick_ratio_avg' ,
    CONVERT((AVG(zc.cash_increase)+AVG(zb.account_receivable))/AVG(zb.total_current_liabilit ),DECIMAL(20,2)) AS 'conserv_quick_ratio_avg' ,
    CONVERT(AVG(zc.cash_increase),DECIMAL(20,2)) AS 'cash_avg' ,
    CONVERT(AVG(zb.total_current_liabilit),DECIMAL(20,2)) AS 'current_liabilities_avg',
    CONVERT((AVG(zp.total_operating_income)-AVG(zp.operating_cost)-AVG(zp.income_tax_expense)),DECIMAL(20,2)) AS 'operating_cash_avg',
    CONVERT((AVG(zp.net_profit)-AVG(zp.no_operating_income)),DECIMAL(20,2)) AS 'kf_net_profit_avg',
    CONVERT((AVG(zb.total_current_assets)-AVG(zb.net_inventory))/AVG(zb.total_current_liabilit),DECIMAL(20,2)) AS 'roe_avg',
    CONVERT(AVG(zp.total_profit),DECIMAL(20,2)) AS 'sale_gross_profit_avg',
    CONVERT(AVG(zp.net_profit),DECIMAL(20,2)) AS 'sale_net_profit_avg',
    CONVERT(AVG(zp.operating_cost)/(AVG(zb.net_inventory)+AVG(zb2.net_inventory)),DECIMAL(20,2)) AS 'inventory_turnover_avg',
    CONVERT(AVG(zp.total_operating_income)/(AVG(zb.long_receivables)+AVG(zb2.long_receivables)),DECIMAL(20,2)) AS 'receive_turnover_avg',
    'Y' AS `enable`,
    NULL AS create_by,
    NOW() AS create_time,
    NULL AS update_by,
    NOW() AS update_time,
    NULL AS remark
    FROM zy_ticket_balance zb LEFT JOIN zy_ticket_info zi ON(zb.zq_code=zi.zq_code)
    LEFT JOIN zy_ticket_cash zc ON(zb.zq_code=zc.zq_code AND zb.report_date=zc.report_date)
    LEFT JOIN zy_ticket_profit zp ON( zb.zq_code=zp.zq_code AND zb.report_date=zp.report_date)
    LEFT JOIN zy_ticket_balance zb2 ON(zb.zq_code=zb2.zq_code AND zb.report_date= DATE_ADD(zb2.report_date,INTERVAL 1 YEAR))
    WHERE zb.zq_code=zi.zq_code AND zi.enable='Y'
    GROUP BY zi.`trade_code` ,zb.report_date
    ORDER BY report_date ASC,zi.trade_code ASC

 

第二个地方计算的是金额，排名，百分比，之前是根据zq_code，report_date，trade_code这两个字段分组循环计算的，基于前面那个，这个我也想通过一个SQL直接计算所有数据然后Insert；在SQL层面去优化；

而事实证明，想过很多种办法，都不行，SQL写出来，查询效率也很低，速度提高不明显，没有实际意义。最终的解决方案是动态校验zq_code，report_date，trade_code，只需执行16次update；

当然也可以更少，将操作表一致的进行合并，我为了SQL语句更清晰点，就分开写了。

（中间想通过SQL解决问题的时候，一到测试环境因为内存溢出，方案报废，而运营因为线上资源有限，不允许修改配置，然后才想到这个最终方案；其实应该一开始就这样做的，被第一步严重影响了思路::>_<::）

SQL：

　　　　 -- 将数据先计算出来，然后在动态判断zq_code,trade_code,report_date，在进行更新。
        UPDATE zy_ticket_report t1 INNER JOIN  (
        
        SELECT tt.*,ROUND((companyCount-totalAssetsRank)/companyCount*100,2) AS totalAssetsTranscend
          FROM (
          SELECT t.*, IF (@pa = t.tradeCode,@curRank := @curRank + 1,@curRank := 1) AS totalAssetsRank,@pa := t.tradeCode AS blTradeCode
          FROM
            ( SELECT zb.report_date AS reportDate,zi.zq_code AS zqCode,zi.trade_code AS tradeCode,zi.zq_jc AS zqJc,com.companyCount,zb.total_assets AS 'totalAssets'
            FROM zy_ticket_info zi JOIN zy_ticket_balance zb ON (zb.zq_code = zi.zq_code )
            JOIN (SELECT trade_code,COUNT(trade_code) AS 'companyCount'  FROM zy_ticket_info zin GROUP BY zin.trade_code) AS com ON(com.trade_code = zi.trade_code)
            WHERE zi.enable = 'Y'
            GROUP BY zb.report_date,zi.trade_code,zb.id
            ORDER BY zb.report_date,zi.trade_code,zb.total_assets DESC ) AS t,
            (SELECT @curRank := 0, @pa := '') b
            ) AS tt
             ) AS t2
        
        SET t1.total_assets=t2.totalAssets,t1.total_assets_rank=t2.totalAssetsRank, t1.total_assets_transcend=t2.totalAssetsTranscend,t1.update_time = NOW()
    
        WHERE t1.`zq_code` = t2.zqCode AND t1.`trade_code` = t2.tradeCode AND t1.`report_date` = t2.reportDate 

 

JAVA：

 

 

 

优化后，数据计算只需60秒左右。o(*￣▽￣*)ブ

 

 

 

记录下这次优化，学到的：

1，查询Map结构的对象集合，根据key获取整个对象。

   XML: 
　　<select id="getZys" resultMap="retMap">
        SELECT
          CONCAT_WS('-',t.zqCode,t.tradeCode,t.reportDate) AS mapKey,t.*,IF (@pa = t.tradeCode,@curRank := @curRank + 1,@curRank := 1) AS totalOperatingIncomeRank, @pa := t.tradeCode AS blTradeCode
        FROM
          (SELECT zp.report_date AS reportDate,zi.zq_code AS zqCode,zi.trade_code AS tradeCode,zi.zq_jc AS zqJc,com.companyCount,zp.total_operating_income AS 'totalOperatingIncome'
          FROM zy_ticket_info zi JOIN zy_ticket_profit zp ON (zp.zq_code = zi.zq_code)
          JOIN (SELECT trade_code,COUNT(trade_code) AS 'companyCount'  FROM zy_ticket_info zin GROUP BY zin.trade_code) AS com ON(com.trade_code = zi.trade_code)
          WHERE zi.enable = 'Y'
          GROUP BY zp.report_date,zi.trade_code,zp.id
          ORDER BY zp.report_date,zi.trade_code,zp.total_operating_income DESC) AS t,
          (SELECT @curRank := 0, @pa := '') b
    </select>


    <resultMap id="retMap" type="com.zy.ticket.model.vo.TicketReportVO">
        <result column="mapKey" property="mapKey" javaType="String"/>
        <result column="zqCode" property="zqCode" javaType="String"/>
        <result column="zqJc" property="zqJc" javaType="String"/>
        <result column="tradeName" property="tradeName" javaType="String"/>
        <result column="tradeCode" property="tradeCode" javaType="String"/>
        <result column="reportDate" property="reportDate" javaType="String"/>
        <result column="companyCount" property="companyCount" javaType="Integer"/>
        <result column="totalAssets" property="totalAssets" javaType="BigDecimal"/>
        <result column="totalAssetsRank" property="totalAssetsRank" javaType="Integer"/>
        <result column="totalAssetsTranscend" property="totalAssetsTranscend" javaType="BigDecimal"/>
    </resultMap>

  Mapper:   
　　@MapKey("mapKey")
    Map<String,TicketReportVO> getZys();

Service:
   Map<String,TicketReportVO> zysMap = zyTicketReportMapper.getZys();
   zysMap.get(vo.getZqCode()+"-"+vo.getTradeCode()+"-"+vo.getReportDate()).getLeverageRatio()

 

2，insert语句插入的字段越少，速度越快。以下4.6万条数据，只需4秒。

<insert id="baseInsert" parameterType="java.util.List">
        <selectKey resultType="java.lang.Long" keyProperty="id" order="AFTER">
            SELECT
            LAST_INSERT_ID()
        </selectKey>
        insert into zy_ticket_report
        (zq_code, zq_jc,trade_name, trade_code, report_date,company_count)
        values
        <foreach collection="list" item="item" separator=",">
            (#{item.zqCode}, #{item.zqJc},#{item.tradeName},
            #{item.tradeCode}, #{item.reportDate},#{item.companyCount})
        </foreach>

        ;COMMIT;
    </insert>

 

3，如果同一个表的insert和updat在一个事务里操作，在insert语句后写commit，手动进行数据库提交，速度会提高很多。

4，只有insert操作的时候，不需要去添加事务处理。