SQL百分比优雅

Question

我刚刚开始熟悉SQL，我发现语言很简单，但也有时非常混乱。例如借此声明

SELECT transaction.ID, 
     dt, 
     bundles.package_name as "Package Name", 
     amount_usd, 
     unit_sales, 
CASE 
    WHEN sum(amount_usd) = 0 THEN 0 
    ELSE round(amount_usd*100/ (SELECT sum(amount_usd) FROM transaction 
      WHERE (transaction.code = 'AA') 
      AND (transaction.country_code = 'US') 
      AND (dt BETWEEN '2016-01-01' and '2016-01-04')),2) 
END as percentage 

FROM transaction 
    JOIN bundles 
    ON transaction.package_id= cast(bundles.id as INT) 

WHERE (transaction.code = 'AA') 
AND (transaction.country_code = 'US') 
AND (dt BETWEEN '2016-01-01' and '2016-01-04') 

GROUP BY transaction.ID, dt, "Package Name" 
ORDER BY transaction.ID 
limit 1000 

我发现浪费一点点不得不重复SELECT加上只是正确计算的百分比。有没有更优雅的解决方案来避免这种情况？

---

下面是我如何修复它。

如果你想在总量的百分数，你不一定需要嵌套

SELECT t.ID, dt, b.package_name as "Package Name", 
     SUM(amount_usd) as amount_usd, SUM(unit_sales) as unit_sales, 
     SUM(amount_usd) * 1.0/SUM(amount_usd) OVER() as ratio 
FROM transaction t JOIN 
    bundles b 
    ON t.package_id = cast(b.id as INT) 
WHERE (t.code = 'AA') AND 
     (t.country_code = 'US') AND 
     (dt BETWEEN '2016-01-01' and '2016-01-04') 
GROUP BY t.ID, dt, "Package Name" 
ORDER BY t.ID 
LIMIT 1000; 

如果你想收入超过每天总的比你做

SELECT t.ID, dt, b.package_name as "Package Name", 
     SUM(amount_usd) as amount_usd, SUM(unit_sales) as unit_sales, 
     SUM(amount_usd) * 1.0/SUM(amount_usd) OVER (PARTITION BY dt) as ratio 
FROM transaction t JOIN 
    bundles b 
    ON t.package_id = cast(b.id as INT) 
WHERE (t.code = 'AA') AND 
     (t.country_code = 'US') AND 
     (dt BETWEEN '2016-01-01' and '2016-01-04') 
GROUP BY t.ID, dt, "Package Name" 
ORDER BY t.ID 
LIMIT 1000; 

Answer 1

最SQL的方言支持ANSI标准窗口函数。另外，格式化通常在应用程序中稍后处理。

所以，这通常被写为：

SELECT t.ID, dt, b.package_name as "Package Name", 
     SUM(amount_usd) as amount_usd, SUM(unit_sales) as unit_sales, 
     SUM(amount_usd) * 1.0/NULLIF(SUM(SUM(amount_usd)) OVER()) as ratio 
FROM transaction t JOIN 
    bundles b 
    ON t.package_id = cast(b.id as INT) 
WHERE (t.code = 'AA') AND 
     (t.country_code = 'US') AND 
     (dt BETWEEN '2016-01-01' and '2016-01-04') 
GROUP BY t.ID, dt, "Package Name" 
ORDER BY t.ID 
LIMIT 1000; 

在您的查询的其他问题：

• 所有符合条件的列名时查询有多个表。
• 使用表缩写作为表别名 - 写和读更容易。
• 不同表格上的JOIN键应该是相同的类型。事实上，他们应该被宣布为适当的外键关系。
• 您错过了SELECT中的某些聚合函数。确保只有未聚集的列是GROUP BY密钥。

如果您的数据库供应商不支持ANSI标准功能，那么您应该采取这些措施。