在 Oracle ROLLUP 中结合 MAX 和 SUM

Question

这是我的 SQL：

WITH source1 AS (
   SELECT 'Fruit' foodtype, 'Apple'  food, 20 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Apple'  food, 30 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Grape'  food, 1  weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Carrot' food, 40 weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Leek'   food, 20 weight FROM dual
)
SELECT grouping(food) lv, foodtype, food, max(weight) weight
FROM   source1
GROUP BY foodtype, ROLLUP (food);

输出如下：

LV FOODTYPE FOOD   WEIGHT
-- -------- ------ ------
0  Veg      Leek       20
0  Veg      Carrot     40
1  Veg                 40
0  Fruit    Apple      30
0  Fruit    Grape       1
1  Fruit               30

我期待它看起来像这样：

LV FOODTYPE FOOD   WEIGHT
-- -------- ------ ------
0  Veg      Leek       20
0  Veg      Carrot     40
1  Veg                 60
0  Fruit    Apple      30
0  Fruit    Grape       1
1  Fruit               31

换句话说，我希望 rollup 总结每种食物的最大重量，而不是取食物类型类别中所有最大值中的最大值。

我确实有各种各样的解决方案，但这意味着必须添加一层额外的 SQL 语句嵌套：

WITH source1 AS (
   SELECT 'Fruit' foodtype, 'Apple'  food, 20 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Apple'  food, 30 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Grape'  food, 1  weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Carrot' food, 40 weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Leek'   food, 20 weight FROM dual
), source_grp AS (
   SELECT s.foodtype, s.food, max(s.weight) max_weight
   FROM   source1 s
   GROUP BY foodtype, food
)
SELECT grouping(g.food) lv, g.foodtype, g.food, sum(g.max_weight) weight
FROM   source_grp g
GROUP BY g.foodtype, ROLLUP (g.food);

有没有办法在没有额外嵌套的情况下做到这一点？

当然，从我的实际情况来看，这个例子已经大大简化了，这就是为什么我试图找到一种方法来减少代码行数。从长远来看，将一条 SQL 语句减少 60 行代码可显着简化其维护。

Answer 1

几天后重温。可以这样做：

WITH source1 AS (
   SELECT 'Fruit' foodtype, 'Apple'  food, 20 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Apple'  food, 30 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Grape'  food, 1  weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Carrot' food, 40 weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Leek'   food, 20 weight FROM dual
)
SELECT grouping(s.food) lv, s.foodtype, s.food,
       CASE WHEN grouping(s.food)=1 THEN
          sum(CASE WHEN grouping(s.food)=1 THEN 0 ELSE max(s.weight) END) OVER (PARTITION BY s.foodtype ORDER BY s.food)
       ELSE
          max(s.weight)
       END weight
FROM   source1 s
GROUP BY s.foodtype, ROLLUP (s.food)

说实话，我也不是 100% 肯定我喜欢这个答案。根据上下文和从维护的角度来看，CASE-WHEN 语句比多级 SELECT 更难理解。

Answer 2

你可以使用窗口函数结合CASE表达式来得到想要的结果：

Demo

WITH cte1 AS
(SELECT a.*, 
SUM(weight) OVER(PARTITION BY foodtype ORDER BY lv, food rows between unbounded preceding and 1 preceding) cum_sum
FROM table1 a)
SELECT lv, foodtype, food, 
CASE WHEN lv = 1 
     THEN cum_sum
     ELSE weight END AS weight
FROM cte1;

您可以使用联合，因此总选择将只有 2。

WITH source1 AS (
   SELECT 'Fruit' foodtype, 'Apple'  food, 20 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Apple'  food, 30 weight FROM dual UNION
   SELECT 'Fruit' foodtype, 'Grape'  food, 1  weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Carrot' food, 40 weight FROM dual UNION
   SELECT 'Veg'   foodtype, 'Leek'   food, 20 weight FROM dual
)
SELECT foodtype, food, max(weight) weight
FROM   source1
GROUP BY foodtype, food
UNION
(SELECT DISTINCT FOODTYPE, NULL, SUM(WEIGHT) TOTAL_WEIGHT  FROM SOURCE1 GROUP BY FOODTYPE);

Answer 3

您可以改用row_number()：

SELECT GROUPING(food) as lv, foodtype, food, SUM(weight) weight
FROM (SELECT s1.*, 
             ROW_NUMBER() OVER (PARTITION BY foodtype, food ORDER BY weight DESC) as seqnum
      FROM source1 s1
     ) s1
WHERE sequm = 1;
GROUP BY foodtype, ROLLUP (food);