sql - 如何循环遍历数据并将前一批与 SQL Server 中的当前批进行比较？

Question

每次我们收到一批新数据时，我们都会将其加载到一个表中，该表存储我们之前收到的每个数据集。每个新批处理都有一个新的 batch_id。仅有的另外两列是 item_id 和 cost。我想构建一个报告，描述添加了多少新项目(基于 item_id)、删除了多少项目以及最新批处理中项目的净变化。考虑到这一点，我想将所有批处理相互比较，但只是为了让较高的 batch_id 比较它之前的 batch_id 。澄清一下，如果我有 3 个 batch_id，那么我希望第 3 批处理与第 2 批处理进行比较，第 2 批处理与第 1 批处理进行比较。第 1 批处理没有什么可比较的。我可以使用下面的脚本执行此操作，但仅限于我指定的批处理(结果中的第二行)，并且对于数百个批处理，这会让人感到 NumPy 。有没有办法使它在 SQL Server 中更加动态？我在下面提供了一个示例数据集和预期结果。

WITH b1
     AS (SELECT *
         FROM example_cumulative
         WHERE batch_id = 2),
     b2
     AS (SELECT *
         FROM example_cumulative
         WHERE batch_id = 3)
     SELECT SUM(CASE
                    WHEN b1.item_id IS NULL
                    THEN 1
                    ELSE 0
                END) [new items], 
            SUM(CASE
                    WHEN b2.item_id IS NULL
                    THEN 1
                    ELSE 0
                END) [removed items], 
            COUNT(b2.item_id) - COUNT(b1.item_id) [net change in item volume]
     FROM b1
          FULL OUTER JOIN b2 ON b1.item_id = b2.item_id;

预期结果

CREATE TABLE example_cumulative
(batch_id INT NOT NULL, 
 item_id  INT NOT NULL, 
 cost     FLOAT NOT NULL
);
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (1,10,100)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (1,11,100)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (1,12,106)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (1,13,142)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (1,14,152)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (1,15,154)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (2,10,110)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (2,11,140)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (2,12,100)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (2,16,100)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (3,10,110)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (3,11,100)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (3,16,195)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (3,17,102)
INSERT INTO example_cumulative (batch_id, item_id, cost) VALUES (3,18,101)

Answer 1

您可以通过动态加入连续的 batch_id 来扩展当前的 full join 逻辑，如下所示:

select
    coalesce(e1.batch_id, e0.batch_id + 1) batch_id,
    sum(case when e0.item_id is null then 1 else 0 end) new_items,
    sum(case when e1.item_id is null then 1 else 0 end) removed_items,
    count(e1.item_id) - count(e0.item_id) net_change
from example_cumulative e1
full join example_cumulative e0 
    on  e1.item_id  = e0.item_id
    and e1.batch_id = e0.batch_id + 1
where coalesce(e1.batch_id, e0.batch_id + 1) <= (select max(batch_id) from example_cumulative)
group by coalesce(e1.batch_id, e0.batch_id + 1)

where 子句用于防止查询返回一个额外的、不需要的记录，其 batch_id 大于最大可用值。

对于您的示例数据， this returns :

batch_id | new_items | removed_items | net_change
-------: | --------: | ------------: | ---------:
       1 |         6 |             0 |          6
       2 |         1 |             3 |         -2
       3 |         2 |             1 |          1