根据查询结果制作可子查询的 UNION ALL

Question

数据

我有几个这样的表：

CREATE TABLE cycles (
  `cycle` varchar(6) COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `cycle_type` varchar(140) COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `start` date DEFAULT NULL,
  `end` date DEFAULT NULL
);

CREATE TABLE rsvn (
  `str` varchar(140) COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `start_date` date DEFAULT NULL,
  `end_date` date DEFAULT NULL
);

INSERT INTO `cycles` (`cycle`, `cycle_type`, `start`, `end`) values
('202013', 'a', '2021-01-04', '2021-01-31'),
('202013', 'b', '2021-01-04', '2021-01-31'),
('202101', 'a', '2021-01-04', '2021-01-31'),
('202101', 'b', '2021-01-04', '2021-01-31'),
('202102', 'a', '2021-02-01', '2021-02-28'),
('202102', 'b', '2021-02-01', '2021-02-28'),
('202103', 'a', '2021-03-01', '2021-03-28'),
('202103', 'b', '2021-03-01', '2021-03-28');

INSERT INTO `rsvn` (str, start_date, end_date) values
('STR01367', '2020-12-07', '2020-06-21'),
('STR00759', '2020-12-07', '2021-04-25'),
('STR01367', '2021-01-04', '2021-09-12'),
('STR01367', '2021-06-21', '2022-02-27');

期望的结果

对于任何给定的周期，我想计算 str 跨周期的出现次数。 所以在周期 2108 - 2108（一个周期）之间，我看到：

字符串	数数
STR01367	1
STR00759	1

从 2108 年到 2109 年（两个周期），我看到：

字符串	数数
STR01367	2
STR00759	1

我试过的

我试图弄清楚如何动态获得这些结果。 我在 UNION ALL 查询（每个周期一个查询）之外看不到任何选项，所以我尝试编写一个 PROCEDURE。 但是，这不起作用，因为我想对查询结果进行后处理，而且我不相信您可以在 CTE 或子查询中使用 PROCEDURE 的结果。

我的程序（有效，不能在子查询中包含结果，例如SELECT * FROM call count_cycles (?)）：

CREATE PROCEDURE `count_cycles`(start_cycle CHAR(6), end_cycle CHAR(6))
BEGIN
    SET @cycles := (
        SELECT CONCAT('WITH installed_cycles_count AS (',
            GROUP_CONCAT(
                CONCAT('
        SELECT rsvn.str, 1 AS installed_cycles
        FROM rsvn
        WHERE "', `cy`.`start`, '" BETWEEN rsvn.start_date AND COALESCE(rsvn.end_date, "9999-01-01")
           OR "', `cy`.`end`, '" BETWEEN rsvn.start_date AND COALESCE(rsvn.end_date, "9999-01-01")
        GROUP BY rsvn.str
    '
                )
                 SEPARATOR ' UNION ALL '
            ),
    ')

    SELECT
             store.chain AS "Chain"
            ,store.division AS "Division"
            ,dividers_store AS "Store"
            ,SUM(installed_cycles) AS "Installed Cycles"
    FROM installed_cycles_count r
    LEFT JOIN store ON store.name = r.dividers_store
    GROUP BY dividers_store
    ORDER BY chain, division, dividers_store, installed_cycles'
        )
        FROM cycles `cy`
        WHERE `cy`.`cycle_type` = 'Ad Cycle'
            AND `cy`.`cycle` >= CONCAT('20', RIGHT(start_cycle, 4))
            AND `cy`.`cycle` <= CONCAT('20', RIGHT(end_cycle, 4))
        GROUP BY `cy`.`cycle_type`
    );

    EXECUTE IMMEDIATE @cycles;
END

或者，我尝试使用递归查询通过增加我的周期来获得我的结果。 这给了我想要的周期：

WITH RECURSIVE xyz AS (
    SELECT cy.`cycle`, cy.`start`, cy.`end`
    FROM cycles cy
    WHERE cycle_type = 'Ad Cycle'
    AND `cycle` = '202101'

    UNION ALL
    
    SELECT cy.`cycle`, cy.`start`, cy.`end`
    FROM xyz
    JOIN cycles cy
        ON cy.`cycle` = increment_cycle(xyz.`cycle`, 1)
        AND cy.`cycle_type` = 'Ad Cycle'
    WHERE cy.`cycle` <= '202110'
)
SELECT * FROM xyz;

但是当我在预订表中添加时，我无法让它工作：无限循环？

WITH RECURSIVE xyz AS (
    SELECT cy.`cycle`, 'dr.dividers_store', 1 AS installed_cycles
    FROM cycles cy
    LEFT JOIN rsvn dr
        ON cy.`start` BETWEEN dr.start_date AND COALESCE(dr.end_date, "9999-01-01")
            OR cy.`end` BETWEEN dr.start_date AND COALESCE(dr.end_date, "9999-01-01")
    WHERE cy.`cycle_type` = 'Ad Cycle'
        AND cy.`cycle` = '202101'

    UNION ALL

    SELECT cy.`cycle`, 'dr.dividers_store', 1 AS installed_cycles
    FROM xyz
    JOIN cycles cy
        ON cy.`cycle` = increment_cycle(xyz.`cycle`, 1)
        AND cy.`cycle_type` = 'Ad Cycle'
    LEFT JOIN rsvn dr
        ON cy.`start` BETWEEN dr.start_date AND COALESCE(dr.end_date, "9999-01-01")
            OR cy.`end` BETWEEN dr.start_date AND COALESCE(dr.end_date, "9999-01-01")
    WHERE cy.`cycle` <= '202102'
)
SELECT * FROM xyz

我必须有哪些选项才能获得我需要的结果，以便我可以在 CTE 或子查询中使用它们？

Answer 1

我正在寻找的结果很容易通过两阶段分组获得。 像这样的东西：

WITH sbc AS (
    SELECT cy.`cycle`, dr.str, 1 AS 'count'
    FROM cycles cy
    LEFT JOIN rsvn dr
        ON cy.`start` BETWEEN dr.start_date AND dr.end_date
            OR cy.`end` BETWEEN dr.start_date AND dr.end_date
    WHERE cy.`cycle_type` = 'Ad Cycle'
        AND cy.`cycle` BETWEEN '202201' AND '202205'
    GROUP BY cy.`cycle`, dr.str
    ORDER BY dr.str, cy.`cycle`
)
SELECT `cycle`, str, SUM(`count`) as `count`
FROM sbc
GROUP BY str

CTE 每个周期每个 rsvn 产生一个结果。 之后所需要做的就是按商店分组并计算出现次数。

除了更简单之外，我怀疑这个查询比我提出问题时坚持的联合概念更快，因为除其他外，服务器不需要对多个分组查询执行联合。 但是，我不明白 MariaDB 如何优化此类查询，虽然我很好奇，但我没有时间运行基准测试来找出答案。