如何在 Postgresql 窗口函数的 PARTITION BY 中包含当前行

Question

我正在尝试执行以下操作；假设我想在给定条件下将表分区为两个分区：

SELECT
    userid,
    ARRAY_AGG(userid) OVER (
        PARTITION BY userid > 100
    ) arr,
    AVG(userid) OVER (
        PARTITION BY userid > 100
    ) avg
FROM users;

我会得到这个：

 userid |                            arr                            |         avg          
--------+-----------------------------------------------------------+----------------------
     46 | {46,23,69,92}                                             |  57.5000000000000000
     23 | {46,23,69,92}                                             |  57.5000000000000000
     69 | {46,23,69,92}                                             |  57.5000000000000000
     92 | {46,23,69,92}                                             |  57.5000000000000000
    552 | {552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 629.2142857142857143
    ... | ...                                                       | ...
    529 | {552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 629.2142857142857143

一切都好，但是如果相反，对于 100 的用户 ID 中：

SELECT
    userid,
    CASE WHEN userid > 100
    THEN ARRAY_AGG(userid) OVER (
        PARTITION BY userid > 100
    )
    ELSE ARRAY_AGG(userid) OVER (
        PARTITION BY userid -- OR userid > 100
        -- PARTITION BY userid > 100 OR CURRENT_ROW
        -- PARTITION BY userid > 100 OR userid = LAG(userid, 0) OVER ()
    )
    END arr
    CASE WHEN userid > 100
    THEN AVG(userid) OVER (
        PARTITION BY userid > 100
    )
    ELSE AVG(userid) OVER (
        PARTITION BY userid -- OR userid > 100
        -- PARTITION BY userid > 100 OR CURRENT_ROW
        -- PARTITION BY userid > 100 OR userid = LAG(userid, 0) OVER ()
    )
    END avg
FROM users;

上面所有注释的代码都是我一直在做的各种尝试。 我所拥有的最好的要么只是没有大于 100 的用户 ID，要么是所有用户 ID：

 userid |                            arr                            |         avg          
--------+-----------------------------------------------------------+----------------------
     23 | {23}                                                      |  23.0000000000000000
     46 | {46}                                                      |  46.0000000000000000
     69 | {69}                                                      |  69.0000000000000000
     92 | {92}                                                      |  92.0000000000000000
    552 | {552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 629.2142857142857143
    ... | ...                                                       | ...
    529 | {552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 629.2142857142857143

有什么方法可以做我正在寻找的东西吗？我也尽量不使用 CTE，因为实际代码中的技术债务太多，仅使用 WITH 来适应它需要相当长的时间。

要清楚，这是预期的结果：

 userid |                             arr                              |         avg
--------+--------------------------------------------------------------|----------------------
     23 | {23,552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 588.6000000000000000
     46 | {46,552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 590.1333333333333334
     69 | {69,552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 591.6666666666666667
     92 | {92,552,506,575,621,644,667,690,759,713,782,828,460,483,529} | 593.2000000000000000
    552 | {552,506,575,621,644,667,690,759,713,782,828,460,483,529}    | 629.2142857142857143
    ... | ...                                                          | ...
    529 | {552,506,575,621,644,667,690,759,713,782,828,460,483,529}    | 629.2142857142857143

这是我一直在寻找的潜在未来内容的参考：nested window functions（但目前尚未实现，截至 Postgresql-11）

编辑：最后但同样重要的是，条件是占位符！它可能与用户 ID 相关，也可能不相关，这里只是为了举例，它可能是

CUME_DIST() OVER (
    PARTITION BY x -- OR CURRENT_USERID
)

Answer 1

这回答了问题的原始版本。

你似乎想要：

select (case when userid < 100
             then array_cat(array[userid],
                            array_agg(userid) filter (where userid > 100) over ()
             else array_agg(userid) filter (where userid > 100) over ()
        end)