无法重置分区组(特别是 Window 函数和 PostgreSQL)
Posted
技术标签:
【中文标题】无法重置分区组(特别是 Window 函数和 PostgreSQL)【英文标题】:Unable to reset partition groups (Window functions and PostgreSQL specifically) 【发布时间】:2019-10-17 18:54:57 【问题描述】:我有一个像这样的简单数据集:
SELECT UNNEST(ARRAY['A', 'A', 'A', 'B', 'B', 'A', 'C', 'B']) AS customer_name, generate_series(8, 1, -1) AS order_time;
+-------------+------------+
| customer_id | order_time |
+-------------+------------+
| "A" | 8 |
+-------------+------------+
| "A" | 7 |
+-------------+------------+
| "A" | 6 |
+-------------+------------+
| "B" | 5 |
+-------------+------------+
| "B" | 4 |
+-------------+------------+
| "A" | 3 |
+-------------+------------+
| "C" | 2 |
+-------------+------------+
| "B" | 1 |
+-------------+------------+
我正在寻找一排:
+-------------+------------+
| customer_id | order_time |
+-------------+------------+
| "A" | 6 |
+-------------+------------+
也就是说,我想获得最新(连续)customer_id
中的第一个order_time
。使用以下 SQL,我只能从 customer_id
A 获得“3”作为 order_time
。我似乎无法“重置”分区。
SELECT customer_name, LAST_VALUE(order_time) OVER W
FROM
(
SELECT UNNEST(ARRAY['A', 'A', 'A', 'B', 'B', 'A', 'C', 'B']) AS customer_name, generate_series(8, 1, -1) AS order_time
) X
WINDOW W AS (PARTITION BY customer_name ORDER BY order_time DESC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING)
ORDER BY order_time DESC
LIMIT 1;
使用 PostgreSQL 11.5
【问题讨论】:
【参考方案1】:你可以使用区别
ROW_NUMBER() OVER (ORDER BY order_time DESC)
和
ROW_NUMBER() OVER (PARTITION BY customer_name ORDER BY order_time DESC)
为gaps-and-islands
结构提供分组:
SELECT XX.customer_name, LAST_VALUE(order_time) OVER W FROM
(
SELECT X.*, ROW_NUMBER() OVER (ORDER BY order_time DESC)-
ROW_NUMBER() OVER (PARTITION BY customer_name ORDER BY order_time DESC)
AS rn
FROM
(
SELECT UNNEST(ARRAY['A', 'A', 'A', 'B', 'B', 'A', 'C', 'B']) AS customer_name,
generate_series(8, 1, -1) AS order_time
) X
) XX
WINDOW W AS (PARTITION BY rn ORDER BY order_time DESC
ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING)
LIMIT 1;
Demo
【讨论】:
谢谢!绝对似乎可以解决问题。可以这么说,我希望有一些不那么复杂/更内置的东西。但似乎我在备用框架规范方面走在了正确的轨道上。 顺便问一句,我不应该需要“最终”ORDER BY
对吗? LIMIT 1
之前的那个。
我们真的需要将customer_name
添加到PARTITION BY rn
吗?这不是矫枉过正吗?
是的 @Jeff ,很好,我们已经在上一步中想到了。以上是关于无法重置分区组(特别是 Window 函数和 PostgreSQL)的主要内容,如果未能解决你的问题,请参考以下文章
Spark 12 GB 数据加载与 Window 函数性能问题
某云elasticsearch节点失效,手动重置primary,迁移分区