如何在PostgreSQL中有效地編寫這樣的查詢：將連續的行聚合為數組（由一對列標識）？

Question

我有一個查詢，如：

(
    SELECT
        t1.person_id,
        t1.created_at,
        't1' AS type,
        t1.extra_data AS extra_data
    FROM table1 AS t1
)
UNION
(
    SELECT
        t2.person_id,
        t2.created_at,
        't2' AS type,
        t2.extra_data AS extra_data
    FROM table2 AS t2
)
UNION
(
    SELECT
        t3.person_id,
        t3.created_at,
        't3' AS type,
        t3.extra_data AS extra_data
    FROM table3 AS t3
)
ORDER BY created_at DESC;

這將導致類似的結果（ created_at是一個時間戳，我省略了具體值，並使用簡單的整數來表示順序）

person_id  | type  |  created_at | extra_data
---------  | ----  |  ---------- | ----------
1          | t1    |  9          | a
1          | t1    |  8          | b
2          | t2    |  7          | c
2          | t2    |  6          | c
2          | t2    |  5          | d
1          | t3    |  4          | e
3          | t3    |  3          | f

我想將連續的（person_id，type）對分組，最大的created_at作為最終的created_at，並將extra_data聚合到一個數組中，即我想得到以下結果：

person_id  | type  |  created_at | extra_data_array
---------  | ----  |  ---------- | ----------
1          | t1    |  9          | [a, b]
2          | t2    |  7          | [c, c, d]
1          | t3    |  5          | e
3          | t3    |  4          | f

我已經嘗試了窗口函數，但未能弄清楚如何實現。

我的問題是：

1）如何編寫一個單查詢來實現我的目標？

2）可以使用索引快速查詢嗎？

我對第二個問題的擔心是，由於基本結果是從UNION查詢中選擇的，因此我懷疑是否有任何機會利用索引。

任何人，感謝各種幫助！

Answer 1

您需要的兩個聚合函數是MAX和array_agg 。

我認為，如果在UNION之前先GROUP ，那么應用索引會更好，所以我會這樣做：

(
    SELECT
        t1.person_id,
        MAX(t1.created_at) AS created_at,
        't1' AS type,
        array_agg(t1.extra_data) AS extra_data
    FROM table1 AS t1
    GROUP BY t1.person_id
)
UNION
(
    SELECT
        t2.person_id,
        MAX(t2.created_at) AS created_at,
        't2' AS type,
        array_agg(t2.extra_data) AS extra_data
    FROM table2 AS t2
    GROUP BY t2.person_id
)
UNION
(
    SELECT
        t3.person_id,
        MAX(t3.created_at) AS created_at,
        't3' AS type,
        array_agg(t3.extra_data) AS extra_data
    FROM table3 AS t3
    GROUP BY t3.person_id
)
ORDER BY created_at DESC;

但是，您也可以先UNION ，然后再將結果GROUP ，只要您將GROUP BY person_id, type 。

還有一些注意事項：

• 這仍將對所有表進行全表掃描，因為您需要從每一行獲取created_at和extra_data 。 就像采訪問題一樣，“打印二叉樹的節點的時間復雜度是多少？”

• 如果要排序數組，則可以使用array_agg(t1.extra_data ORDER BY t1.created_at)或其他方法進行array_agg(t1.extra_data ORDER BY t1.created_at) 。 這是索引可以幫助您的地方。