Question

我有两张桌子：

account_transaction：

+-------------------------------+--------------------------+------------------------+
|            Column             |           Type           |       Modifiers        |
+-------------------------------+--------------------------+------------------------+
| id                            | integer                  | not null               |
| account_id                    | bigint                   | not null               |
| created                       | timestamp with time zone | not null default now() |
| transaction_type              | text                     | not null               |
| amount                        | numeric(5,2)             | not null               |
| external_reference_id         | character varying(60)    |                        |
+-------------------------------+--------------------------+------------------------+

索引：

"idx_account_transaction_created" btree (created)

reporting_period：

+------------+--------------------------+-----------+
|   Column   |           Type           | Modifiers |
+------------+--------------------------+-----------+
| month      | text                     |           |
| created    | timestamp with time zone |           |
| date_range | tstzrange                |           |
+------------+--------------------------+-----------+

我希望获得上一个报告期的所有交易。以下是产生相同结果的两个查询，但是一个执行seq扫描，另一个可以使用idx_account_transaction_created索引。

explain select count(*) from account_transaction where created <@ (select date_range from reporting_period order by created desc limit 1);
+----------------------------------------------------------------------------------------+
|                                       QUERY PLAN                                       |
+----------------------------------------------------------------------------------------+
| Aggregate  (cost=4214.81..4214.82 rows=1 width=0)                                      |
|   InitPlan 1 (returns $0)                                                              |
|     ->  Limit  (cost=13.20..13.20 rows=1 width=40)                                     |
|           ->  Sort  (cost=13.20..13.60 rows=800 width=40)                              |
|                 Sort Key: reporting_period.created                                     |
|                 ->  Seq Scan on reporting_period  (cost=0.00..12.40 rows=800 width=40) |
|   ->  Seq Scan on account_transaction  (cost=0.00..4200.81 rows=1602 width=0)          |
|         Filter: (created <@ $0)                                                        |
+----------------------------------------------------------------------------------------+
(8 rows)

explain select count(*) from account_transaction where created >= '2014-06-01' and created <= '2014-06-30 23:59:59.999999';
+------------------------------------------------------------------------------------------------------------------------------------------------------------------+
|                                                                            QUERY PLAN                                                                            |
+------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Aggregate  (cost=2640.54..2640.54 rows=1 width=0)                                                                                                                |
|   ->  Index Only Scan using idx_account_transaction_created on account_transaction  (cost=0.08..2605.77 rows=69535 width=0)                                      |
|         Index Cond: ((created >= '2014-06-01 00:00:00+00'::timestamp with time zone) AND (created <= '2014-06-30 23:59:59.999999+00'::timestamp with time zone)) |
+------------------------------------------------------------------------------------------------------------------------------------------------------------------+
(3 rows)

我更喜欢第一个查询，因为它似乎更容易阅读和理解，而且只有一次往返。第二个更高效，因为它在创建的字段上使用索引，但意味着应用程序需要走出去并获得最后一个报告周期并获得date_range字段的下限和上限（不是最糟糕的事情）世界）。我想我总是把它写成一个函数或一个视图。但是，我只是有点惊讶PostgreSQL没有发现它可以使用索引。这里有什么我想念的吗？反正有没有让第一个查询使用索引？

我正在使用PostgreSQL 9.3

Answer 1

运算符<@需要使用GIN或GiST索引。不适用于普通的B树索引 Details in the manual here.
相关答案：

替代

对于您的用例，B树索引可能更有效。这应该允许Postgres使用它：

SELECT count(*) AS ct
FROM  (
   SELECT lower(date_range) AS ts_from, upper(date_range) AS ts_to
   FROM   reporting_period
   ORDER  BY created DESC
   LIMIT  1
   ) r
JOIN   account_transaction a ON a.created >= r.ts_from
                            AND a.created <  r.ts_to
;

假设您的所有tstzrange值都包含包括下限和不包括上限（建议的默认值）。要强制我在表格CHECK中建议reporting_period约束：

CHECK (lower_inc(date_range) AND NOT upper_inc(date_range))

否则你需要更复杂的条件。相关回答：

Preventing adjacent/overlapping entries with EXCLUDE in PostgreSQL

查询包含的tstzrange性能不佳

1 个答案:

替代