用于分页的 Linq SQL 生成

Question

使用 EF6、ODP.NET 12.2 和 Oracle 12.2 服务器，我有下表：

id  raw(16)
number  number(19)
name    varchar2(100 char) not null
description varchar2(100 char) not null

（名称）和（名称、描述）的索引

以下 LINQ 查询：

db.SAMPLES.OrderBy(p => p.NAME).ThenBy(p => p.DESCRIPTION).Skip(0).Take(10).ToList();

产生这个 SQL：

SELECT * 
 FROM ( 
 SELECT 
 "Extent1"."ID" AS "ID", 
 "Extent1"."NUMBER" AS "NUMBER", 
 "Extent1"."NAME" AS "NAME"
 FROM ( SELECT "Extent1"."ID" AS "ID", "Extent1"."NUMBER" AS "NUMBER", "Extent1"."NAME" AS "NAME", row_number() OVER (ORDER BY "Extent1"."NAME" ASC) AS "row_number"
  FROM "ZENKI"."SAMPLEs" "Extent1"
 )  "Extent1"
 WHERE ("Extent1"."row_number" > 0)
 ORDER BY "Extent1"."NAME" ASC
 )
 WHERE (ROWNUM <= (10) )

查询运行良好，但性能不佳。显然，由于 (ORDER BY "Extent1"."NAME" ASC) 的（额外？？）顺序，Oracle 没有使用索引。

查询的执行计划是：

----------------------------------------------------------------------------------------------
| Id  | Operation                          | Name      | Rows   | Bytes    | Cost | Time     |
----------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                   |           |     10 |     2250 |    2 | 00:00:01 |
| * 1 |   COUNT STOPKEY                    |           |        |          |      |          |
|   2 |    VIEW                            |           |     10 |     2250 |    2 | 00:00:01 |
| * 3 |     SORT ORDER BY STOPKEY          |           |     10 |     2380 |    2 | 00:00:01 |
| * 4 |      VIEW                          |           |     10 |     2380 |    1 | 00:00:01 |
|   5 |       WINDOW NOSORT                |           |     10 |     1300 |    1 | 00:00:01 |
|   6 |        TABLE ACCESS BY INDEX ROWID | SAMPLEs   | 245376 | 31898880 |    1 | 00:00:01 |
|   7 |         INDEX FULL SCAN            | INDEX_SP1 |     10 |          |    1 | 00:00:01 |
----------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
------------------------------------------
* 1 - filter(ROWNUM<=10)
* 3 - filter(ROWNUM<=10)
* 4 - filter("Extent1"."row_number">0)

如果我删除最后一个订单，查询会很快并产生相同的结果。而执行计划变成：

--------------------------------------------------------------------------------------------
| Id  | Operation                        | Name      | Rows   | Bytes    | Cost | Time     |
--------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                 |           |     10 |     2380 |    1 | 00:00:01 |
| * 1 |   COUNT STOPKEY                  |           |        |          |      |          |
| * 2 |    VIEW                          |           |     10 |     2380 |    1 | 00:00:01 |
|   3 |     WINDOW NOSORT                |           |     10 |     1300 |    1 | 00:00:01 |
|   4 |      TABLE ACCESS BY INDEX ROWID | SAMPLEs   | 245376 | 31898880 |    1 | 00:00:01 |
|   5 |       INDEX FULL SCAN            | INDEX_SP1 |     10 |          |    1 | 00:00:01 |
--------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
------------------------------------------
* 1 - filter(ROWNUM<=10)
* 2 - filter("Extent1"."row_number">0)

经过大量谷歌搜索后，我在 ODP.NET 论坛中询问并得到了这个：

https://community.oracle.com/thread/4126265

似乎 odp.net 无法控制（或限制）生成的 SQL。我认为提供程序从表达式树生成 SQL，那么这怎么可能呢？有其他人对这种类型的查询有这个问题吗？

更新：

LINQ 论坛中的帖子：

https://social.msdn.microsoft.com/Forums/en-US/b6174ef9-a832-45da-8345-b97441d40363/linq-query-pagination-sql-generation?forum=linqtosql

更新 1：

使用 Number 数据类型而不是 varchar 作为 order by 会产生不同的解释计划并且非常快，所以我开始认为这与数据类型有任何关系。

Answer 1

恐怕我帮不上忙，我不使用你做的软件。不过，由于没有人在 12 小时内回复，而且我有一些空闲时间，所以我想说几句话。

以下 LINQ 查询 (...) 生成此 SQL (...)

（稍微格式化以提高可读性；我已经标记了查询的 A、B 和 C 部分，稍后将讨论）：

-- C
SELECT * 
 FROM (-- B
       SELECT 
          "Extent1"."ID" AS "ID", 
          "Extent1"."NUMBER" AS "NUMBER", 
          "Extent1"."NAME" AS "NAME"
       FROM (-- A
             SELECT "Extent1"."ID" AS "ID", 
                    "Extent1"."NUMBER" AS "NUMBER", 
                    "Extent1"."NAME" AS "NAME", 
                    row_number() OVER (ORDER BY "Extent1"."NAME" ASC) AS "row_number"
             FROM "ZENKI"."SAMPLEs" "Extent1"
            )  "Extent1"
       WHERE ("Extent1"."row_number" > 0)
       ORDER BY "Extent1"."NAME" ASC
      )
 WHERE (ROWNUM <= (10) );

如果我们从内部，A查询，这里有一些cmets：

答： ROW_NUMBER“排名”按 NAME ASC 排序的行。没关系。

SELECT "Extent1"."ID" AS "ID", 
       "Extent1"."NUMBER" AS "NUMBER", 
       "Extent1"."NAME" AS "NAME", 
       row_number() OVER (ORDER BY "Extent1"."NAME" ASC) AS "row_number"
  FROM "ZENKI"."SAMPLEs" "Extent1"

B: WHERE 子句是不必要的； ROW_NUMBER 总是 > 0。ORDER BY 也是不必要的，因为 A 查询中使用了 ROW_NUMBER，这使得整个 B 查询成为多余的。

select id, number, name
from A
where row_number > 0
ORDER BY NAME asc

C: 如果 A 查询没有生成 ROW_NUMBER，WHERE 子句才有意义；使用其中一个，而不是两个

select *
from B
where rownum <= 10

基本上，这也可以完成这项工作：

select *
from A
where row_number <= 10

即如果你能让 LINQ 产生这样的东西，性能也可能会受益。

select *
from (SELECT "Extent1"."ID" AS "ID", 
        "Extent1"."NUMBER" AS "NUMBER", 
        "Extent1"."NAME" AS "NAME", 
        row_number() OVER (ORDER BY "Extent1"."NAME" ASC) AS "row_number"
      FROM "ZENKI"."SAMPLEs" "Extent1" 
    )
where row_number <= 10

只是好奇：如果你运行它，它的性能如何？它是否返回所需的结果？如果答案是“好且是”，也许您可​​以尝试说服 LINQ 生成不同的查询。