与 `CROSS APPLY` 和 `OUTER APPLY` 行为不一致

Inconsistent behavior with `CROSS APPLY` and `OUTER APPLY`

我在 Oracle-12c 中有一个架构,类似于具有 accountspostscomments 的典型论坛。我正在编写一个查询以获取...

  1. 一个用户
  2. 该用户的所有帖子
  3. 每个帖子的评论
  4. 以及每条评论的作者。

查询如下所示:

select "accounts".*, "p".*, "c".*, "author".*
from "accounts"
cross apply (
    select * from "posts"
    where "posts"."author_id" = "accounts"."id"
) "p"
cross apply (
    select * from "comments"
    where "comments"."post_id" = "p"."id"
) "c"
left join "accounts" "author" on "author"."id" = "c"."author_id"
where "accounts"."id" = 1

此查询按预期工作。我使用 CROSS APPLY 而不是典型的 JOIN,因为我将添加 OFFSETFETCH 稍后分页。然而,问题是 CROSS APPLY 忽略了没有评论的帖子,这是我不想要的。我想 保留 结果中的帖子,即使它们没有评论。

所以我尝试将 CROSS APPLY 更改为 OUTER APPLY

select "accounts".*, "p".*, "c".*, "author".*
from "accounts"
outer apply (
    select * from "posts"
    where "posts"."author_id" = "accounts"."id"
) "p"
outer apply (
    select * from "comments"
    where "comments"."post_id" = "p"."id"
) "c"
left join "accounts" "author" on "author"."id" = "c"."author_id"
where "accounts"."id" = 1

但现在我得到这个错误:

ORA-00904: "p"."id": invalid identifier
00904. 00000 -  "%s: invalid identifier"
*Cause:    
*Action:
Error at Line: 9 Column: 34

出于某种原因,我的第二个 OUTER APPLY 加入抱怨我引用了第一个的结果中的 "p"."id"。但是当我使用 CROSS APPLY.

时没问题

这是怎么回事?为什么它们之间的行为存在差异?

编辑:OUTER APPLY 在这个基本示例中似乎没有必要。这是从一个更复杂的场景中提炼出来的,在这个场景中,我必须坚持 OUTER APPLY 确实是必要的,但该场景的细节与我要问的实际问题无关——这是关于这种看似未记录的行为差异在 CROSSOUTER 之间 APPLY.

编辑:

Oracle 版本: Database 12c Standard Edition Release 12.1.0.2.0 - 64bit Production

客户端: Oracle SQL 开发者版本 4.2.0.16.356

服务器: uname -a 的输出 - Linux ubuntu-1gb-sfo2-01 4.4.0-64-generic #85-Ubuntu SMP Mon Feb 20 11:50:30 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux

DDL: link

您的查询没有问题。您遇到了 12.1.0.2 中引入的错误 20356733/21547130,如 here 所述。为了克服它,请使用 12.1.0.2 之前的版本或应用最新更新(链接线程声称修复程序在 12.1.0.2.160419 补丁集更新中可用)。

此答案主要由 Matthew McPeak and Martin Smith 找到。我刚刚按照如下所述进行了第一次尝试,发现该问题在 Oracle 12.1.0.1 上无法重现。

第一次尝试回答:

我已经使用您的模式创建了测试数据库,这两个查询对我来说都工作正常。第一个没有return post没有注释,第二个return全是账号post,没有任何ORA-00904错误。我在 Oracle 12c 上做了这个测试。

继续你的问题:

  1. 尝试 copy/paste 并从您的 post 执行第二个查询。有时一些令人讨厌的拼写错误会潜入查询中。来自你的 post 的精确查询按预期对我有效。

  2. 如果您仍然遇到相同的错误,请提供您用于 accountspostscomments 表创建的 DDL。

  3. 请说明您使用的 SQL 客户端。错误肯定是在服务器端,但在这种奇怪的情况下,每个小细节都会造成差异。

我的测试数据库:

CREATE TABLE "accounts"
(
    "id" NUMBER(11) NOT NULL,
    "name" NVARCHAR2(256),
    CONSTRAINT ACCOUNTS_PK PRIMARY KEY ("id")
)
/

CREATE TABLE "posts"
(
    "id" NUMBER(11) NOT NULL,
    "author_id" NUMBER(11) NOT NULL,
    "post_text" NVARCHAR2(1024),
    CONSTRAINT POSTS_PK PRIMARY KEY ("id"),
    CONSTRAINT POST_ACCOUNT_FK FOREIGN KEY ("author_id") REFERENCES "accounts" ("id") ON DELETE CASCADE
)
/

CREATE TABLE "comments"
(
    "id" NUMBER(11) NOT NULL,
    "author_id" NUMBER(11) NOT NULL,
    "post_id" NUMBER(11) NOT NULL,
    "comment_text" NVARCHAR2(1024),
    CONSTRAINT COMMENTS_PK PRIMARY KEY ("id"),
    CONSTRAINT COMMENT_ACCOUNT_FK FOREIGN KEY ("author_id") REFERENCES "accounts" ("id") ON DELETE CASCADE,
    CONSTRAINT COMMENT_POST_FK FOREIGN KEY  ("post_id") REFERENCES "posts" ("id") ON DELETE CASCADE
)
/

INSERT INTO "accounts"("id", "name") VALUES(1, 'testuser')
/
INSERT INTO "posts"("id", "author_id", "post_text") VALUES(1, 1, 'First test post')
/
INSERT INTO "posts"("id", "author_id", "post_text") VALUES(2, 1, 'Second test post')
/
INSERT INTO "comments"("id", "author_id", "post_id", "comment_text") VALUES(1, 1, 2, 'It is a very cool post')
/
COMMIT

对于 CodeFuller's 的回答,我只想补充一点:(A) 有一个补丁可以解决这个错误,并且 (B) 有一个解决方法 SQL 适用于 12.1.0.2 ,虽然我不知道它是否符合您的目的。

解决方法基本上是嵌套连接,如下所示:

SELECT accounts.*,
       p.*,
       author.*
FROM   accounts
       CROSS APPLY (SELECT posts.id,
                           posts.author_id,
                           posts.text,
                           comments.comment_author_id,
                           comments.comment_text
                    FROM   posts
                           OUTER APPLY (SELECT comments.author_id comment_author_id,
                                               comments.text comment_text
                                        FROM   comments
                                        WHERE  comments.post_id = posts.id) comments
                    WHERE  posts.author_id = accounts.id) p
       LEFT JOIN accounts author
         ON author.id = p.comment_author_id
WHERE  accounts.id = 1;


ID   NAME      ID_1 AUTHOR_ID TEXT                                              COMMENT_AUTHOR_ID COMMENT_TEXT                            ID_2  NAME_1                               
---- --------- ---- --------- ------------------------------------------------- ----------------- --------------------------------------- ----- ------------------- 
   1 Fred         1         1 Fred wrote this and it has comments                               3 This is Helen's comment on Fred's post      3 Helen                                
   1 Fred         2         1 Fred wrote this and it does not have any comments
-------- End of Data --------
2 row(s) fetched

参考:table 解决方法的 DDL

CREATE TABLE accounts
(
  id     NUMBER PRIMARY KEY,
  name   VARCHAR2 (30)
);

CREATE TABLE posts
(
  id          NUMBER PRIMARY KEY,
  author_id   NUMBER,
  text        VARCHAR2 (240)
);

CREATE TABLE comments
(
  id          NUMBER PRIMARY KEY,
  post_id     NUMBER,
  author_id   NUMBER,
  text        VARCHAR2 (240)
);

INSERT INTO accounts (id, name)
VALUES (1, 'Fred');

INSERT INTO accounts (id, name)
VALUES (2, 'Mary');

INSERT INTO accounts (id, name)
VALUES (3, 'Helen');

INSERT INTO accounts (id, name)
VALUES (4, 'Iqbal');

INSERT INTO posts (id, author_id, text)
VALUES (1, 1, 'Fred wrote this and it has comments');

INSERT INTO posts (id, author_id, text)
VALUES (2, 1, 'Fred wrote this and it does not have any comments');

INSERT INTO posts (id, author_id, text)
VALUES (3, 4, 'Iqbal wrote this and it does not have any comments');

INSERT INTO comments (id,
                      post_id,
                      author_id,
                      text)
VALUES (1,
        1,
        3,
        'This is Helen''s comment on Fred''s post');