Question

为什么我插入的表格为空时会出现此错误？该表的上一次迭代（自已删除）有一个序列，所以我想知道Postgres是否正在缓存旧表模式？

这是错误：

-- Executing query:
DELETE FROM student
WHERE sno =101;

ERROR:  duplicate key value violates unique constraint "cancel_pkey"
DETAIL:  Key (eno, cdate)=(1, 2014-12-08 21:21:53.710883) already exists.
CONTEXT:  SQL statement "INSERT INTO cancel(eno,excode,sno)
            SELECT eno, excode, sno 
            FROM entry
            WHERE eno = OLD.eno"
PL/pgSQL function cancel_entry() line 3 at SQL statement
SQL statement "DELETE FROM entry
            WHERE sno = OLD.sno"
PL/pgSQL function delete_student() line 8 at SQL statement

实际的插入查询发生在触发器中，如下所示：

CREATE OR REPLACE FUNCTION delete_student()
RETURNS trigger AS $BODY$
    BEGIN
    IF (SELECT EXISTS (SELECT * FROM entry WHERE sno = OLD.sno)) THEN
    INSERT INTO cancel(eno,excode,sno)
        SELECT eno, excode, sno 
        FROM entry
        WHERE sno = OLD.sno;
    DELETE FROM entry
        WHERE sno = OLD.sno;
    END IF;
    RETURN OLD;
    END; $BODY$ 
    LANGUAGE plpgsql;

CREATE TRIGGER DeleteStudent
    BEFORE DELETE ON student
    FOR EACH ROW
    execute procedure delete_student();

cancel表架构如下所示：

CREATE TABLE cancel (
eno     INTEGER     NOT NULL,
excode  CHAR(4)     NOT NULL, --not unique as many instances of cancellations for exam
sno     INTEGER     NOT NULL, --not unique as student may cancel several exams
cdate   TIMESTAMP   NOT NULL    DEFAULT NOW(),
cuser   VARCHAR(128) NOT NULL   DEFAULT CURRENT_USER,
PRIMARY KEY(eno,cdate)
-- sno is not a foreign key, as it must still exist even where student is deleted. 
);

entry表架构是：

CREATE TABLE entry (
eno     INTEGER     NOT NULL    DEFAULT NEXTVAL('eno_sequence'),
excode  CHAR(4)     NOT NULL,
sno     INTEGER     NOT NULL,
egrade  DECIMAL(5,2)    CHECK(egrade BETWEEN 0 AND 100),
PRIMARY KEY(eno),
FOREIGN KEY(excode) REFERENCES exam MATCH FULL ON DELETE RESTRICT,
FOREIGN KEY(sno) REFERENCES student MATCH FULL ON DELETE RESTRICT --handle with stored procedure to retain student reference in cancel table
);

＆＃39;学生＆＃39;表模式

CREATE TABLE student (
   sno          INTEGER         NOT NULL,
   sname            VARCHAR(20) NOT NULL,
   semail      VARCHAR(20)  NOT NULL,
    PRIMARY KEY(sno)
);

Answer 1

由于您正在从附加表entry移动行，并且您首先运行SELECT / INSERT，因此存在潜在的竞争条件：多个并发事务可以同时尝试相同时间。

但是，触发器由表DELETE上的student调用，对受影响的行进行ROW EXCLUSIVE锁定。您DELETE（WHERE sno = 101）的谓词位于唯一列上。尽管问题中缺少表student的实际表定义，但我可以从表entry中的FK定义中看出，它要求对引用列具有唯一或PK约束。这使得并发事务不会删除student中会在竞争行上调用触发器的行。

首先让我感到困惑的是：您显示了触发器/函数DeleteStudent / delete_student()，但异常是从 cancel_entry() 引发的，这是你问题中的不是。走出困境，看起来你在另一个触发器，中调用相同的INSERT命令，这将解释异常。

无论哪种方式，我们看到的功能更昂贵，并且比竞争条件更容易受到影响。首先使用data-modifying CTE删除（锁定行），然后只插入另一个表（这可能会解决您的问题，缺少信息）。还简化：

CREATE OR REPLACE FUNCTION delete_student()
  RETURNS trigger AS
$func$
BEGIN
   IF (SELECT EXISTS (SELECT * FROM entry WHERE sno = OLD.sno)) THEN -- no need
   WITH del AS (
      DELETE FROM entry  -- delete first, locking rows
      WHERE  sno = OLD.sno
      RETURNING eno, excode, sno
      )
   INSERT INTO cancel(eno, excode, sno) -- only then insert
   SELECT eno, excode, sno 
   FROM   del;
   END IF;
   RETURN OLD;
END
$func$  LANGUAGE plpgsql;

您根本不需要IF构造。如果SELECT（或更新版本中的DELETE）在entry中找不到任何行，则INSERT会一无所有。在CTE的情况下，当CTE INSERT没有返回任何行时，永远不会执行最终del。这就像它得到的那样快速和干净。

插入没有排序列的空表时重复键约束错误

2 个答案: