【PostgreSQL里insert on conflict do操作时的冲突报错分析】

news2024/11/23 18:51:42

最近在巡检PostgreSQL的数据库的时候，发现部分数据库里存在大量的如下报错

ERROR:  ON CONFLICT DO UPDATE command cannot affect row a second time
HINT:  Ensure that no rows proposed for insertion within the same command have duplicate constrained values.

一、问题分析

通过报错的字段可以看出是使用了PostgreSQL 9.5 引入的一项新功能，insert on conflict do功能，即UPSERT的效果，当插入遇到约束错误时，直接返回，或者改为执行UPDATE。观察日志可以看出执行的SQL是带有多条记录的，推测多条记录的key值冲突了，因此报了这个错。

PostgreSQL 的 upsert 功能：当记录不存在时，执行插入；否则，进行更新。

PostgreSQL的代码里src/backend/executor/nodeModifyTable.c下的ExecOnConflictUpdate()函数里的注释部分其实解释的很清楚了。当在同一命令中再次更新刚插入的元组时，可能会发生这种情况。例如，因为插入了多个具有相同冲突键值的行。

MERGE 也有同样的问题，SQL-2003标准也类似地规定MERGE，在尝试更新同一行两次时必须引发异常。因为一次请求中对行的处理，顺序是不固定的，数据库不知道应该以哪条为最后需要保留的数据。

注释部分也明确表示出现这种问题属于用户的责任，PostgreSQL不会主动去处理这种报错，这种不在同一条SQL中出现多条相同KEY的数据的问题应该让用户去保障。

二、问题复现

1.建立测试表


postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:09:21)=# create table test(id int primary key, info text, crt_time timestamp);
CREATE TABLE
postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:09:22)=#  insert into test values (1,'test',now()) on conflict (id) do update set info=excluded.info,crt_time=excluded.crt_time;
INSERT 0 1
postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:09:44)=# select * from test;
+----+------+----------------------------+
| id | info |          crt_time          |
+----+------+----------------------------+
|  1 | test | 2024-04-16 14:09:44.405528 |
+----+------+----------------------------+
(1 row)

postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:09:51)=#  insert into test values (2,'hah','2024-04-16 14:01:33.640731') on conflict (id) do update set info=excluded.info,crt_time=excluded.crt_time;
INSERT 0 1
postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:10:01)=# select * from test;
+----+------+----------------------------+
| id | info |          crt_time          |
+----+------+----------------------------+
|  1 | test | 2024-04-16 14:09:44.405528 |
|  2 | hah  | 2024-04-16 14:01:33.640731 |
+----+------+----------------------------+
(2 rows)

2.模拟连续两次插入同样的主键值

可以看到，id是主键，我连续两次插入id=7，但是没有主键冲突，原本的insert变成了更新，把id=7的其他字段进行了更新。

postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:11:22)=#  insert into test values (7,'test',now()) on conflict (id) do update set info=excluded.
info,crt_time=excluded.crt_time;
INSERT 0 1
postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:11:38)=# select * from test where id=7;
+----+------+----------------------------+
| id | info |          crt_time          |
+----+------+----------------------------+
|  7 | test | 2024-04-16 14:11:38.229476 |
+----+------+----------------------------+
(1 row)

postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:11:51)=#  insert into test values (7,'ha','2024-04-16 14:15:38.229476') on conflict (id) do upda
te set info=excluded.info,crt_time=excluded.crt_time;
INSERT 0 1
postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:12:11)=# select * from test where id=7;
+----+------+----------------------------+
| id | info |          crt_time          |
+----+------+----------------------------+
|  7 | ha   | 2024-04-16 14:15:38.229476 |
+----+------+----------------------------+
(1 row)

3.模拟一条SQL包含相同的key值

可以看到，一个SQL里插入多个行记录，并且key值相同(违反主键冲突)的时候，报了ERROR: ON CONFLICT DO UPDATE command cannot affect row a second time。上文我们也通过代码的注释部分，知道了这种问题属于用户的责任，需要用户保障，因此这部分的内容可能需要结合业务侧进行分析，检查下业务逻辑了。

postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:10:03)=# insert into test values (3,'hah','2024-04-16 14:01:33.640731'),(3,'hah','2024-04-16 14:01:33.640731') on conflict (id) do update set info=excluded.info,crt_time=excluded.crt_time;
ERROR:  ON CONFLICT DO UPDATE command cannot affect row a second time
HINT:  Ensure that no rows proposed for insertion within the same command have duplicate constrained values.

postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:10:27)=# insert into test values (4,'hah','2024-04-16 14:01:33.640731'),(5,'hah','2024-04-16 14:
01:33.640731') on conflict (id) do update set info=excluded.info,crt_time=excluded.crt_time;
INSERT 0 2
postgres<16.1>(ConnAs[postgres]:PID[1103345] 2024-04-16/14:10:38)=# select * from test;
+----+------+----------------------------+
| id | info |          crt_time          |
+----+------+----------------------------+
|  1 | test | 2024-04-16 14:09:44.405528 |
|  2 | hah  | 2024-04-16 14:01:33.640731 |
|  4 | hah  | 2024-04-16 14:01:33.640731 |
|  5 | hah  | 2024-04-16 14:01:33.640731 |
+----+------+----------------------------+
(4 rows)