删除PostgreSQL中的表而不删除关联的序列

Question

我有一张桌子foo.为了快速升级/部署我的网站,我创建了一个新表tmp_foo,以包含一些新数据:

create table tmp_foo (like foo including constraints including defaults including indexes);

现在每个表都有一个PK id列,如下所示:

   Column    |         Type          |                                Modifiers                                 
-------------+-----------------------+--------------------------------------------------------------------------
 id          | integer               | not null default nextval('foo_id_seq'::regclass)

重要的是,两个表都依赖于完全相同的序列foo_id_seq.没有tmp_foo_id_seq.这对我来说似乎没问题.

在此之后,我加载tmp_foo了新数据并重命名了表格,以便tmp_foo接管真实foo,并且原始版本foo成为了foo_old.现在我试着放弃foo_old:

db=> drop table foo_old ;
ERROR:  cannot drop table foo_old because other objects depend on it
DETAIL:  default for table foo_old column id depends on sequence foo_id_seq

很公平,id列默认仍然取决于序列.

db=> alter table foo_old alter column id drop default;

这是踢球者.

db=> drop table foo_old ;
ERROR:  cannot drop table foo_old because other objects depend on it
DETAIL:  default for table foo column id depends on sequence foo_id_seq

因此foo_old不再对序列有任何可见的依赖性,但它仍然试图将序列与表一起删除(显然不会因为新表依赖于它).

所以问题是两部分:

1. 为什么序列仍与旧表相关联？
2. 有没有办法绕过这个不涉及使新表依赖于一个新的或不同的序列(如果这甚至会有帮助)？

(在PostgreSQL 8.4上)

Answer 1

试试这个:

ALTER SEQUENCE foo_id_seq OWNED BY NONE

然后你应该能够放弃桌子.

要检索序列的"所有者",请使用以下查询

SELECT s.relname as sequence_name,  
       n.nspname as sequence_schema,  
       t.relname as related_table, 
       a.attname as related_column 
  FROM pg_class s, pg_depend d, pg_class t, pg_attribute a, pg_namespace n 
  WHERE s.relkind     = 'S' 
    AND n.oid         = s.relnamespace 
    AND d.objid       = s.oid 
    AND d.refobjid    = t.oid 
    AND (d.refobjid, d.refobjsubid) = (a.attrelid, a.attnum)