On 2010-12-02 01:18, Jim Nasby wrote:
> On Dec 1, 2010, at 8:07 AM, Yeb Havinga wrote:
>> FK's cannot refer to rows in inheritance childs.
> We have partially solved this issue at work. In our scenario, we're not using inheritance for partitioning, we're using it for, well, inheriting. As part of that, we have a field in the parent table that tells you what "type" of object each row is, and constraints on the child tables that enforce that. We've created triggers that perform the same operations that the built-in RI triggers do, namely grabbing a share lock on the target row. The difference is that our trigger looks at the "type" field to determine exactly what table it needs to try and grab shared locks on (we need to do this because the backend doesn't allow you to SELECT ... FROM parent FOR SHARE).
That part is exactly what the current WIP patch takes care of: grabbing
share locks on the right relation.
> Our solution is not complete though. Offhand, I know it doesn't support cascade, but I think there's more stuff it doesn't do. AFAIK all of those shortcomings could be handled with whats available at a user level though, so someone with enough motivation could produce an entire RI framework that worked with inheritance (though the framework would need a way to work around the uniqueness issue).
But is 'it can be solved on the user level' enough reason to not
implement it in the server code? Foreign key and unique constraint
checking are features the server should provide.
In response to
pgsql-hackers by date
|Next:||From: Robert Haas||Date: 2010-12-02 14:37:37|
|Subject: Re: crash-safe visibility map, take three|
|Previous:||From: Alvaro Herrera||Date: 2010-12-02 13:52:35|
|Subject: Re: improving foreign key locks|