Re: Deadlocks caused by referential integrity checks

On Aug 27, 2004, at 5:16 PM, Joshua D. Drake wrote:

>
>> If PG had a way for me to tell it the action logger transaction was
>> "less important" and should be the one killed, I'd live with that,
>> since the other transaction is usually more expensive.
>
> Hello set the check deferrable.

Thanks... On which transaction would I set that?  The short one or the
long one?  Or both?

And how do you specify that a FK check is deferrable?  Do I need to
drop and recreate the FK with the deferrable attribute?