Rod Taylor wrote:

>On Tue, 2004-11-30 at 12:23 -0500, Jan Wieck wrote:
>  
>
>>On 11/30/2004 11:29 AM, Rod Taylor wrote:
>>
>>    
>>
>>>I've been trying various methods of making Slony fail, and have noticed
>>>that after numerous attempts that the database is growing quickly in
>>>size without any data.
>>>
>>>The reason of course is the initial COPY, when rolled back, leaves a ton
>>>of dead tuples in the structures an indexes.
>>>
>>>Please consider adding a VACUUM of the subscriber in between retries.
>>>      
>>>
>>Unfortunately it is impossible to add a vacuum to slon right in between 
>>the delete and the copy, since everything is done in one transaction 
>>(and has to). So this can only be an administrative suggestion.
>>    
>>
>
>I mean between the 2 attempts of slony to go through the initial copy --
>so after slony issues a rollback() on the subscriber.
>  
>
Actually, the even better answer involves submitting a TRUNCATE request 
on the table; that clears everything out posthaste.  That was briefly in 
pre-1.0.3.

Unfortunately, that falls prey to two issues:
 1.  It doesn't work out so well on 7.3, where TRUNCATE must run 
_outside_ a transaction

 2.  Foreign key constraints bite the truncate if they exist.

Those factors mean that it's troublesome to use TRUNCATE, which would be 
the "better" answer...