Ed L. wrote:

>On Tuesday November 9 2004 9:52, Jan Wieck wrote:
>  
>
>>>I see the madness, no way to bookmark progress on the master without
>>>getting way into MVCC.
>>>      
>>>
>>And with that MVCC mess, make "sure" that no vacuum on that master
>>accidentially removes old rows. These are not locked by any existing
>>transaction any more, because the COPYing slon got clobbered ... but we
>>would like them to hang in there longer anyway. It is madness, indeed.
>>    
>>
>
>By "old rows" removed by vacuum, I assume you're talking about dead tuples 
>that get reused by vacuum or reclaimed by vacuum full?  Why do we care 
>about those?  If they were dead, I thought they didn't exist as far as any 
>live db activity was concerned?
>  
>
Ah, but they _aren't_ dead with respect to the COPY SET event, because 
we need them copied over in order for them to subsequently be modified 
by the entries already queued up in sl_log_1.

They're gone, but they can't be forgotten...