Jim C. Nasby wrote:
> On Mon, Apr 24, 2006 at 02:51:22PM +0000, Christopher Browne wrote:
>   
>> As far as lower bound goes, I don't see too much problem there.  The
>> value is used to determine how long the cleanup thread sleeps between
>> invocations.  The shortest that could *possibly* fall to is zero, in
>> which case it would run continually.
>>     
>
> Wouldn't that result in the thread continuously polling the database,
> presenting an unreasonable load?
>   
Well, yes...  That's the inherent risk... 

At some point, this will become counterproductive, for sure.
> I guess in some cases if you're paranoid about database size and don't
> care about performance that might be ok, but ISTM there should at least
> be a warning in the docs. Another concern is that running in that mode
> could cause an XID rollover relatively quickly.
I don't think that would be too much of a problem; the transactions
involved in doing the deletes would run for long enough that you'd not
be contributing anywhere as much to XID rollover from this as the
regular traffic would give.

The reason for sl_log_1 to be growing immensely is if either:
a) You're hitting the origin with quite a lot of pretty big updates, or
b) You're hitting the origin with Really A Lot of little updates.

If it's the latter, the cleanup loop would only be contributing a couple
dozen XIDs, over a not-exactly-momentary period of time.  I don't see
this being *big* trouble, as far as XID rollover is concerned.