On 9/4/2007 9:53 PM, Dan Falconer wrote:
> 	I was then struck by yet another surprise: Slony seems to like keeping all 
> it's log data, even when it's the only running node.  I've noticed there have 

The case of having a replication set but no subscribers at all is a 
little pathologic and not optimized for especially.

Slony indeed likes to keep logs. It never vacuumes the sl_log_* tables. 
Instead it periodically switches them and truncates the old when all 
tuples have been deleted out of it. This could even be optimized by not 
even deleting them in the first place but deciding that none of them is 
needed any more. This switching is a process that can take a while if 
you for example have long running transactions. The log trigger only 
decides which log table to write to once per transaction. After the log 
switch is initiated, the system basically waits until all transactions 
that have been in progress during that log switch start have ended and 
then it waits until the log later on is completely empty because all 
known subscribers have replicated everything. There might be a bug in 
that code that actually prevents the delete of old log rows if you don't 
have any subscribers at all.

As said, it is a pathological case that should not happen in a normal 
setup where you have at least one single subscriber.


Jan

-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck at Yahoo.com #