On 8/15/2007 4:00 PM, Darcy Buskermolen wrote:
> On Wednesday 15 August 2007 09:49:52 Tom Lane wrote:
>> Heikki Linnakangas <heikki at enterprisedb.com> writes:
>> > Maybe we can do something to reduce the xid consumption? For example,
>> > reuse xids for read-only queries.
>>
>> Hmm, that's an idea.
>>
>> More simply, just keep the current transaction open (resetting
>> everything but the XID) if we have made no changes by the time we're
>> told to commit or rollback ... which is something we track already,
>> so as not to waste cycles on useless commit XLOG records.
> 
> Jan and myself were discussing something like this as it would relate to a 
> subscribe process for slony.
> 
> Jan care to summerize your thoughts on this?

That was a different issue, where the catch up got stuck because the 
index scan over sl_log_* got stuck on the minxid of the copy_set and 
returned ever growing sets to be filtered through the snapshot logic. 
This has since been resolved by changing the log selecting query.

The idea to reuse readonly XID's is interesting. Like vacuum, Slony 
would suffer from long running transactions though, so I would urge to 
make sure Tom's suggestion is followed. Which isn't as easy as it sounds 
since the decision to keep the transaction open is done right before the 
backend would wait for the next query. Since there is no guarantee it 
ever does, this feature would need a timeout mechanism or it would lead 
to idle transactions where the client didn't even ask for one.


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 #