On Mon, 2007-11-19 at 15:26 +0100, Cyril SCETBON wrote:
> 
> Stéphane A. Schildknecht wrote:
> 
> > Cyril SCETBON a écrit :
> >> I think it would be better to be able to modify SLON_DATA_FETCH_SIZE
> >> by using a parameter on the slon command line. For example, we got a
> >> lot of updates (400 per second) between far geographical sites. I had
> >> to modify this variable and to recompile the slon binary cause I
> >> didn't find another way until now.
> >
> > How did you modify this variable? Increase or decrease ? 
> I've modified the value of SLON_DATA_FETCH_SIZE in src/slon/slon.h and 
> rebuilt it.
> I've changed it from 10 to 50 to make fetchs of 500 lines.
> > What was the
> > problem you encountered ?
> >   
> When vaccum was happening on slony tables the db rate of the replication 
> decreased and slony had to get 3000 DML a few minutes later to make the 
> subscriber up to date. But fetching 3000 DMLs orders by fetching 100 
> lines was too slow. That's why I've increased the value to make fetchs 
> of 500 lines.

Chris,

I didn't explicitly make the connection before but my investigation of
whether we needed to VACUUM at all was connected to Cyril's performance
decrease described above.

We discussed increasing SLON_DATA_FETCH_SIZE, but we probably wouldn't
need to do that if VACUUM didn't run and slow us down occasionally. 

So maybe we can knock one thing off your task list, at least?

I can't remember where we left the earlier discussion on the vacuuming.
Did you want me to make a patch, or can I leave it with you guys? No
reason to rush at all my end, just trying to remember where we were.

-- 
  Simon Riggs
  2ndQuadrant  http://www.2ndQuadrant.com