Re: Query times change by orders of magnitude as DB ages

On Wed, Nov 25, 2009 at 4:58 PM, Kevin Grittner <Kevin(dot)Grittner(at)wicourts(dot)gov
> wrote:

> Grzegorz Jaœkiewicz<gryzman(at)gmail(dot)com> wrote:
>
> > Other way, is to perform regular cluster && reindex
>
> If you CLUSTER there is no reason to REINDEX; indexes are rebuilt by
> the CLUSTER command.
>
> Also, if you do a good job with regular VACUUMs, there isn't any bloat
> to fix. In that case a regular CLUSTER would only be needed if it was
> worth the cost to keep data physically organized in the index
> sequence.
>
> the out of order data layout is primary reason for index bloat. And that
happens , and gets worse over time once data is more and more distributed.
("random" deletes, etc).
Thus suggestion of partitioning. I for one, hope in 8.5 we will get much
more user friendly partitioning interface - and we would no longer have to
write custom triggers. Which is probably the only reason I am only going to
partition a table only if it is really really really ... needed.

--
GJ