On 2013-01-25, Tim Uckun <timuckun@gmail.com> wrote:
>> I agree that seems like the most likely cause. Each update to the
>> row holding the hstore column requires adding new index entries for
>> all the hstore elements, and autovacuum will need to clean up the
>> old ones in the background. The best solution would be to either
>> normalize the data instead of using hstore, or move the hstore to a
>> separate table which is referenced by some sort of ID from the
>> frequently-updated table.
>
>
> That's very interesting. I can certainly split up the table, no big
> deal there. So would the index be redone even if I am not updating
> the hstore field itself?
Absolutely! see MVCC.
http://www.postgresql.org/docs/current/static/mvcc-intro.html
--
⚂⚃ 100% natural