Re: Broken index?

Paul Green <traktion(at)webleicester(dot)co(dot)uk> writes:
> I didn't actually compile it at all - I used the default RPMs provided by
> Redhat. I'm unsure whether they have compiled in support or not,

They do...

> but I know
> I always start/stop postgres using the init.d scripts, which always start
> the database under the user postgres. As far as I'm aware, I've never
> changed the locale of the system/db either. I just checked using 'locale'
> to see how the system was set up under the usernames I use to log in to the
> machine and they are all en_GB anyway. I also checked /etc/sysconfig/i18n
> and that too was set to en_GB, so I think I'm safe on this front.

Nonetheless, it certainly appears that you've got a corrupt index, so
I'm guessing that at one time or another the wrong locale environment
was used. Even one out-of-order entry in the index will cause
long-lasting problems.

You could try a direct observation of index ordering, say

SELECT name FROM table ORDER BY name;

If you run this first with enable_sort turned off and second with
enable_indexscan turned off, you should get indexscan and explicit-sort
plans respectively (use EXPLAIN to make sure). Dump the outputs into
files and compare...

> Any ideas?

My recommendation is to update to 7.1.3 in any case.

regards, tom lane