> > More importantly, PostgreSQL 6.5.3 works very, very well without
> > VACUUM'ing.
>
> 6.5 effectively assumes that "foo = constant" will select exactly one
> row, if it has no statistics to prove otherwise.
I thought we had agreed upon a default that would still use
the index in the above case when no statistics are present.
Wasn't it something like a 5% estimate ? I did check
that behavior, since I was very concerned about that issue.
Now, what is so different in his case?
Andreas