Bill Moran <wmoran(at)collaborativefusion(dot)com> writes:
> However, the part I wanted to comment on (and got busy yesterday so
> am only getting to it now) is that there's no guarantee that SELECT
> isn't modifying rows.
Another way that SELECT can cause disk writes is if it sets hint bits on
recently-committed rows. However, if the tables aren't actively being
modified any more, you'd expect that sort of activity to settle out pretty
quickly.
I concur with the temporary-file theory --- it's real hard to see how
analyzing the tables would've fixed it otherwise.
regards, tom lane