Julien Rouhaud <rjuju123(at)gmail(dot)com> writes:
> +1, and more generally please follow
> https://wiki.postgresql.org/wiki/Slow_Query_Questions.
Yeah. FWIW, the most likely explanation for the change in behavior is
that by the time of the second execution, auto-analyze has managed to
update the table's statistics, and that for some reason that's changing
the plan for the worse. But then the next question is why and what
can be done about that. See the wiki entry for info that would be
helpful in diagnosing this.
regards, tom lane