On 2018/05/17 14:19, Amit Langote wrote:
> Looking at this for a bit, I wondered if this crash wouldn't have occurred
> if the "propagation" had also considered join relations in addition to
> simple relations. For example, if I changed inheritance_planner like the
> attached (not proposing that we consider committing it), reported crash
> doesn't occur. The fact that it's not currently that way means that
> somebody thought that there is no point in keeping all of those joinrels
> around until plan creation time. If that is so, is it a bit worrying that
> a FDW function invoked from createplan.c may try to look for one?
Oops, I forgot to attach the patch that I had used in the experiment.
Thanks,
Amit