Am Donnerstag, den 07.12.2017, 18:23 -0800 schrieb Peter Geoghegan:
> Does "set replacement_sort_tuples = 0" change anything on
> REL9_6_STABLE? If you only get a crash when there is very little
> work_mem, then that might be a good avenue of investigation. Note
> that
> my changes to external sorting started in REL9_6_STABLE, so they
> can't
> be involved here.
>
replacement_sort_tuples = 0 changes the picture, indeed. With that
setting, the query runs without problems in REL9_6_STABLE.
> Are you aware of commit 512f67c8d02cc558f9c269cc848b0f0f788c4fe1,
> which fixed a bug affecting external sorts? Are you sure that you
> have
> that fix on REL9_5_STABLE + REL9_6_STABLE?
My test instances are build against a fresh pull from yesterday.