On Sat, 5 Jan 2019 at 03:12, Tomas Vondra <tomas.vondra@2ndquadrant.com> wrote:
> >>
> >> partitions 0 100 1000 10000
> >> --------------------------------------------
> >> master 19 1590 2090 128
> >> patched 18 1780 6820 1130
> >>
> >> So, that's nice. I wonder why the throughput drops so fast between 1k
> >> and 10k partitions, but I'll look into that later.
> >
> > Those look strange. Why is it so slow with the non-partitioned case?
> > I'd have expected that to be the fastest result.
> >
>
> Because there are 1M rows in the table, and it's doing a seqscan.
Of course. My test did the same, but I didn't consider that because I
had so few rows per partition. Likely just adding an index would have
it make more sense.
--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services