On Thu, Aug 28, 2025 at 9:10 PM Andres Freund <andres@anarazel.de> wrote:
> Same. Tomas, could you share what you applied?
Tomas posted a self-contained patch to the list about an hour ago?
> > I'm not sure that Thomas'/your patch to ameliorate the problem on the
> > read stream side is essential here. Perhaps Andres can just take a
> > look at the test case + feature branch, without the extra patches.
> > That way he'll be able to see whatever the immediate problem is, which
> > might be all we need.
>
> It seems caused to a significant degree by waiting at low queue depths. If I
> comment out the stream->distance-- in read_stream_start_pending_read() the
> regression is reduced greatly.
IIUC, that is very roughly equivalent to what the patch actually does.
The fastest configuration of all, independent of io_method, is
"enable_indexscan_prefetch=off". So it's hard to believe that the true
underlying problem is low queue depth. Though I certainly don't doubt
that higher queue depths will help *when io_method=worker*.
--
Peter Geoghegan
