On Thu, Apr 8, 2021 at 1:05 PM Merlin Moncure <mmoncure(at)gmail(dot)com> wrote:
> This effect is only noticeable when the remote query is returning
> volumes of data. My question is, is there any way to sleep loop
> client side without giving up 3x performance penalty? Why is that
> that when more local sleep queries are executed, performance improves?
Looking at this more, it looks like that when sleeping with pg_sleep,
libpq does not receive the data. I think for this type of pattern to
work correctly, dblink would need a custom sleep function wrapping
poll (or epoll) that consumes input on the socket when signalled read
ready.
merlin