Re: pg_stat_bgwriter

On Mon, Oct 14, 2019 at 1:25 PM dangal <danielito(dot)gallo(at)gmail(dot)com> wrote:

> Do you think it should increase bgwriter_lru_maxpages due to the value of
> maxwritten_clean?
>

I find the background writer to be pretty unimportant these days. If the
kernel is freely accepting writes without blocking, the backends can
probably write their own buffers without it being a meaningful bottleneck.
On the other hand, if the kernel is constipated, no tweaking of the
background writer parameters is going to insulate the backends from that
fact. That said, I would increase bgwriter_lru_maxpages (or decrease
bgwriter_delay) anyway. It probably won't make much difference, but if it
does it is more likely to help than to hurt.

> Do you think it should increase bgwriter_lru_maxpages,
> bgwriter_lru_multiplier, and decrease bgwriter_delay due to the value of
> buffers_backend compared to buffers_alloc?
>

I don't think that that comparison is meaningful, so wouldn't make changes
based on it.

> Do you think a modification is necessary?
> What values would you recommend?
> thank you
>

If you are experiencing a problem, this is probably not the right way to
investigate it. If a particular query is slow, try EXPLAIN (ANALYZE,
BUFFERS). If lots of user-facing things are slow, try sampling "select
wait_event, wait_event_type from pg_stat_activity where
backend_type='client backend';"