On Mon, Dec 30, 2019 at 12:46:40PM -0800, Robert Treat wrote:
On Mon, Dec 23, 2019 at 9:11 PM Robert Haas <robertmh...@gmail.com> wrote:
On Sat, Dec 21, 2019 at 1:25 PM Tom Lane <t...@sss.pgh.pa.us> wrote:
> > What is the overhead here except the memory consumption?
>
> The time to copy those strings out of shared storage, any time
> you query pg_stat_activity.
It seems like you're masterminding this, and I don't know why. It
seems unlikely that anyone will raise the value unless they have very
long queries, and if those people would rather pay the overhead of
copying more data than have their queries truncated, who are we to
argue?
If increasing the maximum imposed some noticeable cost on
installations the kept the default setting, that might well be a good
argument for not raising the maximum. But I don't think that's the
case. I also suspect that the overhead would be pretty darn small even
for people who *do* raise the default setting. It looks to me like
both reading and write operations on st_activity_raw stop when they
hit a NUL byte, so any performance costs on short queries must come
from second-order effects (e.g. the main shared memory segment is
bigger, so the OS cache is smaller) which are likely irrelevant in
practice.
I'm generally in favor of the idea of allowing people to make
trade-offs that best work for them, but Tom's concern does give me
pause, because it isn't clear to me how people will measure the
overhead of upping this setting. If given the option people will
almost certainly start raising this limit because the benefits are
obvious ("I can see all my query now!") but so far the explanation of
the downsides have been either hand-wavy or, in the case of your
second paragraph, an argument they are non-existent, which doesn't
seem right either; so how do we explain to people how to measure the
overhead for them?
I think there are two questions that we need to answer:
1) Does allowing higher values for the GUC mean overhead for people who
don't actually increase it?
I don't think so.
2) What's the overhead for increasing the value for short/long queries?
My assumption is that for short queries, it's going to be negligible.
For longer queries it may be measurable, but I'd expect longer queries
to be more expensive in general, so maybe it's still negligible.
Of course, the easiest thing we can do is actually measuring this.
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
