Re: data_checksums enabled by default (was: Move --data-checksums to common options in initdb --help)

Hi,

On 2021-01-06 12:02:40 -0500, Stephen Frost wrote:
> * Andres Freund (andres(at)anarazel(dot)de) wrote:
> > On 2021-01-04 19:11:43 +0100, Michael Banck wrote:
> > > Am Samstag, den 02.01.2021, 10:47 -0500 schrieb Stephen Frost:
> > > > I agree with this, but I'd also like to propose, again, as has been
> > > > discussed a few times, making it the default too.
> >
> > FWIW, I am quite doubtful we're there performance-wise. Besides the WAL
> > logging overhead, the copy we do via PageSetChecksumCopy() shows up
> > quite significantly in profiles here. Together with the checksums
> > computation that's *halfing* write throughput on fast drives in my aio
> > branch.
>
> Our defaults are not going to win any performance trophies and so I
> don't see the value in stressing over it here.

Meh^3. There's a difference between defaults that are about resource
usage (e.g. shared_buffers) and defaults that aren't.

> > > This looks much better from the WAL size perspective, there's now almost
> > > no additional WAL. However, that is because pgbench doesn't do TOAST, so
> > > in a real-world example it might still be quite larger. Also, the vacuum
> > > runtime is still 15x longer.
> >
> > That's obviously an issue.
>
> It'd certainly be nice to figure out a way to improve the VACUUM run but
> I don't think the impact on the time to run VACUUM is really a good
> reason to not move forward with changing the default.

Vacuum performance is one of *THE* major complaints about
postgres. Making it run slower by a lot obviously exascerbates that
problem significantly. I think it'd be prohibitively expensive if it
were 1.5x, not to even speak of 15x.

> imv, enabling page checksums is akin to having fsync enabled by default.
> Does it impact performance? Yes, surely quite a lot, but it's also the
> safe and sane choice when it comes to defaults.

Oh for crying out loud.

Greetings,

Andres Freund