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

On Thu, Jan 7, 2021 at 3:32 AM Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
>
> On Wed, Jan 6, 2021 at 1:29 PM Michael Banck <michael(dot)banck(at)credativ(dot)de> wrote:
> > That one seems to be 5min everywhere, and one can change it except on
> > Azure.
>
> Okay, thanks for clearing that up. Looks like all of the big 3 cloud
> providers use Postgres checksums in a straightforward way.
>

But they might have done something to reduce the impact of enabling
checksums like by using a different checksum (for data blocks) and or
compression (for WAL) technique.

> I don't have much more to say on this thread. I am -1 on the current
> proposal to enable page-level checksums by default.
>

-1 from me too with the current impact on performance and WAL it can
have. I was looking at some old thread related to this topic and came
across the benchmarking done by Tomas Vondra [1]. It clearly shows
that enabling checksums can have a huge impact on init time, WAL, and
TPS.

Having said that, if we really want to enable checksums, can't we
think of improving performance when it is enabled? I could think of
two things to improve (a) a better algorithm for wal compression
(currently we use pglz), this will allow us to enable wal_compression
at least when data_checksums are enabled (b) a better technique for
checksums to reduce the cost of PageSetChecksumCopy. I don't have good
ideas to offer to improve things in these two areas but I think it is
worth investigating if we want to enable checksums.

[1] - https://www.postgresql.org/message-id/20190330192543.GH4719%40development

--
With Regards,
Amit Kapila.