maria-discuss team mailing list archive

[Gordan Bobic <gordan.bobic@xxxxxxxxx>]

On Fri, Dec 9, 2022 at 2:40 AM Reindl Harald <h.reindl@xxxxxxxxxxxxx> wrote:

> >>> 2.2) Database doesn't crash because the damage merely corrupts a
> >>> single value but the record structure remains sound.
> >>>
> >>> So it is that 2.2) point where the InnoDB checksum gives you anything
> >>
> >> moron it don't matter if you find it useful - the whole point was that
> >> you pretended the filesystem can do the same with it's checksums which
> >> is nonsense
> >
> > You are conveniently ignoring the fact that in the vast majority of
> > cases what InnoDB checksums will catch is silent disk corruption
> > rather than database internals corruption.
>
> i ignore nothing but filesystem corruption is still a different topic
>
> > So the one narrow edge case you are clinging to as the full
> > justification of your abusive behaviour and delusions of grandeur are
> > a tiny fraction of a percent of the errors that will cause InnoDB
> > checksums to fail - and outside that narrow edge case all of the rest
> > of them will be caught and handled better at layers below the database
> > itself.
>
> the topic was and still is "Some of us run MariaDB on file systems that
> do their own block checksumming, and thus run
> innodb_checksum_algorithm=none" where you mix two completly independent
> layers
>
> > So either you are arguing in bad faith, or you really are extensively
> > ignorant of typical failure patterns.
>
> the topic was and still is "Some of us run MariaDB on file systems that
> do their own block checksumming, and thus run
> innodb_checksum_algorithm=none" where you mix two completly independent
> layers

They are different layers, but 99.9%+ of corruption that InnoDB
checksums ever detects occur in the storage layer, not in the database
internals layer.
So in terms of the overall picture of the corruption cause probability
landscape which you seem to be struggling to see, you are about 0.1%
correct. I'll grant you that.

I'll go as far as hazarding a guess that InnoDB checkums were
originally added with the main motivation of detecting disk corruption
rather than internals debugging. Unfortunately, the code tree back in
3.23.24 (as far as I can tell the release where InnoDB was merged)
doesn't seem to contain any annotations on the subject that might shed
light on the original motivation.