Re: Use read streams in pg_visibility

Hi,

On Thu, 5 Sept 2024 at 18:54, Noah Misch <noah(at)leadboat(dot)com> wrote:
>
> On Thu, Sep 05, 2024 at 03:59:53PM +0300, Nazir Bilal Yavuz wrote:
> > On Wed, 4 Sept 2024 at 21:43, Noah Misch <noah(at)leadboat(dot)com> wrote:
> > > https://postgr.es/m/CAEudQAozv3wTY5TV2t29JcwPydbmKbiWQkZD42S2OgzdixPMDQ@mail.gmail.com
> > > then observed that collect_corrupt_items() was now guaranteed to never detect
> > > corruption. I have pushed revert ddfc556 of the pg_visibility.c changes. For
> > > the next try, could you add that testing we discussed?
> >
> > Do you think that corrupting the visibility map and then seeing if
> > pg_check_visible() and pg_check_frozen() report something is enough?
>
> I think so. Please check that it would have caught both the blkno bug and the
> above bug.

The test and updated patch files are attached. In that test I
overwrite the visibility map file with its older state. Something like
this:

1- Create the table, then run VACUUM FREEZE on the table.
2- Shutdown the server, create a copy of the vm file, restart the server.
3- Run the DELETE command on some $random_tuples.
4- Shutdown the server, overwrite the vm file with the #2 vm file,
restart the server.
5- pg_check_visible and pg_check_frozen must report $random_tuples as corrupted.

Do you think this test makes sense and enough?

--
Regards,
Nazir Bilal Yavuz
Microsoft