Re: CREATE DATABASE with filesystem cloning

Hi,

On Thu, 9 May 2024 at 19:29, Robert Haas <robertmhaas@gmail.com> wrote:
>
> On Wed, May 8, 2024 at 10:34 AM Nazir Bilal Yavuz <byavuz81@gmail.com> wrote:
> > > I'm not so sure about the GUC name. On the one hand, it feels like
> > > createdb should be spelled out as create_database, but on the other
> > > hand, the GUC name is quite long already. Then again, why would we
> > > make this specific to CREATE DATABASE in the first place? Would we
> > > also want alter_tablespace_file_copy_method and
> > > basic_archive.file_copy_method?
> >
> > I agree that it is already quite long, because of that I chose the
> > createdb as a prefix. I did not think that file cloning was planned to
> > be used in other places. If that is the case, does something like
> > 'preferred_copy_method' work? Then, we would mention which places will
> > be affected with this GUC in the docs.
>
> I would go with file_copy_method rather than preferred_copy_method,
> because (1) there's nothing "preferred" about it if we're using it
> unconditionally and (2) it's nice to clarify that we're talking about
> copying files rather than anything else.

Done.

>
> My personal enthusiasm for making platform-specific file copy methods
> usable all over PostgreSQL is quite limited. However, it is my
> observation that other people seem far more enthusiastic about it than
> I am. For example, consider how quickly it got added to
> pg_combinebackup. So, I suspect it's smart to plan on anything we add
> in this area getting used in a bunch of places. And perhaps it is even
> best to think about making it work in all of those places right from
> the start. If we build support into copydir and copy_file, then we
> just get everything that uses those, and all that remains is to
> document was is covered (and add comments so that future patch authors
> know they should further update the documentation).

I updated the documentation and put a comment on top of the copydir()
function to inform that further changes and uses of this function may
require documentation updates.

I also changed O_RDWR to O_WRONLY flag in the clone_file() function
based on Raniers' feedback. Also, does this feature need tests? I
thought about possible test cases but since this feature requires
specific file systems to run, I could not find any.

v6 is attached.

--
Regards,
Nazir Bilal Yavuz
Microsoft