Re: CREATE DATABASE with filesystem cloning - Mailing list pgsql-hackers

Hi,

On Tue, 21 May 2024 at 15:08, Ranier Vilela <ranier.vf@gmail.com> wrote:
>
> Em ter., 21 de mai. de 2024 às 05:18, Nazir Bilal Yavuz <byavuz81@gmail.com> escreveu:
>>
>> Hi,
>>
>> On Thu, 16 May 2024 at 17:47, Robert Haas <robertmhaas@gmail.com> wrote:
>> >
>> > On Thu, May 16, 2024 at 9:43 AM Nazir Bilal Yavuz <byavuz81@gmail.com> wrote:
>> > > Actually, the documentation for the file_copy_method was mentioning
>> > > the things it controls; I converted it to an itemized list now. Also,
>> > > changed the comment to: 'Further uses of this function requires
>> > > updates to the list that GUC controls in its documentation.'. v7 is
>> > > attached.
>> >
>> > I think the comments need some wordsmithing.
>>
>> I changed it to 'Uses of this function must be documented in the list
>> of places affected by this GUC.', I am open to any suggestions.
>>
>> > I don't see why this parameter should be PGC_POSTMASTER.
>>
>> I changed it to 'PGC_USERSET' now. My initial point was the database
>> or tablespace to be copied with the same method. I thought copying
>> some portion of the database with the copy and rest with the clone
>> could cause potential problems. After a bit of searching, I could not
>> find any problems related to that.
>>
>> v8 is attached.
>
> Hi,
> I did some research on the subject and despite being an improvement,
> I believe that some terminologies should be changed in this patch.
> Although the new function is called *clone_file*, I'm not sure if it really is "clone".
> Why MacOS added an API called clonefile [1] and Linux exists
> another called FICLONE.[2]
> So perhaps it should be treated here as a copy and not a clone?
> Leaving it open, is the possibility of implementing a true clone api?
> Thoughts?

Thank you for mentioning this.

1- I do not know where to look for MacOS' function definitions but I
followed the same links you shared. It says copyfile(...,
COPYFILE_CLONE_FORCE) means 'Clone the file instead. This is a force
flag i.e. if cloning fails, an error is returned....' [1]. It does the
cloning but I still do not know whether there is a difference between
'copyfile() function with COPYFILE_CLONE_FORCE flag' and 'clonefile()'
function.

2- I read a couple of discussions about copy_file_range() and FICLONE.
It seems that the copy_file_range() function is a slightly upgraded
version of FICLONE but less available since it is newer. It still does
the clone [2]. Also, FICLONE is already used in pg_upgrade and
pg_combinebackup.

Both of these copyfile(..., COPYFILE_CLONE_FORCE) and
copy_file_range() functions do the cloning, so I think using clone
terminology is correct. But, using FICLONE instead of
copy_file_range() could be better since it is more widely available.

[1] https://www.unix.com/man-page/mojave/3/copyfile/
[2] https://elixir.bootlin.com/linux/v5.15-rc5/source/fs/read_write.c#L1495

--
Regards,
Nazir Bilal Yavuz
Microsoft