Re: Sequence Access Methods, round two

On Fri, 28 Mar 2025 at 09:31, Michael Paquier <michael(at)paquier(dot)xyz> wrote:
>
> On Tue, Feb 25, 2025 at 04:25:37PM +0900, Michael Paquier wrote:
> > Rebased.
>
> Conflict in pg_class.h. Rebased.
> --
> Michael

Looks like we entered a rebase loop here. I'm really interested in
moving this forward; furthermore, I am inclined to include these
patches into our PostgreSQL fork with modifications that we use for
sharding and cloud.

So here is my 2c:

At first, general thought: Access Method is a term we use for
relations. Table access method is something that stores data and
knowledge on how to read this data, and index access method is
something that hints what data we should retrieve using some predicate
(or quals). But sequence... is generating data primitive? So, maybe
not Sequence access method but Sequence generate method? OTOH this
patch provides a way for generating sequence values by _accessing_
remote storage or procedures, so maybe we are fine with this wording.

patches:

0003:
Patch uses "local" wording for build-in sequence, which is not
precise. As I understand, snowflake sequence state is local too.

0004:

> + /*
> + * Retrieve table access method used by a sequence to store its metadata.
> + */
> + const char *(*get_table_am) (void);

Looks like we force the sequence to have some table storage. I can
imagine cases where local storage is not used (and thus, not logged),
so can we have this optional?

0007:
So, we use generic xlog for logging very small actual changes. I
understand this is out of topic here, but can we add a second option
for generic log to store FormData_snowflake_data which is a few bytes?
Maybe we should start different thread for that

--
Best regards,
Kirill Reshke