From: | Pavel Borisov <pashkin(dot)elfe(at)gmail(dot)com> |
---|---|
To: | Aleksander Alekseev <aleksander(at)timescale(dot)com> |
Cc: | Postgres hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org> |
Subject: | Re: Add 64-bit XIDs into PostgreSQL 15 |
Date: | 2022-01-28 14:43:06 |
Message-ID: | CALT9ZEHvtJ1a7OzFydyHRz631e_40+w+ad7iu9N21Rey80W1cw@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
>
> Did you consider using 4 bytes for pd_xid_base and another 4 bytes for
> (pd_xid_base/pd_multi_base)? This would allow calculating XMIN/XMAX
> as:
>
> XMIN = (t_min_extra_bits << 32) | t_xmin
> XMAX = (t_max_extra_bits << 32) | t_xmax
>
> ... and save 8 extra bytes in the pd_special area. Or maybe I'm
> missing some context here?
>
Hi, Alexander!
In current design it is not possible, as pd_xid_base is roughly just a
minimum 64-xid of all tuples that may fit this page. So we do not make any
extra guess that it should be in multiples of 2^32.
If we make pd_xid_base in multiples of 2^32 then after current XID crosses
the border of 2^32 then pages that contains tuples with XMIN/XMAX before
this point are not suitable for tuple inserts anymore. In effect we will
then have "sets" of the pages for each 2^32 "epoch" with freed space that
can not be used anymore.
I think it's too big a loss for gain of just 8 bytes per page.
Thank you for your dive into this matter!
--
Best regards,
Pavel Borisov
Postgres Professional: http://postgrespro.com <http://www.postgrespro.com>
From | Date | Subject | |
---|---|---|---|
Next Message | Andrew Dunstan | 2022-01-28 14:46:02 | Re: A test for replay of regression tests |
Previous Message | Robert Haas | 2022-01-28 14:42:17 | Re: warn if GUC set to an invalid shared library |