Re: LISTEN/NOTIFY bug: VACUUM sets frozenxid past a xid in async queue

On Sun Oct 19, 2025 at 2:14 PM -03, Joel Jacobson wrote:
> On Fri, Oct 17, 2025, at 15:51, Álvaro Herrera wrote:
>> I have the impression that this thread has lost focus on the idea of
>> producing a backpatchable bugfix.  The last proposed patch has a lot of
>> new mechanism that doesn't seem suitable for backpatch.  I could be
>> wrong of course.
>
> I've tried to create a minimal isolated fix, hopefully suitable for
> backpatching, with no new mechanisms, other than the added
> GetOldestQueuedNotifyXid used by vac_update_datfrozenxid.
>
> It's based on the approach discussed earlier in this thread, that just
> goes through the notification queue from QUEUE_TAIL to QUEUE_HEAD, to
> find the oldestXid in the current database.
>
> Implementation:
>
> * Break out SLRU read page code from asyncQueueReadAllNotifications into
>   new helper-function asyncQueueReadPageToBuffer.
>
> * Add GetOldestQueuedNotifyXid which uses the new helper-function
>   asyncQueueReadPageToBuffer.
>
> It passes the 001_xid_freeze.pl test, not included in this patch.
>
This is similar to what was already proposed at [1]. This approach was
abandoned because a notification on the queue may block datfrozenxid
advance and clog truncation which can cause other issues for the users [2].


[1] https://www.postgresql.org/message-id/CAFY6G8cJm73_MM9SuynZUqtqcaTuepUDgDuvS661oLW7U0dgsg%40mail.gmail.com
[2] https://www.postgresql.org/message-id/d186fba0-dc65-4274-aa96-3906bbb2e530%40postgrespro.ru

--
Matheus Alcantara