Horton

Re: Add 64-bit XIDs into PostgreSQL 15

Peter Eisentraut <peter@eisentraut.org>

View thread
From: Peter Eisentraut <peter@eisentraut.org>
To: Aleksander Alekseev <aleksander@timescale.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Cc: Maxim Orlov <orlovmg@gmail.com>, Shubham Khanna <khannashubham1197@gmail.com>, Daniel Gustafsson <daniel@yesql.se>, Pavel Borisov <pashkin.elfe@gmail.com>, Robert Haas <robertmhaas@gmail.com>, Chris Travers <chris@orioledata.com>, Bruce Momjian <bruce@momjian.us>, Chris Travers <chris.travers@gmail.com>, Peter Geoghegan <pg@bowt.ie>, Fedor Sigaev <teodor@sigaev.ru>, Alexander Korotkov <aekorotkov@gmail.com>, Konstantin Knizhnik <knizhnik@garret.ru>, Nikita Glukhov <n.gluhov@postgrespro.ru>, Yura Sokolov <y.sokolov@postgrespro.ru>
Date: 2024-07-25T10:19:39Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Add SLRU tests for 64-bit page case

  2. Make use FullTransactionId in 2PC filenames

  3. Use larger segment file names for pg_notify

  4. Index SLRUs by 64-bit integers rather than by 32-bit integers

On 23.07.24 11:13, Aleksander Alekseev wrote:
>> Here is the fix. It can be tested like this:
>> [...]
> 
> PFA the rebased patchset.

I'm wondering about the 64-bit GUCs.

At first, it makes sense that if there are settings that are counted in 
terms of transactions, and transaction numbers are 64-bit integers, then 
those settings should accept 64-bit integers.

But what is the purpose and effect of setting those parameters to such 
huge numbers?  For example, what is the usability of being able to set

vacuum_failsafe_age = 500000000000

I think in the world of 32-bit transaction IDs, you can intuitively 
interpret most of these "transaction age" settings as "percent toward 
disaster".  For example,

vacuum_freeze_table_age = 150000000

is 7% toward disaster, and

vacuum_failsafe_age = 1600000000

is 75% toward disaster.

However, if there is no more disaster threshold at 2^31, what is the 
guidance for setting these?  Or more radically, why even run 
transaction-count-based vacuum at all?

Conversely, if there is still some threshold (not disaster, but 
efficiency or something else), would it still be useful to keep these 
settings well below 2^31?  In which case, we might not need 64-bit GUCs.

Your 0004 patch adds support for 64-bit GUCs but doesn't actually 
convert any existing GUCs to use that.  (Unlike the reloptions, which 
your patch coverts.)  And so there is no documentation about these 
questions.