Horton

Re: First draft of PG 17 release notes

Peter Geoghegan <pg@bowt.ie>

View thread
From: Peter Geoghegan <pg@bowt.ie>
To: Andres Freund <andres@anarazel.de>
Cc: Alvaro Herrera <alvherre@alvh.no-ip.org>, Bruce Momjian <bruce@momjian.us>, Melanie Plageman <melanieplageman@gmail.com>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2024-05-16T14:55:25Z
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. Revert support for ALTER TABLE ... MERGE/SPLIT PARTITION(S) commands

  2. When creating materialized views, use REFRESH to load data.

  3. Revert temporal primary keys and foreign keys

  4. Avoid needless large memcpys in libpq socket writing

  5. Enhance nbtree ScalarArrayOp execution.

  6. Introduce a non-recursive JSON parser

  7. Combine freezing and pruning steps in VACUUM

  8. Allow SIGINT to cancel psql database reconnections.

  9. Provide API for streaming relation data.

  10. Add hash support functions and hash opclass for contrib/ltree.

  11. Pull up ANY-SUBLINK with the necessary lateral support.

  12. Read WAL directly from WAL buffers.

  13. Introduce the dynamic shared memory registry.

  14. Add macros for looping through a List without a ListCell.

  15. Support +/- infinity in the interval data type.

  16. Extend ALTER OPERATOR to allow setting more optimization attributes.

  17. Consider cheap startup paths in add_paths_to_append_rel

On Wed, May 15, 2024 at 11:48 PM Andres Freund <andres@anarazel.de> wrote:
> On 2024-05-15 10:38:20 +0200, Alvaro Herrera wrote:
> > I disagree with this.  IMO the impact of the Sawada/Naylor change is
> > likely to be enormous for people with large tables and large numbers of
> > tuples to clean up (I know we've had a number of customers in this
> > situation, I can't imagine any Postgres service provider that doesn't).
> > The fact that maintenance_work_mem is no longer capped at 1GB is very
> > important and I think we should mention that explicitly in the release
> > notes, as setting it higher could make a big difference in vacuum run
> > times.
>
> +many.

TIDStore/the lifting of the maintenance_work_mem cap is likely to make
the performance of VACUUM a lot more predictable, overall. While most
VACUUM operations don't hit the limit, the limit is disproportionately
involved in cases where (for whatever reason) vacuuming becomes a long
and painful process. Even if you as a user never run into such a
problem, you still spend time worrying about it, and/or taking
measures to make sure it doesn't affect you.

The justification for not including mention of these items is that
they're not very relevant to users. I find that hard to square with
what does get included. For example, the "Source Code" section is full
of highly niche items. Items that are low impact, even for users
that'll benefit the most. Also, "Monitoring" often mentions monitoring
improvements that expose low-level implementation details (e.g. SLRU
statistics), even though there's a good chance that Bruce wouldn't
include an item for some improvement to the SLRU subsystem itself.

If somebody puts in an enormous amount of effort to get a big
performance improvement over the line, then ISTM that that effort is a
useful signal when the time comes to write the release notes (at least
up to a point). For example, Masahiko and John spent about 2 years on
the TIDStore thing, on and off. These things do not happen in a vacuum
(no pun intended). Common sense tells me that they went to those
lengths precisely because they understood that it very much was
relevant to users. That belief would have been reinforced by both
experience, and by discussion on the list during the development of
the feature.

To be fair to Bruce, it probably really is true that most individual
users won't care about (say) TIDStore. But it's probably also true
that most individual users don't care about the release notes, or at
most skim the major items.

-- 
Peter Geoghegan