Commitfest 2021-11 Patch Triage - Part 3 - Mailing list pgsql-hackers

From Daniel Gustafsson
Subject Commitfest 2021-11 Patch Triage - Part 3
Date
Msg-id CDAE89A0-DD74-44C6-8D6F-AB9FF750737B@yesql.se
Whole thread Raw
Responses Re: Commitfest 2021-11 Patch Triage - Part 3
List pgsql-hackers
Welcome to the third installment of the CF 2021-11 patch triage, looking at the
patches which have been in four CF's and parts of the ones with three CF's.  As
per usual I might very well have misunderstood something (or many things), but
there seem to be a fair few ones which should be closeable in this CF here.


2962: Fix DROP TABLESPACE on Windows with ProcSignalBarrier?
============================================================
This fixes an old bug which can prevent DROP TABLESPACE from working on
Windows.  There has been a fair bit of review and testing, by among other
myself, and I feel this patch is in good enough shape to be considered for
Ready for Committer.  There is a goalpost movement request in the thread for
DROP DATABASE support, but I think we can get this in in two commits to move us
forward.  What are your thoughts on this one Thomas?


2979: Support tab completion for upper character inputs in psql
===============================================================
There is broad concensus that this patch fixes a less than ideal behavior in
psql, and the patch has seen a lot of review and rework.  It currently needs a
review of the latest version, but skimming it and judging by the thread I'd say
it's close to Ready for Committer (if not already).


2942: increase size of pg_commit_ts buffers
===========================================
Production use has validated this patch as fixing a performance regression, and
the patch has support from another committer.  There is mention of patch 2627
(More scalable multixacts buffers and locking) in the thread as an alternative
approach, that buffers potentially should be configurable instead.  Since that
patch seems like further out from landing, should we go ahead with this to have
at least a partial solution for 15?  Do have any plans to go ahead with this
patch Álvaro or are you awaiting results on the configurable approach?


2992: Allow batched insert during cross-partition updates
=========================================================
This is a follow-up to 927f453a94106 to support batched inserts of internal
INSERT commands which are used in cross-partition updates for local/remote
partitions.  The patch was after a lot of review initially marked Ready for
Committer but went to WOA after failing to apply.  Once updated it went to Need
Review which IMO is misleading given it's status and a skim so I'm moving it to
RFC again.  Georgios, having reviewed it earlier do you any thoughts on the
latest version?


2970: Set default transactions to read-only at servers start in pg_upgrade
==========================================================================
Goalpost movement saw this patch go from read-only transactions to stopping
bgworkers from running at all during pg_upgrade.  There is agreement on the
approach, but discussion on the exact implementation (which has since trailed
off).  I have on my TODO to bring this over the finish line.


2948: Simplify some RI checks to reduce SPI overhead
====================================================
Marked ready for committer it seems that raised concerns have been addressed.
Peter, being listed as reviewer do you have any thoughts on this?


2957: Identify missing publications from publisher while create/alter
subscription.
=====================================================================
Patch is being actively reviewed, has been kept up to date, and was marked RFC
back in May.  Hopefully we can close this soon then.


2968: Minimal logical decoding on standbys (take 6)
===================================================
The latest revision of the patch has had feedback from Robert (and Andres)
which needs to be addressed, and the patch no longer applies (after the recent
ThisTimeLineID changes perhaps?).  I'm moving this patch to WoA for now.


2999: Catalog version access
============================
The initial patch in this thread was dismissed for not providing enough
information to solve the stated problem, but there has since been a patch
posted which performs ReadControlFile as recommended in the thread.  Vik, can
you confirm if this patch solves your original problem?


2996: Speed up verifying UTF-8
==============================
An IMO interesting approach to speeding up input validation with promising
results posted.  The discussion seem to be quite confident with the patch, but
the latest version takes out a portion of the previous patch which was deemed
overly complicated.  Heikki, are you up for reviewing this version?


2932: Partial foreign key updates in referential integrity triggers
===================================================================
A recently posted update from Peter indicates that this is close to being
committable, so fingers crossed.


3025: postgres_fdw: suppress explicit casts in text:text comparisons
====================================================================
Currently undergoing active review and receiving updates with a new version
posted today by Tom.  Hopefully it can be closed in this CF.


2947: Full support for index LP_DEAD hint bits on standby
=========================================================
This has potential to be a nice performance boost for a common usecase for
scalable production setups.  The patch changed course half-way through but has
since been marked Ready for Committer.  Any takers on this?


3120: Rewriting the test of pg_upgrade as a TAP test - take three
=================================================================
Currently this is stuck on a discussion around how to deal with test.sh and the
effects on the buildfarm.  It seems like the questions raised have been
answered, but we need to settle on how to continue.


3230: Column filtering in logical replication
=============================================
This fails badly to apply, which is likely because there has been a lot of
discussion on the grammar making a new patch not really an option until
settled.  That discussion ended in late September so hopefully there is a new
version during this CF, else we'll have to close this with RwF awaiting a new
patch in a later CF.


3223: Delegating superuser tasks to new security roles
======================================================
There is a lot of ongoing discussion on this one, but skimming the thread it's
not really clear (to me) where it leaves the patch in question.


3138: Support for NSS as a libpq TLS backend
============================================
I'm clearly fairly biased here but I hope that at 48 revisions in the patch is
in good enough shape to be close to RFC.  Joshua Brindle has recently picked up
reviewing it so hope we can reach somewhere for 15 with this work.


3035: Removing unused trailing linepointers from heap pages
===========================================================
An alternative to this was committed for 14, which led the author to change the
patch into a follow-up to that commit.  The merit and value of this second
patch is debated, with seemingly good arguments both for and against.  There
was an ask for testcases to provide evidence of correctness back in August
which has gone unanswered, so unless this moves during this CF I think we
should mark it RWF awaiting a new version of the patch.


3058: Eliminating "Permission denied" errors on stat() (on Windows)
===================================================================
This was committed back in July, and then promptly reverted as there were
issues with mingw.  There has since been a patch posted addressing the problem
which needs review.  Michael, having looked at the earlier version, do you feel
up for taking a look at the reworked one?


3172: Consistent use of SSL/TLS in docs
=======================================
The proposal to use SSL/TLS instead of just SSL in the docs came about after a
small discussion in the NSS thread, but it wasn't to anyones liking so the
reworked patch just adds notes that SSL and TLS are used interchangeably in the
docs.  This version is a pretty small addition that I think we should go ahead
with, but more eyes are needed to validate that.

--
Daniel Gustafsson        https://vmware.com/




pgsql-hackers by date:

Previous
From: Isaac Morland
Date:
Subject: Re: Should AT TIME ZONE be volatile?
Next
From: Michael Paquier
Date:
Subject: Re: Clean up build warnings of plperl with clang-12+