Date: Sat, 1 Oct 2022 19:58:01 -0500
From: Justin Pryzby <pryzby@telsasoft.com>
To: Andres Freund <andres@anarazel.de>
Cc: Andrew Dunstan <andrew@dunslane.net>,
	Thomas Munro <thomas.munro@gmail.com>,
	pgsql-hackers@lists.postgresql.org, Noah Misch <noah@leadboat.com>,
	Michael Paquier <michael@paquier.xyz>,
	Anastasia Lubennikova <lubennikovaav@gmail.com>,
	Tom Lane <tgl@sss.pgh.pa.us>, Robert Haas <robertmhaas@gmail.com>,
	Melanie Plageman <melanieplageman@gmail.com>,
	Peter Eisentraut <peter.eisentraut@enterprisedb.com>,
	Daniel Gustafsson <daniel@yesql.se>
Subject: Re: CI and test improvements
Message-ID: <20221002005801.GB7745@telsasoft.com>
References: <20220708002232.GF13040@telsasoft.com>
 <20220623193125.GB22452@telsasoft.com>
 <20220528153741.GK19626@telsasoft.com>
 <20220828144447.GA21897@telsasoft.com>
 <20220828160752.l5l66k3eptokzhzj@awork3.anarazel.de>
 <20220828171029.GO2342@telsasoft.com>
 <20220828212802.r6eymfffrgr3lxxt@awork3.anarazel.de>
 <20220910200542.GX31833@telsasoft.com>
 <20221002004501.w4ilvthshmuz6642@awork3.anarazel.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20221002004501.w4ilvthshmuz6642@awork3.anarazel.de>
User-Agent: Mutt/1.9.4 (2018-02-28)
Archived-At: 
 <https://www.postgresql.org/message-id/20221002005801.GB7745%40telsasoft.com>
Precedence: bulk

On Sat, Oct 01, 2022 at 05:45:01PM -0700, Andres Freund wrote:
> Hi,
> 
> On 2022-09-10 15:05:42 -0500, Justin Pryzby wrote:
> > From 4ed5eb427de4508a4c3422e60891b45c8512814a Mon Sep 17 00:00:00 2001
> > From: Justin Pryzby <pryzbyj@telsasoft.com>
> > Date: Sun, 3 Apr 2022 00:10:20 -0500
> > Subject: [PATCH 03/23] cirrus/ccache: disable compression and show stats
> > 
> > Since v4.0, ccache enables zstd compression by default, saving roughly
> > 2x-3x.  But, cirrus caches are compressed as tar.gz, so we could disable
> > ccache compression, allowing cirrus to gzip the uncompressed data
> > (better than ccache's default of zstd-1).
> 
> I wonder whether we could instead change CCACHE_COMPRESSLEVEL (maybe 3, zstd's
> default IIRC). It'd be good if we could increase cache utilization.

I considered that (and I think that's what I wrote initially).

I figured that if cirrus is going to use gzip-6 (tar.gz) in any case, we
might as well disable compression.  Then, all the tasks are also doing
the same thing (half the tasks have ccache before 4.0).

> > From 0bd5f51b8c143ed87a867987309d66b8554b1fd6 Mon Sep 17 00:00:00 2001
> > From: Justin Pryzby <pryzbyj@telsasoft.com>
> > Date: Thu, 14 Apr 2022 06:27:07 -0500
> > Subject: [PATCH 05/23] cirrus: enable various runtime checks on macos and
> >  freebsd
> > 
> > windows is slower than freebsd and mac, so it's okay to enable options which
> > will slow them down some.  Also, the cirrusci mac instances always have lot of
> > cores available.
> 
> > See:
> > https://www.postgresql.org/message-id/20211217193159.pwrelhiyx7kevgsn@alap3.anarazel.de
> > https://www.postgresql.org/message-id/20211213211223.vkgg3wwiss2tragj%40alap3.anarazel.de
> > https://www.postgresql.org/message-id/CAH2-WzmevBhKNEtqX3N-Tkb0gVBHH62C0KfeTxXzqYES_PiFiA%40mail.gmail.com
> > https://www.postgresql.org/message-id/20220325000933.vgazz7pjk2ytj65d@alap3.anarazel.de
> > 
> > ci-os-only: freebsd, macos
> > ---
> >  .cirrus.yml | 8 +++++---
> >  1 file changed, 5 insertions(+), 3 deletions(-)
> > 
> > diff --git a/.cirrus.yml b/.cirrus.yml
> > index 183e8746ce6..4ad20892eeb 100644
> > --- a/.cirrus.yml
> > +++ b/.cirrus.yml
> > @@ -113,7 +113,9 @@ task:
> >          \
> >          CC="ccache cc" \
> >          CXX="ccache c++" \
> > -        CFLAGS="-Og -ggdb"
> > +        CPPFLAGS="-DRELCACHE_FORCE_RELEASE -DCOPY_PARSE_PLAN_TREES -DWRITE_READ_PARSE_PLAN_TREES -DRAW_EXPRESSION_COVERAGE_TEST" \
> > +        CXXFLAGS="-Og -ggdb -march=native -mtune=native" \
> > +        CFLAGS="-Og -ggdb -march=native -mtune=native"
> 
> What's reason for -march=native -mtune=native here?

No particular reason, and my initial patch didn't have it.
I suppose I added it to test its effect and never got rid of it.

> >      EOF
> >    build_script: |
> >      su postgres -c "ccache --zero-stats"
> > @@ -336,8 +338,8 @@ task:
> >        CC="ccache cc" \
> >        CXX="ccache c++" \
> >        CLANG="ccache ${brewpath}/llvm/bin/ccache" \
> > -      CFLAGS="-Og -ggdb" \
> > -      CXXFLAGS="-Og -ggdb" \
> > +      CFLAGS="-Og -ggdb -DRANDOMIZE_ALLOCATED_MEMORY" \
> > +      CXXFLAGS="-Og -ggdb -DRANDOMIZE_ALLOCATED_MEMORY" \
> >        \
> >        LLVM_CONFIG=${brewpath}/llvm/bin/llvm-config \
> >        PYTHON=python3
> 
> I'd also use CPPFLAGS here, given you'd used it above...
> 
> I'm planning to commit an updated version of this change soon, without the
> -march=native -mtune=native bit, unless somebody protests...

One other thing is that your -m32 changes caused the linux/meson task to
take an additional 3+ minutes (total ~8).  That's no issue, except that
the Warnings task depends on the linux/mason task, and itself can take
up to 15 minutes.

So those two potentially take as long as the windows task.
I suggested that CompileWarnings could instead "Depend on: Freebsd",
which currently takes 6-7min (and could take 4-5min if given more CPUs).

-- 
Justin