Re: Draft for basic NUMA observability

On 4/9/25 17:51, Andres Freund wrote:
> Hi,
> 
> On 2025-04-09 17:28:31 +0200, Tomas Vondra wrote:
>> On 4/9/25 17:14, Andres Freund wrote:
>>> I'd mention that the includes of postgres.h/fmgr.h is what caused missing
>>> build-time dependencies and via that failures on buildfarm member dogfish.
>>>
>>
>> Not really, I also need to include "c.h" instead of "postgres.h" (which
>> is also causing the same failure).
> 
> I did mention postgres.h :)
> 

D'oh, I missed that. I was focused on the fmgr one.

> 
> 
>>> I think this may not be needed anymore, that was just there for
>>> GetSystemInfo(), right?  Conversely, I suspect it may now be needed in the new
>>> location of pg_numa_get_pagesize()?
>>>
>>
>> Good question. But if it's needed there, shouldn't it have failed on CI?
> 
> Oh. No. It shouldn't have - because that include was completely
> unnecessary. We always include windows.h on windows.
> 

Makes sense. I'll get rid of the windows.h include.

> 
> 
>>>> From 201f8be652e9344dfa247b035a66e52025afa149 Mon Sep 17 00:00:00 2001
>>>> From: Tomas Vondra <tomas@vondra.me>
>>>> Date: Wed, 9 Apr 2025 13:29:31 +0200
>>>> Subject: [PATCH 2/2] ci: Check for missing dependencies in meson build
>>>>
>>>> Extends the meson build on Debian to also check for missing dependencies
>>>> by executing
>>>>
>>>>     ninja -t missingdeps
>>>>
>>>> right after the build. This highlights unindended dependencies.
>>>>
>>>> Reviewed-by: Andres Freund <andres@anarazel.de>
>>>> https://postgr.es/m/CALdSSPi5fj0a7UG7Fmw2cUD1uWuckU_e8dJ+6x-bJEokcSXzqA@mail.gmail.com
>>>
>>> FWIW, while I'd prefer it as a meson.build visible test(), I think it's ok to
>>> have it just in CI until we have that.  I would however also add it to the
>>> windows job, as that's the most "different" type of build / source of missed
>>> dependencies that wouldn't show up on our development systems.
>>>
>>
>> We can add it as a meson.build test, sure. I was going for the CI first,
>> because then it fires no matter what build I do locally (I'm kinda still
>> used to autotools).
> 
> A meson test would do the same thing, it'd fail while running the tests, no?
> 

Sure, but only if you use meson. Which I still mostly don't, so I've
been thinking about the CI first, because I use that very consistently
before pushing something.

> 
>> If you agree adding it to build_script is the right way to do that, I'll
>> do the same thing for the windows job.
> 
> WFM.

Thanks. I'll polish this a bit more and push.


regards

-- 
Tomas Vondra