Re: Regression tests fail with musl libc because libpq.so can't be loaded

On Wed, Mar 20, 2024 at 08:29:21PM +0100, Wolfgang Walther wrote:
> Bruce Momjian:
> > > The library/postgres docker image has been pulled about 8 billion times
> > > since 2014 [1]. While we can't really tell how many of those pulled the
> > > alpine variant of the image, comparing the alpine [2] and ubuntu/debian
> > > [3,4] base images gives a rough estimate of >50% using alpine in general.
> >
> > Uh, what is the current behavior of Postgres on musl? It just fails if
> > the process title is longer than argv[0] plus the environment space to
> > the LD_ environment variable, and then linking fails for certain
> > extensions? If there are many downloads, why would we only be getting
> > this report now?
>
> The process title works fine. It's just the way how space is cleared for the
> process title, that is causing problems elsewhere.
>
> The thing that is broken when running postgres on alpine/musl is, to put
> libpq in a custom path and use LD_LIBRARY_PATH to find it when loading
> libpqwalreceiver (+ some contrib modules). Nobody does that, especially not
> in a container environment where postgres is likely the only thing running
> in that container, so there is no point in using any custom library paths or
> anything - the image is built once and made to work, and everybody else is
> just using that working image.
>
> The much more practical problem is that the test suite doesn't run, because
> it makes use of LD_LIBRARY_PATH for that purpose. In the past, the packagers
> for alpine only disabled the failing tests, but IIRC they have given up on
> that and just disabled the whole test suite by now.

Thanks, that is very clear.

--
Bruce Momjian <bruce(at)momjian(dot)us> https://momjian.us
EDB https://enterprisedb.com

Only you can decide what is important to you.