On 12/30/20 4:18 PM, Insurgo wrote:
> Hello there,
>
> Since a couple of days, it seems that http://ftpmirror.gnu.org/ randomly
> points to non-functional mirrors to download coreboot relative packages
> in CI builds.
>
> I would say that CircleCI fails 2/3 of the time in current builds, at
> different places.
>
> As you may know, building coreboot requires to download different
> tarballs prior of building its own musl-cross toolchain prior of
> building coreboot from it.
>
> In the case of Heads project, which builds boards ROMs for different
> coreboot versions, it means that each time that a Heads' module
> definition changes, the original coreboot musl-cross related tarballs
> need to be downloaded again prior of being in CircleCI's cache for next
> builds only related to scripts changes.
>
> Consequently, in the past (3?) days, CircleCI attempted to download
> packages from different ftpmirror.gnu.org mirrors, sometimes being
> successful, and sometimes failing after 2h of building time. I'm trying
> to build for more then 24 hours now with really low success rate from
> that situation, and i'm seeking at least acknowledgement and/or some
> advices, since pinning coreboot to selected mirrors would be a hack and
> not interesting in the long term.
>
>
> The output exerpt of CircleCI looks like this, which doesn't help much:
>
> make -C util/crossgcc build-i386 build_iasl SKIP_GDB=1
> bash ./buildgcc -p i386-elf -j 4 \
> -d /root/project/build/coreboot-4.8.1/util/crossgcc/xgcc
> Welcome to the coreboot cross toolchain builder v1.52 (June 11th, 2018)
>
> Building toolchain using 4 thread(s).
>
> Target architecture is i386-elf
>
> Found compatible Ada compiler, enabling Ada support by default.
>
> Downloading and verifing tarballs ...
> * gmp-6.1.2.tar.xz (downloading from
> https://ftpmirror.gnu.org/gmp/gmp-6.1.2.tar.xz)... 0% 2%
> 5% 7% 10% 13% 15% 18% 21% 23% 26%
> 28% 31% 34% 36% 39% 42% 44% 47% 49%
> 52% 55% 57% 60% 63% 65% 68% 71% 73%
> 76% 78% 81% 84% 86% 89% 92% 94% 97%
> 99%100%... hash verified ("9dc6981197a7d92f339192eea974f5eca48fcffe")
> * mpfr-3.1.5.tar.xz (downloading from
> https://ftpmirror.gnu.org/mpfr/mpfr-3.1.5.tar.xz)... 0%... Failed to
> download mpfr-3.1.5.tar.xz.
> make[3]: *** [Makefile:26: build_gcc] Error 1
> make[2]: *** [Makefile:48: build-i386] Error 2
> make[1]: *** [util/crossgcc/Makefile.inc:46: crossgcc-i386] Error 2
> make[1]: Leaving directory '/root/project/build/coreboot-4.8.1'
> make: *** [modules/coreboot:75:
> "/root/project/build/coreboot-4.8.1/.xcompile"] Error 2
>
> CircleCI received exit code 0
>
> And where local tests, happening throuhg QubesOS and whonix, tries
> different mirrors at each attempt which normally succeed.
>
> To I guess it comes to where CircleCI's instance is located which
> selects failing mirrors.
>
>
> If you can do anything about this, that would be helpful. Currently, i'm
> sure everyone using CircleCI to build stuff depending on your mirrors
> are having big migraines.
>
>
> Thanks!
>
> Thierry Laurion
>
> https://insurgo.ca
>
All failing builds here are link to this problem (if you want to take a
look, would just give proof, but not much more information, since
coreboot doesn't output on screen (console) the actual server being
used: https://app.circleci.com/pipelines/github/osresearch/heads
Thanks!
>
