Control: tags -1 + pending
On Sun, 27 Oct 2024 at 15:56:11 +0100, Aurelien Jarno wrote:
> I have tracked down the issue to the use of systemd >= 256~rc3-3 on the
> host, which bumps the maximum number of open files hard limit from
> 1048576 to 1073741816 [1].
Good catch! I have been looking into this but had not diagnosed it
successfully yet.
> Restoring the bookworm value using "ulimit -Hn 1048576" before the build
> is enough to fix the issue.
I believe this is a timeout caused by test instrumentation iterating
over all possibly-open fds before we spawn a subprocess, so that we can
assert that they are correctly close-on-exec (which is part of why the
test for the instrumented debug build is failing, but the test for the
production build that we actually use in the OS is passing).
Because fds are allocated lowest-first, it would be enough to stop at
some reasonable limit like 1024 - in practice the test suite isn't going
to get anywhere near that high anyway.
smcv
