1. Forum
  2. Usenet
  3. LINUX.DEBIAN.KERNEL

  • Kernel related problem (randomly failing tests), where to discuss?

    From Diederik de Haas@21:1/5 to All on Sat Jan 15 00:00:40 2022
    Hi,

    In https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1003536 I described the problem in more detail, but I'll give the TL;DR here to determine the best place/ML to discuss the issue further.

    TL;DR: The iwd program deliberately/explicitly uses kernel features/modules
    for some of its functionality. It also has a number of tests, but they fail
    (or succeed) a bit (too) random, because whether a kernel module is loaded or not depends on several factors:
    - kernel configuration, which differs per ARCH and kernel version
    - detected (virtual) hardware which causes the (auto-)loading of loadable (=m) kernel modules
    (- 'environmental' factors like other processes that run or ran on that machine?)

    In 'extremis' this can mean that a test performed on arm64-buildd-1 fails, while it succeeds on arm64-buildd-2, because of a (slight) HW difference.

    On my own hardware I have complete freedom to fix that by modprobe-ing needed kernel modules. But (AFAIK) I have less freedom on Salsa CI, even less on autopkgtest/ci.debian.net and (essentially?) none on buildd, assuming it's
    even (technically) possible.

    OTOH, you want tests to consistently fail or succeed and not be dependent on chance. (Reproducible Builds may be relevant here too)

    Has this problem been discussed before? If so, could someone point me to that? If not, where would the best place be to discuss this?

    Cheers,
    Diederik
    -----BEGIN PGP SIGNATURE-----

    iHUEABYIAB0WIQT1sUPBYsyGmi4usy/XblvOeH7bbgUCYeIAmAAKCRDXblvOeH7b biX2AQDXJo9fqobnd2497D6LrSnOBX7wCrmR/HFPWn+virrE0wEAyfJUN4rTUzHX GnEuFQMUq+zoqVE3bpMGTsCaQrbqFwc=
    =o85C
    -----END PGP SIGNATURE-----

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Ben Hutchings@21:1/5 to Diederik de Haas on Sun Jan 23 16:50:03 2022
    On Sat, 2022-01-15 at 00:00 +0100, Diederik de Haas wrote:
    Hi,

    In https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1003536 I described the problem in more detail, but I'll give the TL;DR here to determine the best place/ML to discuss the issue further.

    TL;DR: The iwd program deliberately/explicitly uses kernel features/modules for some of its functionality. It also has a number of tests, but they fail (or succeed) a bit (too) random, because whether a kernel module is loaded or
    not depends on several factors:
    [...]
    Has this problem been discussed before? If so, could someone point me to that?
    If not, where would the best place be to discuss this?

    You shouldn't run any tests like this at build time.

    For autopkgtests, if you set the needs-root and isolation-machine
    restrictions then the tests will run as root on a VM. But currently
    neither salsa-ci nor ci.debian.net implements this, so those tests will
    be skipped.

    Another option in autopkgtests is to depend on qemu and start the VM
    yourself. This is not easy to do, but I implemented it for initramfs-
    tools.

    Ben.

    --
    Ben Hutchings
    I'm always amazed by the number of people who take up solipsism because
    they heard someone else explain it. - E*Borg on alt.fan.pratchett

    -----BEGIN PGP SIGNATURE-----

    iQIzBAABCgAdFiEErCspvTSmr92z9o8157/I7JWGEQkFAmHtdtsACgkQ57/I7JWG EQmYrg/+ISN9Gd5+1Rxe+/e1VOxeq/JXoqn2X4ZPCxEapcbi4RnyFmzVLP2UyEAf atLQoQrNiUnDVsHcKL4BlPXOjBI/c/88UDnAxnlz4spwazFoYYHrAcvexiGuxlbF T5uU0Tj1sxc3VC2NMRM/k/KgqS6yEFd6hOxhVyYnyQ6s9QOXjJqFBff3wv9tgij2 ytlB+4+pY2Fh+daNuIxYvOORanQdqXe5tUeY6BPF12pODE6YNc6I9+2nL5rSWJsp 6jiNz0Hk8Ei/Dp+L5k8X/Nx/FpgXdo5E7Vovw9tegtjEMbt6aUJv3fXICXNiOGgQ LS07ARC2p1Rot2y1lOuAAwGRJJE1AxlE6rEDeF/+/HIqT6nPiCiwOqTNJ5KFtLFM IBnU3lqvxhMP/ZZCxEEycHjJx9GfG/15xlVcSekiUoehcb8+35pmT6Ra0DamOOnq rjXHmj0c4sc1smzsEwBYYzgPh3qLigE/xE8RJb3201IaeBDacpJlg+QmeADUv8uj aLqm3+ic8NIq6bzG+xiRxfv+HahDBIgnWNLMuJR7JCsrerGMYujIMHj0JBjDcSK4 qNuoeWOqLgQKJAd3NYisfyCls6HfoOITigIaUvu3nbKE8YP4nGrgeM1f/53W8g2l vyIiQGD6+2bqZPkanyxMXuIG6MhiN7xOQH5+6V4d7rvvtJaYLc4=
    =gQN1
    -----END PGP SIGNATURE-----

    --- SoupGat
  • From Diederik de Haas@21:1/5 to All on Mon Jan 24 16:38:26 2022
    On zondag 23 januari 2022 16:40:11 CET Ben Hutchings wrote:
    You shouldn't run any tests like this at build time.

    For autopkgtests, if you set the needs-root and isolation-machine restrictions then the tests will run as root on a VM. But currently
    neither salsa-ci nor ci.debian.net implements this, so those tests will
    be skipped.

    Another option in autopkgtests is to depend on qemu and start the VM yourself. This is not easy to do, but I implemented it for initramfs-
    tools.

    Thank you :)

    Cheers,
    Diederik
    -----BEGIN PGP SIGNATURE-----

    iHUEABYIAB0WIQT1sUPBYsyGmi4usy/XblvOeH7bbgUCYe7H8gAKCRDXblvOeH7b bu1OAQDXp9U+M3RLcMSNjHgd7LD1UISEEzPpnDpmKfua4aJE6wD9FARO7cjLxxbj iUoIQRSg/79E46phpX8ieu90YqRGzAc=
    =y1Pt
    -----END PGP SIGNATURE-----

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)