After some discussions with Christian Heimes, it seems this has something to do with the latest kernel (4.9), as the kernel's api used in this test changed [0][1]. This will need to be addressed in python, but it is weird that it happens only on specific architectures.
[0] https://github.com/smuellerDD/libkcapi/commit/be242c387b7030cbccae2c183107efa86d9a3cd6 [1] http://bugs.python.org/issue29324 Anyway I don't think any more attention is required by infra for this specific case. Regards, Charalampos Stratakis Associate Software Engineer Python Maintenance Team, Red Hat ----- Original Message ----- From: "Kevin Fenzi" <ke...@scrye.com> To: devel@lists.fedoraproject.org Sent: Wednesday, January 18, 2017 7:55:38 PM Subject: Re: Is there something wrong with the Koji builders? On Wed, 18 Jan 2017 07:22:49 -0500 (EST) Charalampos Stratakis <cstra...@redhat.com> wrote: > Python 3 started failing on i686, x86_64 and arm only. > > https://koji.fedoraproject.org/koji/taskinfo?taskID=17313924 > > The failures come from the test_socket.py [0], test_aead_aes_gcm [1] > and more specifically this line [2] with the message: OSError: [Errno > 22] Invalid argument > > This test is checking Python's interface for the kernel crypto API > (added in 3.6 [3]). > > [0] > https://hg.python.org/cpython/file/3.6/Lib/test/test_socket.py#l5473 > [1] > https://hg.python.org/cpython/file/3.6/Lib/test/test_socket.py#l5472 > [2] > https://hg.python.org/cpython/file/3.6/Lib/test/test_socket.py#l5497 > [3] http://bugs.python.org/issue27744 Thats nothing to do with the buildsystem. Or at least if it is, it's not any of the known issues I was working on. As far as I know now everything should be back to working/normal. There were 3 issues: 1. The buildvm's were updated and started crashing under load. They would spew oopses and finally reboot. This would result in builds on them getting restarted or dying in strange ways. This seems to be something particular to their hardware/setup that changed in later 4.8.x and early 4.9.x kernels. We are currently running 4.10rc4 on them and it's been stable (knock on wood). 2. Sometimes, very rarely dnf would fail downloading packages for the build root. We have worked around this by passing dnf via koji several mirrors where it can download from, so if it fails on one it should fall back to the next and so on. https://pagure.io/fedora-infrastructure/issue/5689 3. In some configurations (where there was more than I host to download from) koji would fail to download the src.rpm correctly and error out. We are working around this now by just pointing koji at one place for these downloads for now until we can get things fixed. https://pagure.io/fedora-infrastructure/issue/5694 Sorry for all the instability the last few days. If you see any further issues like the above, let me know. kevin _______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org _______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
