21 Nov 2026

feedFedora People

Kevin Fenzi: misc fedora bits: third week of november 2026

Kevin Fenzi's avatar Scrye into the crystal ball

The last week was a very busy one, while next week will be hopefully very quiet.

Mass updates/reboots

We managed to get everything updated and rebooted last week, but the outage on thursday didn't go at all nicely. Here's a few of the big things we hit:

I fumbled a ad-hock ansible command horribly. I was meaning to upgrade a few builders from fedora 44 to fedora 45. ansible lets you run ad-hock commands by passing the command and a host pattern to run that command on. You can use wildcards in those patterns as well as groups or hostnames or anything ansible can understand. You need to seperate patterns with a sperator, : and escape wildcards from the shell. I meant to use:

buildvm-a64-0*:!buildvm-a64-01*:!staging

That is, all the buildvm-a64-0* hosts, except not bvmhost-a64-01 or any hosts in the staging group. Instead what I typed:

buildvm-a64-0*:!buildvm-a64-01:*:!staging

That is, all buildvm-a64-0* hosts, not a host called "buildvm-a64-01" which does not exist (we use fqdn everywhere), all hosts, and not staging. Oops. I saw it right after I ran it, and hit control-c, but it already started the dnf command on some hosts. ;( Luckily our rhel hosts all failed because we have some repo files expanding versions and 'epel45' and the like doesn't exist there. Unfortunately it did affect some fedora hosts I didn't plan on. ;(

Due to the above, the wiki, openqa and some copr hosts all got upgraded. The wiki broke, but we were able to back out the upgrade there and get things working again, openqa just needed a playbook run and the copr hosts are planned to upgrade next week anyhow, so things were not too bad.

For some reason some of the builders got into a bad state. I think it may be because it started the upgrade and then something caused it to stop in the middle. This resulted in needing to reinstall a bunch of the buildvm's. Basically all the aarch64, ppc64le, and s390x builders were reinstalled. Luckily thats just a playbook away.

The f45 change moving repo files bit us. In the past we just copy over via ansible new repo files that point to our internal mirror. However, until those are moved over newly install buildvm's try and use the default metalink and were getting mirrors that are completely not reachable by them. Additionally, we disable the openh264 repo by copying over a repo file with just a note in it. Until ansible copies that over however, builders see the enabled openh264 and try and hit the external cisco cdn and of course cannot reach it. Finally our local repo file copies all point to gpg keys that are no longer in the same place. All this is fixable, and I'm kicking myself for not doing fresh installs in staging sooner to catch this sort of thing.

Next, our aarch64 reinstalls were not working right. About 50-75% of the time they would boot, but then not be able to download the initramfs from the network. I still need to try and figure out whats going on here. In the mean time I just brute forced installing them all until they booted correctly and installed/came up.

Finally, the new fedora 45 builders were unable to build epel packages. This was emitting a signature error in buildSRPM tasks. This issue was caused by the Fedora 45 rpm enforcing signatures change. We had tested Fedora and dnf5 handles this just fine, unfortunately, epel builds use dnf-3 and it doesn't handle this. For now at least I switched rpm on all the builders back to the old digests verify only. I'm not sure there's going to be much else to do, although I suppose we could try and move epel to use dnf5 for bootstrapping / srpm builds. It could well change some behavior, but would be a lot more future proof.

Anubis challenges

The anubis valkey backend seems to be working fine from all I can tell. I haven't heard anyone say that they are getting less challenges, but it sure seems like that here.

Thanksgiving next week

Next week is the thanksgiving holidays in the us and I am taking all week off. Of course I will still be around in the Fedora community, but don't expect me to be at meetings or do things that aren't either urgent or not things I enjoy.

I've got a lot of personal projects piled up and I hope to work on some of those, but also just decompress and relax some.

As always, comment on the fediverse: https://fosstodon.org/@nirik/117656544865011334

21 Nov 2026 8:43pm GMT

Guillaume Kulakowski: Migration de Gmail vers plMail : reprendre le contrĂ´le sur mes e-mails

21 Nov 2026 1:47pm GMT

20 Nov 2026

feedFedora People

Filipe Rosset: dkms vs r8168-8.057.00 issues

Filipe Rosset's avatar

r8168 Makefile does not honor the KVER variable, there some issues there:

BASEDIR := /lib/modules/$(shell uname -r)
KERNELDIR ?= $(BASEDIR)/build
KVER = $(shell uname -r)

as KERNELDIR is using BASEDIR, and BASEDIR does use "uname -r" hardcoded, our workaround here is to pass a dynamic value for KERNELDIR, the precedence for our value do the trick.

rosset@xeon:# diff -ruN /usr/src/r8168-8.057.00-1/dkms.conf.bak /usr/src/r8168-8.057.00-1/dkms.conf
--- /usr/src/r8168-8.057.00-1/dkms.conf.bak 2026-09-10 21:00:00.000000000 -0300
+++ /usr/src/r8168-8.057.00-1/dkms.conf 2026-11-20 20:36:00.398148359 -0300
@@ -1,6 +1,6 @@
PACKAGE_NAME=r8168
PACKAGE_VERSION=8.057.00-1
-MAKE[0]="'make' -j$(nproc) KVER=${kernelver}"
+MAKE[0]="'make' -j$(nproc) KVER=${kernelver} KERNELDIR=/lib/modules/${kernelver}/build"
BUILT_MODULE_NAME[0]=r8168
DEST_MODULE_LOCATION[0]=/kernel/drivers/net
AUTOINSTALL="YES"
rosset@xeon:#

rosset@xeon:~# sudo dkms status

for K in /lib/modules/; do
K=${K##
/}
echo "===== $K ====="
modinfo -k "$K" r8168 2>/dev/null |
grep -E '^(filename|rhelversion|vermagic):'
done


gpu-unlock/2.0.0, 6.12.0-208.112.5.3.el10uek.x86_64, x86_64: installed
gpu-unlock/2.0.0, 6.12.0-211.74.1.el10_2.x86_64, x86_64: installed
gpu-unlock/2.0.0, 6.12.0-264.14.1.el10_3.x86_64, x86_64: installed
nvidia/620.70.06, 6.12.0-208.112.5.3.el10uek.x86_64, x86_64: installed
nvidia/620.70.06, 6.12.0-211.74.1.el10_2.x86_64, x86_64: installed
nvidia/620.70.06, 6.12.0-264.14.1.el10_3.x86_64, x86_64: installed
r8168/8.057.00-1, 6.12.0-208.112.5.3.el10uek.x86_64, x86_64: installed
r8168/8.057.00-1, 6.12.0-211.74.1.el10_2.x86_64, x86_64: installed
r8168/8.057.00-1, 6.12.0-264.14.1.el10_3.x86_64, x86_64: installed
===== 6.12.0-207.111.5.3.el10uek.x86_64 =====
===== 6.12.0-207.111.5.4.el10uek.x86_64 =====
===== 6.12.0-207.111.5.5.el10uek.x86_64 =====
===== 6.12.0-207.111.5.6.el10uek.x86_64 =====
===== 6.12.0-208.112.5.1.el10uek.x86_64 =====
===== 6.12.0-208.112.5.2.el10uek.x86_64 =====
===== 6.12.0-208.112.5.3.el10uek.x86_64 =====
filename: /lib/modules/6.12.0-208.112.5.3.el10uek.x86_64/extra/r8168.ko.xz
vermagic: 6.12.0-208.112.5.3.el10uek.x86_64 SMP preempt mod_unload modversions
===== 6.12.0-211.62.1.el10_2.x86_64 =====
===== 6.12.0-211.65.1.el10_2.x86_64 =====
===== 6.12.0-211.66.1.el10_2.x86_64 =====
===== 6.12.0-211.68.1.el10_2.x86_64 =====
===== 6.12.0-211.69.1.el10_2.x86_64 =====
===== 6.12.0-211.72.1.el10_2.x86_64 =====
===== 6.12.0-211.73.1.el10_2.x86_64 =====
===== 6.12.0-211.74.1.el10_2.x86_64 =====
filename: /lib/modules/6.12.0-211.74.1.el10_2.x86_64/extra/r8168.ko.xz
rhelversion: 10.2
vermagic: 6.12.0-211.73.1.el10_2.x86_64 SMP preempt mod_unload modversions
===== 6.12.0-264.14.1.el10_3.x86_64 =====
filename: /lib/modules/6.12.0-264.14.1.el10_3.x86_64/extra/r8168.ko.xz
rhelversion: 10.3
vermagic: 6.12.0-264.14.1.el10_3.x86_64 SMP preempt mod_unload modversions

20 Nov 2026 11:58pm GMT