I tried reproducing this line locally and I do hit some 504 errors, but dnf seems to handle it locally - this is only reproducible in our CI environment at the moment.
[MIRROR] gpgme-1.20.0-5.fc39.x86_64.rpm: Status code: 404 for https://d2lzkl7pfhq30w.cloudfront.net/pub/fedora/linux/releases/39/Everything/x86_64/os/Packages/g/gpgme-1.20.0-5.fc39.x86_64.rpm (IP: 3.167.217.183)
[MIRROR] libmetalink-0.1.3-32.fc39.x86_64.rpm: Status code: 404 for https://d2lzkl7pfhq30w.cloudfront.net/pub/fedora/linux/releases/39/Everything/x86_64/os/Packages/l/libmetalink-0.1.3-32.fc39.x86_64.rpm (IP: 3.167.217.183)
Fedora 39 has been EOL for almost a year and gets no updates of any kind
Fedora 40 is also EOL and receives no updates.
If you wish to remain on f39 then updates are non-existent
If you wish to update then please perform the upgrades using dnf at the command line with sudo dnf system-upgrade download --releasever=XX to download the upgrades followed by sudo dnf system-upgrade reboot (Replace the XX with the release version to use for the upgrade and remember that fedora only supports upgrades of 1 or 2 release versions at a time)
Once you have upgraded to f41 then the system will be running dnf5 and the second command would change to sudo dnf offline reboot
Currently your CI tests only against Fedora 38 and 39, which are out of support. It doesn’t have any coverage for Fedora 41 and 42, which are the currently supported Fedora versions.
Probably it’s best to modify the CI test cases to cover F41 and F42 instead of the obsolete versions?
These are container images (e.g. ghcr.io/nvidia/fedora:39@sha256:blah) so rather than trying to upgrade, they can just pull an F41 or F42 container image instead.
What about using RHEL or a derivative? RHEL is based on Fedora, but rather than about 13 months of support as you get with Fedora, RHEL and derivatives will provide up to ten years of support.
I mean, it’s good that they test their product against Fedora! (We’re looking at a Fedora container instance used for integration testing of code changes here, not at a production system.)
So this is actually probably a bug in how the viable mirrorlist is constructed for F39.
Yes F39 is EOL’d and is not receiving updates.. and users are not encouraged to use it. But users are gonna user for user reasons.
The intent is that metalink should be returning valid mirrors for all the releases… for some reason the mirrorlist for F39 is being populated incorrectly.
I also have an F39 laptop… weirdly I’ve had to boot it like 3 times this week to differential analysis something that’s come up in here…hmm.. anyways…
Im able to pull packages from the F39 repos without a problem because for whatever reason the mirrorlist selection algorithm isn’t choosing the problematic cloudfront mirror…but its in the list.. i checked the metalink XML file pulled.
Not sure how or if this will get sorted out, in the meantime you could stop using metalink and select a specific other mirror that you can confirm is valid.
Hello, I just wanted to update that I have removed[0] the EOL fedora:{38,39} from our pipelines, and am now using fedora:41. No issues now. I appreciate the fast responses from everyone.