Yesterday's containers-common update [1] to 0.58.0-1.fc41 changed the dependencies [2], in particular it dropped slirp4netns (Suggests: doesn't count, as it doesn't get installed by default). This broke user containers, see below. This is a freshly booted Rawhide VM, i.e. no old config files. So either the dependency was dropped prematurely, or it was deliberate and podman shouldn't use slirp any more (which is my understanding, as it wants to use pasta now?) -- so please feel free to reassign to podman. [1] https://koji.fedoraproject.org/koji/buildinfo?buildID=2420115 [2] https://src.fedoraproject.org/rpms/containers-common/c/17934d87b2686ab5c18cb56277967c926ded61f3?branch=rawhide Reproducible: Always Steps to Reproduce: $ sudo dnf install -y podman $ podman run -it --rm quay.io/libpod/busybox Error: could not find slirp4netns, the network namespace can't be configured: exec: "slirp4netns": executable file not found in $PATH Seen by cockpit-podman tests, e.g. https://artifacts.dev.testing-farm.io/71f500ce-c7a0-4b7e-b9ef-47ded095378e/ https://artifacts.dev.testing-farm.io/8ed028ba-50e2-4970-90c2-046a560910b3/ https://artifacts.dev.testing-farm.io/86c12d23-18f5-4392-aad6-d49d9180c1ff/
@pholzing I pulled the containers.conf file from c/common v0.58.0 in this containers-common update and set pasta as a hard dep and slirp4netns to a weaker dep Suggests. Looks like that wasn't enough and podman was still searching for slirp4netns on a fresh rawhide VM. Is there any other configuration required?
Looks like f40/41 is still on podman 5.0.0-rc1??? The pasta default change was merged for 5.0.0-rc4.
(In reply to Paul Holzinger from comment #2) > Looks like f40/41 is still on podman 5.0.0-rc1??? The pasta default change > was merged for 5.0.0-rc4. Thanks Paul. @mpitt would you be able to rerun the tests with the latest rawhide rpm ?
This morning's `dnf update` run in my rawhide VM still doesn't see anything newer than podman-5.0.0~rc1-3.fc40.x86_64. Maybe it's because the rawhide updates happen in such a rapid succession [1] that hardly any update actually makes it into a compose and the mirror delay? This is certainly a dependency bug in containers-common, it should Conflicts: to podman < (version_with_pasta). Downloading it directly from koji: dnf update -y https://kojipkgs.fedoraproject.org//packages/podman/5.0.0~rc6/2.eln136/x86_64/podman-5.0.0~rc6-2.eln136.x86_64.rpm But now `podman run -it --rm quay.io/libpod/busybox` hangs completely, doesn't give me a shell, can't be ^Ced, and there's neither journal nor any other messages. systemd-cgls shows └─podman-1353.scope ├─1353 podman run -it --rm quay.io/libpod/busybox └─1360 /usr/bin/pasta --config-net -t none -u none -T none -U none --no-map-gw --dns none --quiet --netns /run/user/1000/netns/netns-6fc583c9-1160-0549-12c5-2d5fe1b6951f but that's about it, nothing happens. This still remains after a reboot. This is another bug which has started to break all upstream podman rawhide tests, I'll file that separately. But it means that I can neither confirm nor reject this slirp4netns issue with newer versions. [1] https://bodhi.fedoraproject.org/updates/?packages=podman
I reported the new hang here: https://github.com/containers/podman/issues/22052
@pholzing The "could not find slirp4netns" seems to still occur on rawhide gating tests with 5.0.0-1 with: containers-common-0.58.0-4.fc41.noarch and passt-0^20240320.g71dd405-1.fc41.x86_64 as of Mar 25 https://artifacts.dev.testing-farm.io/a4045627-d621-474c-acc5-a221b3297782/work-tests.ymlnj5vrrlr/tests-csu495bm/test.podman-rootless.bats.log . The link mentions all the installed packages on top and all are the latest available AFAIK. And given the pasta default was set in 5.0.0-rc4, this is strange.
@mheon @santiago CC
Today is March 26. I cannot reproduce the slirp-error nor the hang, either with or without --enablerepo=updates-testing . The gating-test failures are in tests that explicitly use --network=slirp4netns. Those need to be fixed, but their failure has nothing to do with this BZ.
@(In reply to Ed Santiago from comment #8) > Today is March 26. I cannot reproduce the slirp-error nor the hang, either > with or without --enablerepo=updates-testing . > > The gating-test failures are in tests that explicitly use > --network=slirp4netns. Those need to be fixed, but their failure has nothing > to do with this BZ. Thanks Ed. I'm gonna waive the test failures then. The tests are timing out on toolbx and Rishi is investigating it, but I'd rather we get these bodhi in asap.
Can this be closed now? The default is pasta (and passt should be installed by default), only users that use --network slirp4netns have to install slirp4netns, that is the case for our gating tests as we want to test both the pasta and the slirp4netns podman integration.
Yeah, I suppose. Dependencies in Fedora are a lost cause anyway, so there's nothing else to do here. Thanks!