http://bugzilla.opensuse.org/show_bug.cgi?id=1122024 http://bugzilla.opensuse.org/show_bug.cgi?id=1122024#c1 Richard Brown changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |CONFIRMED Component|Kubic |Containers Assignee|kubic-bugs@opensuse.org |containers-bugowner@suse.de --- Comment #1 from Richard Brown --- Bug confirmed, the flannel CNI config breaks podman networking. The CNI config from podman-cni-config remains incompatible with flannel's config - if it is forced on the system them both flannel's networking and podmans networking break, compared to this report where only podmans networking is broken and flannel works fine for k8s. Assigning to the containers team because a) They're responsible and most knowledgable in podman+kubernetes and at least partially responsible/knowledgable about the CNI stack b) they need all of the above working in CaaSP c) If my suspicion is correct, they should have a strong voice in deciding how we fix flannel or what we replace it with, because of the above. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1122024 http://bugzilla.opensuse.org/show_bug.cgi?id=1122024#c2 Richard Brown changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |containers-bugowner@suse.de Flags| |needinfo?(containers-bugown | |er@suse.de) --- Comment #2 from Richard Brown --- https://build.opensuse.org/request/show/666428 This change makes podman use its "podman" named CNI network, if it's present, regardless of what other CNI networks are present. (This is the modern upstream behaviour) This means that on a kubeadm Kubic system, if podman-cni-config is installed (currently requiring forced installation due to intentional conflicts in the rpms), podman still works, even if flannel has been configured However, kubernetes/flannel doesn't seem to be able to (as far as my knowledge shows) do a similar hard-binding to a single CNI network Therefore, the above approach does not solve this bug, and would just changing it symptoms if we removed the podman-cni-config rpm conflicts with kubeadm. In short, this means I need input from the containers team/anyone on the following 1) Is there any way for kubernetes/CNI to be configured to use a specific CNI network? This would mean we could tell k8s to use the "cbr0" network configured by flannel and this bug could be closed. 2) If 1) is not feasible - what alternatives to flannel or what changes to flannel can be made so we can have a CNI network configuration that will work even in the presence of another CNI network? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1122024 Richard Brown changed: What |Removed |Added ---------------------------------------------------------------------------- Component|Containers |Kubic -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=1122024 Richard Brown changed: What |Removed |Added ---------------------------------------------------------------------------- Status|CONFIRMED |RESOLVED Resolution|--- |UPSTREAM Flags|needinfo?(rbrown@suse.com) | -- You are receiving this mail because: You are on the CC list for the bug.