kubeadm not creating /etc/cni/net.d .. pods will fail to load.

[adiri]

I tested kubeadm on CentOS 7, did exactly what the document stated and everything work as expected.

when I created a pod I had the following error:

  1m        1m      4   {kubelet server.name}           Warning     FailedSync  Error syncing pod, skipping: failed to "SetupNetwork" for "appname-a94fb6130669dde66c7ebe8cd697c498463fa721-5rg6i_default" with SetupNetworkError: "Failed to setup network for pod \"appname-a94fb6130669dde66c7ebe8cd697c498463fa721-5rg6i_default(45543fa1-b1fe-11e6-af4d-5254002465aa)\" using network plugins \"cni\": cni config unintialized; Skipping pod"

creating at the master and slaves /etc/cni/net.d fixed the problem.

[fieg]

I'm experiencing the same on ubuntu xenial.

@adiri what do you mean with "creating /etc/cni/net.d"? Just create an empty directory?

[mikedanese]

kubeadm doesn't create this, a network daemonset should. Are your running weave or flannel in a daemonset?

[adiri]

@mikedanese Yes, I applied the weave daemonset (kubectl apply -f https://git.io/weave-kube)

I also tried kubeadm init --pod-network-cidr=10.244.0.0/16 but same problem.

[mikedanese]

Then this might be a problem with weave-kube cc @errordeveloper @lukemarsden

[matthewrobertbell]

I'm also experiencing this with a fresh Digital Ocean 4GB 16.04 VM (single node cluster). I followed http://kubernetes.io/docs/getting-started-guides/kubeadm/ and didn't install a network daemonset, as it wasn't mentioned as required.

kube-dns and kube-dashboard are the pods that fail to come up.

I hope that helps.

Edit: I see a network daemonset is required, my mistake.

Second edit: Starting from scratch, after kubeadm init, I installed flannel with kubectl create -f https://raw.githubusercontent.com/coreos/flannel/master/Documentation/kube-flannel.yml

After this, kube-dns has still failed, presumably because it didn't retry. Is it a bug to try and setup kube-dns (which relies on a network daemonset), before there is a network daemonset?

[damaspi]

On an Ubuntu 16.04.1 LTS, with a master and 3 nodes in an internal static network (192.168.2.99 to 102) but having each also an IP address directly on the Internet, I have installed k8 (1.5.1), then initiated with
kubeadm init --api-advertise-addresses=192.168.2.99
to link the cluster on the internal network and not expose it.

I then installed my weave-kube by typing
kubectl apply -f https://git.io/weave-kube

kubectl get pods --namespace=kube-system shows that everything seems to be ok, but when I add successfully the nodes, I see regular crashes of the weave-net-xxxx pods.

When I deploy some demo application, I have the same message as above. (Error syncing pod, skipping: failed to "SetupNetwork" ).

When I check the logs of the proxy pod, kubectl logs kube-proxy-g7qh1 --namespace=kube-system I get the following info: proxier.go:254] clusterCIDR not specified, unable to distinguish between internal and external traffic

The test for an empty clusterCIDR in pkg/proxy/iptables/proxier.go seems rather recent, so is kubeadm adapted to ensure that this flag is properly set? In the templates (.json), I don't find it and I don't know where to add this flag to ensure that it is properly taken by the proxy...

[pires]

@damaspi i've opened #102 to address your issue as it's not related to this one.

[luxas]

kubeadm should not create the /etc/cni/net.d directory; that's a task for the CNI DaemonSet plugins.

And the other issue that's faced here is handled in #102

[funnydevnull]

I am having similar problems but _only_ when I use an internal IP address for kubeadm init --api-advertise (I have two interfaces, one internal one external). Is this related to #102?

[luxas]

@funnydevnull I can't say, please open a new issue with more details and we'll see