08/08/2022, 12:17 PM
Hello people. I'm trying the offline installation of k3s and I would also use cilium as network provider (and metallb, but that's another story). The k3s images are correctly imported, putting the airgap tar file into
, but when I do the same thing with cilium operator and cilium "client" images tar files they seems to be not imported automatically during the cluster init. Do the auto-import only works for k3s airgap package? Or am I making something wrong? I put here also the list of images directory and the crictl image list after the k3s cluster init:
Copy code
[root@rocky1 srv]# ls /var/lib/rancher/k3s/agent/images/
cilium-operator.tar  cilium.tar  k3s-airgap-images-amd64.tar.gz  metallb-controller.tar  metallb-speaker.tar
[root@rocky1 srv]# crictl image list
IMAGE                                        TAG                    IMAGE ID            SIZE
<|>               v0.7.3-build20220613   38b3b9ad736af       239MB
<|>                 v0.3.5                 dbd43b6716a08       8.51MB
<|>     v0.0.21                fb9b574e03c34       35.3MB
<|>   1.9.1                  99376d8f35e0a       49.7MB
<|>   1.34.1                 62aedd01bd852       1.47MB
<|>   2.6.2                  72463d8000a35       103MB
<|>    v0.5.2                 f73640fb50619       65.7MB
<|>             3.6                    6270bb605e12e       686kB
Thanks a lot for your help!


08/09/2022, 3:25 PM
Do you have a manifest configured in
as well that specifies to use cilium, and have the flags in config to not use the built in cni? If not, you’ll need those as well to make sure k3s knows to use cilium (and its images you’ve loaded there) instead of flannel


08/10/2022, 7:20 AM
I do have the flag for the network (indeed when k3s starts it starts without the network pods) but I miss the manifests files. So this could be the reason. Thanks a lot for your help Max!
👍 1
one last question: I'm having problems using the manifest to deploy cilium because it find that the node is not ready (and oc it's not ready because the network is missing 😄 :
Copy code
0/1 nodes are available: 1 node(s) had untolerated taint {<|>: }. preemption: 0/1 nodes are available: 1 Preemption is not helpful for scheduling.
is there a way to avoid this? I'm thinking about the manual removal of the not-ready taint, but I guess it's not really safe to do
the other option could be add a toleration in the helmchart values, but again I'm not sure to do the correct thing