This message was deleted.
# k3sa
adamant-kite-43734
11/07/2024, 7:16 PMThis message was deleted.
c
creamy-pencil-82913
11/07/2024, 7:23 PMI am confused, what does the container runtime socket path have to do with the kubelet root dir?
creamy-pencil-82913
11/07/2024, 7:25 PMthe only thing that root-dir changes is where /var/lib/kubelet is at, that has nothing to do with containerd
l
late-needle-80860
11/07/2024, 7:29 PMsorry yes … of course … containerd data was “moved” via symbolic links from the default /run/k3s to our defined new folder on an LVM2.
late-needle-80860
11/07/2024, 7:29 PMnot by specifying the root.dir
c
creamy-pencil-82913
11/07/2024, 7:36 PMyou might consider using bind mounts instead of symlinks. that tends to work better for things that want to monitor the original paths.
🙏 1
🤔 1
l
late-needle-80860
11/07/2024, 7:36 PMhmm check … okay interesting.
late-needle-80860
12/03/2024, 10:26 PMHad a check and found that yes indeed the CCm does several lookups for the node on the hci ( google compute engine ). That are unsuccessful - because the name of the node does match its vm name on the hci.
However, downgrading to v4.14 of the ccm and at least it’s allowed to join as expected. Without me having to restart the k3s service multiple times quickly after each other 🤣 .
Now I need to find a way to tell the ccm the VM name pattern to look after.
And also see if upgrading to a version after v4.16 is better.
late-needle-80860
12/04/2024, 7:12 PMI finally found the smoking gun in the ccm logs.
I created this issue in the kubernetes org. On the cloud-provider-gcp repo: https://github.com/kubernetes/cloud-provider-gcp/issues/791
All of this makes me think that’s going to be hard to use the with-node-id on the cloud.
Am I correct in saying that this is a k3s specific feature @creamy-pencil-82913 ?
c
creamy-pencil-82913
12/04/2024, 7:14 PMyes it absolutely is
👍 1
creamy-pencil-82913
12/04/2024, 7:15 PMI would probably try to figure out why the node password secret isn’t getting cleaned up when the node is deleted.
🤔 1
l
late-needle-80860
12/04/2024, 7:15 PMIt’s a cool feature though! I like the thinking. I just can’t find a way to tell the ccm to use another way of deriving what name to use to identify the vm on the underlying hci
late-needle-80860
12/04/2024, 7:16 PMYeah … I have colleagues suggestion the same. Maybe I’m to stubborn here in trying to get the --with-node-id to work on GCP GCE
c
creamy-pencil-82913
12/04/2024, 7:17 PMyeah I would probably try to address the root cause, instead of trying to get your workaround working
l
late-needle-80860
12/04/2024, 7:17 PMcheck.
late-needle-80860
12/04/2024, 7:18 PMI’m grateful for your help and pointers on this. And patience 😄
late-needle-80860
12/04/2024, 7:18 PMHelped a great deal.
late-needle-80860
12/04/2024, 7:19 PMIs it K3s itself that, before a node is removed ensures that the node-password.k3s secret is removed?
c
creamy-pencil-82913
12/04/2024, 7:50 PMnot before. When. When the node resource is deleted, it should trigger a delete of the node password secret
✅ 1
9 Views