Closed kgts23 closed 5 months ago
I've changed the CNI from Canal to Calico. Now the metric-server is reachable from all Nodes.
root@k8s-node01:~# nc -uzv 192.168.1.21 4789
Connection to 192.168.1.21 4789 port [udp/*] succeeded!
root@k8s-node01:~# nc -uzv 192.168.1.22 4789
Connection to 192.168.1.22 4789 port [udp/*] succeeded!
root@k8s-node01:~# nc -uzv 192.168.1.23 4789
Connection to 192.168.1.23 4789 port [udp/*] succeeded!
Environmental Info: RKE2 Version: rke2 version v1.29.3+rke2r1 (1c82f7ed292c4ac172692bb82b13d20733909804) go version go1.21.8 X:boringcrypto
Node(s) CPU architecture, OS, and Version: x86_64, Ubuntu, 22.04.4 LTS
Cluster Configuration: 3 Server Nodes (for Rancher later) No CIS profile
Describe the bug: When calling up kubectl top nodes, I noticed that only every third request was answered via the load balancer (layer 4, round robin). I then ran kubectl top node on every single node and realized that only the node where the metrics server was also running as a pod was responding. I then set up a busybox and tried to reach the internal cluster IP of the metrics-server pod. Only the node itself, where metrics-server was deployed, responded.
Steps To Reproduce:
Actual behavior: Only reachable on the node, where the metrics-server pod is running
Additional context / logs: From node2 (trying to reach metrics-server on node3)
Busybox from node02
Canal Log on node2 for the busybox: