Hello,
I've been trying to enable IPVS in kube-proxy on an EKS cluster, using the managed kube-proxy addon. Here are the cluster details:
Kubernetes: v1.27 - kube-proxy addon: v1.27.4-eksbuild.2 - Nodes with AMI: v1.27.3-20230816
This is the advanced configuration provided on the kube-proxy addon:
{
"mode": "ipvs",
"ipvs": {
"scheduler": "rr"
}
}
However, it appears that it is not working at all since one of our backend pods, exposed as ClusterIP is never reached (all requests are logged, and it is at zero even though some other replicas reach 65k when using round robin as a scheduler). I noticed that the field "udpIdleTimeout" caused a validation warning in the configuration map, but removing it yielded the same results. After all of this, I tried to enable all the IPVS modules by using custom user data in a launch template, as indicated in this now older issue:
https://github.com/aws/containers-roadmap/issues/142#issuecomment-1170010088 . Even indicating the scheduler to use in the daemon set did not help.
Am I missing any important step to ensure IPVS set to round robin correctly balances traffics between all pods (Ideally I would use "sed", but "rr" is ideal to see if IPVS is working as expected). It was all working correctly until a month ago, using ipvsadm installed via userdata as mentioned on the github issue.
AWS OFFICIALUpdated 3 months ago
