Marcel Fest
Marcel Fest
I don't think so, but I think that configuring kube-proxy using `strictARP=true, mode=ipvs` and using kube-vip was never tested.
@jennydaman we had a pull-request which changed it from physical to macvlan when DHCP is used see #897 and the issue #871. Macvlan itself should not cause packet loss, maybe...
@jennydaman what you can try is running a trace route using `mtr` to your TCP Port of the service and check with `pwru` if all packets hit the node. Maybe...
@jennydaman that is a good question. I don't know if configuring rp_filter with `0` is correct or maybe even wanted. I think this topic needs a short investigation. @chdxD1 maybe...
@jennydaman as the Macvlan interface is managed by kube-vip we can add this without any side effects to the other network configurations on the node.
@jennydaman it would be awesome if you couldtry the new functionality in version v.1.0.0 :)
@jennydaman good point, the manifest generation needs some updates in general.
Hey @CafeLungo, could you add the `kube-vip-ds` logs as well? They have more information and hopefully better logs when you set the loglevel to `-4`.
@CafeLungo the `EndpointSlice` topic is already on my radar as the tests failed after an api and client upgrade to k8s version 1.33. For that to disappear you could enable...
Yes the pod needs to live on the same host for ingress and egress to work.