cri-o: Ubuntu 18.04: DNS resolution broken after scheduling hostport enabled pods
Description
Cri-O breaks DNS resolution once pod with “Hostport” is scheduled. I’m logging this issue here, as i was unable to reproduce this issue on Docker 17.03 and Containerd 1.1.3/1.2.0-beta.
The symptoms are similar to https://github.com/kubernetes/kubeadm/issues/273 .
Steps to reproduce the issue:
- Schedule a pod with
hostPortcontainer spec configured. - Kubernetes shows the pod in ImagePullBack error.
- DNS resolution is not working anymore. However kubelet and crio services later disabled.
Additional information you deem important (e.g. issue happens only occasionally):
Output of crio --version:
cri-o-1.11 Xenial Package
Additional environment details (AWS, VirtualBox, physical, etc.):
- Hetzner Cloud and Scaleway
- OS: Ubuntu 18.04 Bionic
- Kubernetes 1.11.2
- CNI 0.7.1
- CNI runtime: Issue observed using Canal, Cillium and Weave
About this issue
- Original URL
- State: closed
- Created 6 years ago
- Reactions: 2
- Comments: 15 (7 by maintainers)
@runcom, @Bregor The usual way to tacle the issue with
containerdanddockeris to enforce the path to the proper resolv.conf in thekubelet.This can be acheived using via the command line argument
--resolv-conf=/run/systemd/resolve/resolv.confor via Kubelet configuration file.But it does not work with cri-o.
@runcom I did not try with
containerd, butdocker-18.06“broken” just like cri-o