Bug 1685118 - VMIs with delegated IPs are stopped by the kubelet if kubelet is restarted
Summary: VMIs with delegated IPs are stopped by the kubelet if kubelet is restarted
Keywords:
Status: CLOSED WONTFIX
Alias: None
Product: Container Native Virtualization (CNV)
Classification: Red Hat
Component: Networking
Version: 1.4
Hardware: Unspecified
OS: Unspecified
medium
high
Target Milestone: ---
: 2.1.0
Assignee: Petr Horáček
QA Contact: Meni Yakove
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2019-03-04 11:39 UTC by Roman Mohr
Modified: 2019-06-20 11:57 UTC (History)
4 users (show)

Fixed In Version:
Doc Type: If docs needed, set a value
Doc Text:
Clone Of:
Environment:
Last Closed: 2019-06-20 11:57:25 UTC
Target Upstream Version:
Embargoed:


Attachments (Terms of Use)


Links
System ID Private Priority Status Summary Last Updated
Github kubevirt kubevirt pull 2100 0 None closed [WIP] keep original pod ip on a dummy interface 2020-10-05 01:45:18 UTC

Description Roman Mohr 2019-03-04 11:39:40 UTC
Description of problem:

When connecting the VMI via `bridge` mode and IP delegation to the pod network, CNI status checks fail and as a consequence the kubelet stops all containers of that pod when the kubelet is restarted.

I checked it on k8s 1.10.4 and k8s 1.13.3. I *think* that the hint in the logs is this:

```
Mar 04 09:28:43 node01 kubelet[29530]: W0304 09:28:43.167534   29530 docker_sandbox.go:384] failed to read pod IP from plugin/docker: NetworkPlugin cni failed on the status hook for pod "virt-launcher-vmi-nocloud-v4j2m_default": Unexpected address output
```

I did not try it with ovn-cni, crio, just flannel.


Version-Release number of selected component (if applicable):


How reproducible:

```bash
$ cluster/kubectl.sh create -f cluster/examples/vmi-nocloud.yaml
$ cluster/kubectl.sh create -f cluster/examples/vmi-slirp.yaml
$ cluster/kubectl.sh get vmis
NAME          AGE   PHASE     IP            NODENAME
vmi-nocloud   15s   Running   10.244.0.32   node01
vmi-slirp     1m    Running   10.244.0.31   node01
$ cluster/cli.sh ssh node01
$ sudo systemctl restart kubelet
$ cluster/kubectl.sh get vmis
NAME          AGE   PHASE     IP            NODENAME
vmi-nocloud   34s   Failed    10.244.0.32   node01
vmi-slirp     1m    Running   10.244.0.31   node01
```


Steps to Reproduce:
1.
2.
3.

Actual results:

The VMI with the default `bridge` options gets stopped.

Expected results:

No VMI gets restarted if the kubelet restarts.

Additional info:

https://github.com/kubevirt/kubevirt/issues/2076

Comment 1 Dan Kenigsberg 2019-06-20 11:57:25 UTC
Given our plans to eliminate the "bridge" binding mechanism for the Pod network ( https://jira.coreos.com/browse/KNIP-570 ) there is not reason track this bug further.


Note You need to log in before you can comment on or make changes to this bug.