ipv6 + Calico + out of tree cloud-provider is broken #3401

CecileRobertMichon · 2023-02-22T21:09:10Z

While moving all CAPZ templates to out-of-tree CCM, I was not able to get a reliably passing test for ipv6. This test is very flaky (it fails most of the time). Looking into it I found that the error is that kubelet can't patch the status of one of the kube-system pods (sometimes kube-apiserver, sometimes kube-scheduler, sometimes kube-controller-manager). One example of the error can be seen here. The error is

	Pod \"kube-controller-manager-capz-e2e-4laer3-ipv6-control-plane-6v7dn\" is invalid: [status.podIPs: Invalid value: []core.PodIP{core.PodIP{IP:\"2001:1234:5678:9abc::5\"}, core.PodIP{IP:\"2001:1234:5678:9abc::5\"}}: may specify no more than one IP for each IP family, status.podIPs[1]: Duplicate value: core.PodIP{IP:\"2001:1234:5678:9abc::5\"}]"

The same error does not repro with the in-tree cloud-provider (see passing ipv6 test with CAPZ and in-tree cloud-provider here).

The cloud-provider-azure test for ipv6 is also failing (although the failure is different: https://testgrid.k8s.io/provider-azure-cloud-provider-azure#cloud-provider-azure-master-ipv6-capz)

The text was updated successfully, but these errors were encountered:

feiskyer · 2023-02-23T01:13:13Z

/assign @lzhecheng

feiskyer · 2023-02-23T01:15:02Z

The error is invalid Pod IPs, which doesn't seem to be related to ccm/kcm. But let's check the logs why there was duplicated IPs in pod status.

CecileRobertMichon · 2023-02-23T02:17:13Z

I agree it doesn't seem related to ccm, I think it might have to do with Calico somehow, but the weird thing is, I can't reproduce this error with the exact same template (Calico CNI too) when clou-provider is "azure" but as soon as I switch cloud-provider to "external" and install CCM I can repro easily.

lzhecheng · 2023-02-23T08:27:49Z

I created a cluster with 1) CAPZ of your branch (oot-templates); 2) templates/cluster-template-ipv6.yaml; 3) ci-entrypoint.sh. The cluster looks good. Could you please share how to reproduce locally?

zhechengli@devbox:~$ kgpa
NAMESPACE          NAME                                                         READY   STATUS    RESTARTS   AGE     IP                       NODE
                       NOMINATED NODE   READINESS GATES
calico-apiserver   calico-apiserver-78c5b59ff9-8v9r4                            1/1     Running   0          3m51s   2001:1234:5678:9a40::6   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
calico-apiserver   calico-apiserver-78c5b59ff9-v7hg9                            1/1     Running   0          3m51s   2001:1234:5678:9a42::3   zhecheng-223-5-md-0-5cqbp            <none>           <none>
calico-system      calico-kube-controllers-f7574cc46-jn87s                      1/1     Running   0          6m5s    2001:1234:5678:9a40::5   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
calico-system      calico-node-cts6q                                            1/1     Running   0          6m5s    2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
calico-system      calico-node-pv52g                                            1/1     Running   0          4m51s   2001:1234:5678:9abd::4   zhecheng-223-5-md-0-5cqbp            <none>           <none>
calico-system      calico-node-qxvsn                                            1/1     Running   0          4m55s   2001:1234:5678:9abd::5   zhecheng-223-5-md-0-rb7nl            <none>           <none>
calico-system      calico-typha-7859969dd9-rgxsx                                1/1     Running   0          6m6s    2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
calico-system      calico-typha-7859969dd9-z6ck2                                1/1     Running   0          4m47s   2001:1234:5678:9abd::5   zhecheng-223-5-md-0-rb7nl            <none>           <none>
calico-system      csi-node-driver-c7gtq                                        2/2     Running   0          4m10s   2001:1234:5678:9a42::2   zhecheng-223-5-md-0-5cqbp            <none>           <none>
calico-system      csi-node-driver-dgl62                                        2/2     Running   0          4m14s   2001:1234:5678:9a41::2   zhecheng-223-5-md-0-rb7nl            <none>           <none>
calico-system      csi-node-driver-h855n                                        2/2     Running   0          5m27s   2001:1234:5678:9a40::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
calico-system      tigera-operator-64db64cb98-p4k66                             1/1     Running   0          6m14s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        cloud-controller-manager-5c447487c8-j9695                    1/1     Running   0          6m10s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        cloud-node-manager-5rqwt                                     1/1     Running   0          4m55s   2001:1234:5678:9abd::5   zhecheng-223-5-md-0-rb7nl            <none>           <none>
kube-system        cloud-node-manager-dg8mh                                     1/1     Running   0          6m10s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        cloud-node-manager-jrpzb                                     1/1     Running   0          4m51s   2001:1234:5678:9abd::4   zhecheng-223-5-md-0-5cqbp            <none>           <none>
kube-system        coredns-565d847f94-d2btx                                     1/1     Running   0          6m23s   2001:1234:5678:9a40::2   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        coredns-565d847f94-q9cx4                                     1/1     Running   0          6m23s   2001:1234:5678:9a40::3   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        etcd-zhecheng-223-5-control-plane-xrhvp                      1/1     Running   0          6m29s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        kube-apiserver-zhecheng-223-5-control-plane-xrhvp            1/1     Running   0          6m28s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        kube-controller-manager-zhecheng-223-5-control-plane-xrhvp   1/1     Running   0          6m28s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        kube-proxy-j4bnp                                             1/1     Running   0          4m51s   2001:1234:5678:9abd::4   zhecheng-223-5-md-0-5cqbp            <none>           <none>
kube-system        kube-proxy-r9bzj                                             1/1     Running   0          4m55s   2001:1234:5678:9abd::5   zhecheng-223-5-md-0-rb7nl            <none>           <none>
kube-system        kube-proxy-wftz6                                             1/1     Running   0          6m23s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
kube-system        kube-scheduler-zhecheng-223-5-control-plane-xrhvp            1/1     Running   0          6m28s   2001:1234:5678:9abc::4   zhecheng-223-5-control-plane-xrhvp   <none>           <none>
...
zhechengli@devbox:~$ kgno -owide
NAME                                 STATUS   ROLES           AGE     VERSION   INTERNAL-IP              EXTERNAL-IP   OS-IMAGE             KERNEL-VERSION      CONTAINER-RUNTIME
zhecheng-223-5-control-plane-xrhvp   Ready    control-plane   7m8s    v1.25.3   2001:1234:5678:9abc::4   <none>        Ubuntu 22.04.1 LTS   5.15.0-1021-azure   containerd://1.6.2
zhecheng-223-5-md-0-5cqbp            Ready    <none>          5m28s   v1.25.3   2001:1234:5678:9abd::4   <none>        Ubuntu 22.04.1 LTS   5.15.0-1021-azure   containerd://1.6.2
zhecheng-223-5-md-0-rb7nl            Ready    <none>          5m32s   v1.25.3   2001:1234:5678:9abd::5   <none>        Ubuntu 22.04.1 LTS   5.15.0-1021-azure   containerd://1.6.2

CecileRobertMichon · 2023-02-27T20:08:45Z

@lzhecheng in my experience this doesn't repro every time but chances of reproing are a lot higher with 3+ control planes. I am usually able to repro easily within the first 4-5 clusters built.

lzhecheng · 2023-03-02T12:07:14Z

This time I created clusters with 3 control-plane nodes, for 3 times. They all look good. Is the problem still there?

CecileRobertMichon · 2023-03-03T01:13:08Z

@lzhecheng yes, see kubernetes-sigs/cluster-api-provider-azure#3221, e2e test failed ipv6 scenario: https://prow.k8s.io/view/gs/kubernetes-jenkins/pr-logs/pull/kubernetes-sigs_cluster-api-provider-azure/3221/pull-cluster-api-provider-azure-e2e/1631444499202838528

The only change compared to kubernetes-sigs/cluster-api-provider-azure#3105 where e2e test is passing for ipv6 is kubernetes-sigs/cluster-api-provider-azure@e3701e0

lzhecheng · 2023-03-03T09:45:41Z

Update:
There's another issue: machinedeployment cannot be available even if all pods and nodes are ready. E.g. https://prow.k8s.io/view/gs/kubernetes-jenkins/pr-logs/pull/kubernetes-sigs_cluster-api-provider-azure/3221/pull-cluster-api-provider-azure-e2e/1631562894430703616

I observed such issue locally as well.

Status:
  Conditions:
    Last Transition Time:  2023-03-03T06:48:48Z
    Message:               Minimum availability requires 1 replicas, current 0 available
    Reason:                WaitingForAvailableMachines
    Severity:              Warning
    Status:                False
    Type:                  Ready
    Last Transition Time:  2023-03-03T06:48:48Z
    Message:               Minimum availability requires 1 replicas, current 0 available
    Reason:                WaitingForAvailableMachines
    Severity:              Warning
    Status:                False
    Type:                  Available
  Observed Generation:     1
  Phase:                   ScalingUp
  Replicas:                1
  Selector:                cluster.x-k8s.io/cluster-name=zhecheng-303-1,cluster.x-k8s.io/deployment-name=zhecheng-303-1-md-0
  Unavailable Replicas:    1
  Updated Replicas:        1

CecileRobertMichon · 2023-03-03T21:42:16Z

@lzhecheng you're right this was a different error on the worker node because I forgot to add back the "external" cloud-provider configuration on the worker node kubelet config (fixed in kubernetes-sigs/cluster-api-provider-azure@051ae83)

Now that it's fixed, it is still failing and this time repro'ing the same issue (on the first try) I was seeing before where 1 of the control planes has a pending pod for one of the k8s components, in this case it's the second control plane's scheduler pod: https://storage.googleapis.com/kubernetes-jenkins/pr-logs/pull/kubernetes-sigs_cluster-api-provider-azure/3221/pull-cluster-api-provider-azure-e2e/1631738500053209088/artifacts/clusters/bootstrap/resources/capz-e2e-6l4j4b/Machine/capz-e2e-6l4j4b-ipv6-control-plane-m8ktt.yaml

in the kubelet log on that VM, we can see

Mar 03 20:15:26.475236 capz-e2e-6l4j4b-ipv6-control-plane-d5d4j kubelet[1476]: I0303 20:15:26.475205 1476 status_manager.go:691] "Failed to update status for pod" pod="kube-system/kube-scheduler-capz-e2e-6l4j4b-ipv6-control-plane-d5d4j" err="failed to patch status \"{\\\"metadata\\\":{\\\"uid\\\":\\\"0881123a-ea6b-401b-9459-8d7c2214f0d9\\\"},\\\"status\\\":{\\\"conditions\\\":[{\\\"lastProbeTime\\\":null,\\\"lastTransitionTime\\\":\\\"2023-03-03T19:47:56Z\\\",\\\"status\\\":\\\"True\\\",\\\"type\\\":\\\"Initialized\\\"},{\\\"lastProbeTime\\\":null,\\\"lastTransitionTime\\\":\\\"2023-03-03T19:48:15Z\\\",\\\"status\\\":\\\"True\\\",\\\"type\\\":\\\"Ready\\\"},{\\\"lastProbeTime\\\":null,\\\"lastTransitionTime\\\":\\\"2023-03-03T19:48:15Z\\\",\\\"status\\\":\\\"True\\\",\\\"type\\\":\\\"ContainersReady\\\"},{\\\"lastProbeTime\\\":null,\\\"lastTransitionTime\\\":\\\"2023-03-03T19:47:56Z\\\",\\\"status\\\":\\\"True\\\",\\\"type\\\":\\\"PodScheduled\\\"}],\\\"containerStatuses\\\":[{\\\"containerID\\\":\\\"containerd://3bb38c78e25b21aa91ebaa30ca67397e453970b68cb666648d81d4f94dd1625a\\\",\\\"image\\\":\\\"registry.k8s.io/kube-scheduler:v1.25.6\\\",\\\"imageID\\\":\\\"registry.k8s.io/kube-scheduler@sha256:f41301881252779d21dde86aac5a45e9acfe560643b5a28cef1286eabb187e26\\\",\\\"lastState\\\":{},\\\"name\\\":\\\"kube-scheduler\\\",\\\"ready\\\":true,\\\"restartCount\\\":0,\\\"started\\\":true,\\\"state\\\":{\\\"running\\\":{\\\"startedAt\\\":\\\"2023-03-03T19:48:01Z\\\"}}}],\\\"hostIP\\\":\\\"10.0.0.5\\\",\\\"phase\\\":\\\"Running\\\",\\\"podIP\\\":\\\"2001:1234:5678:9abc::5\\\",\\\"podIPs\\\":[{\\\"ip\\\":\\\"2001:1234:5678:9abc::5\\\"},{\\\"ip\\\":\\\"2001:1234:5678:9abc::5\\\"}],\\\"startTime\\\":\\\"2023-03-03T19:47:56Z\\\"}}\" for pod \"kube-system\"/\"kube-scheduler-capz-e2e-6l4j4b-ipv6-control-plane-d5d4j\": Pod \"kube-scheduler-capz-e2e-6l4j4b-ipv6-control-plane-d5d4j\" is invalid: [status.podIPs: Invalid value: []core.PodIP{core.PodIP{IP:\"2001:1234:5678:9abc::5\"}, core.PodIP{IP:\"2001:1234:5678:9abc::5\"}}: may specify no more than one IP for each IP family, status.podIPs[1]: Duplicate value: core.PodIP{IP:\"2001:1234:5678:9abc::5\"}]"

lzhecheng · 2023-03-06T01:29:38Z

@CecileRobertMichon thank you for the fix.
I have a question: Is the 3rd control plane node not coming up a separate issue or a result of this "invalid podIPs" issue? Do you think if it is expected?

CecileRobertMichon · 2023-03-06T17:32:59Z

@lzhecheng the 3rd control plane node is not coming up because of the "invalid pod IPs" issue in the 2nd control plane: Cluster API checks the health of all control plane components before scaling up and in this case the check is failing because the scheduler pod is not healthy

lzhecheng · 2023-03-08T13:29:23Z

@CecileRobertMichon Update: I ran ci-e2e.sh locally several times and have 2 findings:

It succeeded most of times when only running the ipv6 test
When running all 3 required tests, I saw such logs:

E0308 10:04:59.429233       1 controller.go:326]  "msg"="Reconciler error" "error"="admission webhook \"validation.azurecluster.infrastructure.cluster.x-k8s.io\" denied the request: AzureCluster.infrastructure.cluster.x-k8s.io \"zhecheng-308-1\" is invalid: spec.networkSpec.subnets[1].CIDRBlocks: Invalid value: []string{\"10.1.0.0/16\", \"2001:1234:5678:9abd::/64\"}: field is immutable" "AzureCluster"={"name":"zhecheng-308-1","namespace":"capz-e2e-c0c9z3"} "controller"="azurecluster" "controllerGroup"="infrastructure.cluster.x-k8s.io" "controllerKind"="AzureCluster" "name"="zhecheng-308-1" "namespace"="capz-e2e-c0c9z3" "reconcileID"="c8f083d8-d0ef-420d-b950-025ee090b76d"

Here capz-e2e-c0c9z3 is not the namespace for ipv6 cluster.
My guess is that there's a bug when creating clusters simultaneously. I am not 100% sure so I just want to highlight the possibility.
Could you please:

Rebase your PR. Then I can also test it with CI pipeline by cherrypicking your commits.
Do experiments with CI pipeline: Try to only run ipv6 test or run the 3 required tests sequentially.
Thank you.

CecileRobertMichon · 2023-03-08T20:06:33Z

I just rebased the PR so it now only has the diff for switching ipv6. Running the job again then will disable everything but ipv6 in a test commit and run that.

lzhecheng · 2023-03-15T12:10:32Z

Update: Reproduced on a manually created cluster, restarting these static manifest Pods will get recovered.

lzhecheng · 2023-03-17T15:25:34Z

@CecileRobertMichon is it possible I can create a cluster with a kubelet binary built locally by myself?

kubelet seems to update Pod status with unexpected host IPs and I'd like to add logs to check it.

CecileRobertMichon · 2023-03-17T17:33:54Z

@lzhecheng yes but you'll need to create a template based on the custom-builds.yaml template that uses out of tree + ipv6 (that's not a combination of features we have in the templates in the repo unfortunately): https://capz.sigs.k8s.io/developers/kubernetes-developers.html#kubernetes-117

lzhecheng · 2023-03-22T13:20:22Z

Update:
I think I found the root cause: kubelet is not properly processing host IPs. OOT cloud-provider behaves little differently from before, which directly leads to failure but I think it is kubelet that should handle the situation.
Will do further verification tomorrow. And then make a fix and detailed root cause analysis.

lzhecheng · 2023-03-23T03:08:22Z

Root cause: kubelet always adds the second host IP into PodIPs without checking if it is the same IP as PodIPs[0] or the same IP family. While the order of hostIPs is not ensured.

			if kubecontainer.IsHostNetworkPod(pod) {
				// Primary IP is not set
				if s.PodIP == "" {
					s.PodIP = hostIPs[0].String()
					s.PodIPs = []v1.PodIP{{IP: s.PodIP}}
				}
				// Secondary IP is not set #105320
				if len(hostIPs) == 2 && len(s.PodIPs) == 1 {
					s.PodIPs = append(s.PodIPs, v1.PodIP{IP: hostIPs[1].String()})
				}
			}

https://github.com/kubernetes/kubernetes/blob/v1.25.6/pkg/kubelet/kubelet_pods.go#L1534-L1544
In our case, IPv6 IP is added to Node first, hostIPs: [<IPv6>]. Then the IPv4 one is added, hostIPs: [<IPv4>, <IPv6>]. As a result, IPv6 addr is added to PodIPs twice. I added more log code into kubelet and the result:

# hostIPs: IPv6
Mar 22 11:48:08 zhecheng-322-18-control-plane-759sz kubelet[1479]: I0322 11:48:08.326325    1479 kubelet_pods.go:1536] DEBUG hostIPs [2001:1234:5678:9abc::6]
Mar 22 11:48:08 zhecheng-322-18-control-plane-759sz kubelet[1479]: I0322 11:48:08.326434    1479 kubelet_pods.go:1546] DEBUG s.PodIPs [{2001:1234:5678:9abc::6}]

# hostIPs: IPv4, IPv6
Mar 22 11:50:05 zhecheng-322-18-control-plane-759sz kubelet[1479]: I0322 11:50:05.222657    1479 kubelet_pods.go:1536] DEBUG hostIPs [10.0.0.6 2001:1234:5678:9abc::6]
Mar 22 11:50:05 zhecheng-322-18-control-plane-759sz kubelet[1479]: I0322 11:50:05.222673    1479 kubelet_pods.go:1546] DEBUG s.PodIPs [{2001:1234:5678:9abc::6} {2001:1234:5678:9abc::6}]

The fix: Check IP family before adding the IP into PodIPs
kubernetes/kubernetes#116879

Why after switching to OOT cloud-provider-azure, the problem appears:
I think the order of in-tree cloud-provider is hostIPs: [<IPv6>, <IPv4>] so the problem is not revealed. It can be further investigated.

@CecileRobertMichon @feiskyer

CecileRobertMichon · 2023-03-23T23:49:33Z

I think the order of in-tree cloud-provider is hostIPs: [, ] so the problem is not revealed. It can be further investigated.

Would it be possible to switch the order in out of tree cloud-provider too so the order is consistent with in tree cloud-provider?

lzhecheng · 2023-03-24T01:28:15Z

I think the order of in-tree cloud-provider is hostIPs: [, ] so the problem is not revealed. It can be further investigated.

Would it be possible to switch the order in out of tree cloud-provider too so the order is consistent with in tree cloud-provider?

Yes, I am also working on it.

lzhecheng · 2023-03-24T12:08:39Z

@CecileRobertMichon this is the PR to fix the order issue in cloud-provider-azure code: #3643
But I wonder if we can make the fix easier. As you can see, the problem is that IPv4 is always handled first then IPv6.

cloud-provider-azure/pkg/provider/azure_instances.go

Lines 127 to 152 in cdd5003

    
           if len(netInterface.IPV4.IPAddress) > 0 && len(netInterface.IPV4.IPAddress[0].PrivateIP) > 0 { 
        
           	address := netInterface.IPV4.IPAddress[0] 
        
           	addresses = append(addresses, v1.NodeAddress{ 
        
           		Type:    v1.NodeInternalIP, 
        
           		Address: address.PrivateIP, 
        
           	}) 
        
           	if len(address.PublicIP) > 0 { 
        
           		addresses = append(addresses, v1.NodeAddress{ 
        
           			Type:    v1.NodeExternalIP, 
        
           			Address: address.PublicIP, 
        
           		}) 
        
           	} 
        
           } 
        
           if len(netInterface.IPV6.IPAddress) > 0 && len(netInterface.IPV6.IPAddress[0].PrivateIP) > 0 { 
        
           	address := netInterface.IPV6.IPAddress[0] 
        
           	addresses = append(addresses, v1.NodeAddress{ 
        
           		Type:    v1.NodeInternalIP, 
        
           		Address: address.PrivateIP, 
        
           	}) 
        
           	if len(address.PublicIP) > 0 { 
        
           		addresses = append(addresses, v1.NodeAddress{ 
        
           			Type:    v1.NodeExternalIP, 
        
           			Address: address.PublicIP, 
        
           		}) 
        
           	} 
        
           }

I want to add logic there to avoid reorder later. So, here's the question:
Is it the process same for dualstack and IPv6 to add addresses to network interface? IPv6 cluster is IPv6 first then IPv4. Is dualstack cluster IPv4 first then IPv6? If so, we have to use my fix above. If not, means the process is the same, then as long as we have 2 internal IPs, we can always put IPv6 first.

CecileRobertMichon · 2023-03-24T16:36:08Z

Is it the process same for dualstack and IPv6 to add addresses to network interface? IPv6 cluster is IPv6 first then IPv4. Is dualstack cluster IPv4 first then IPv6?

I don't know the answer to this. Maybe @aramase does?

Edit: I just realized you meant in CAPZ. For network interfaces, it looks like IPv4 is always first: https://github.com/kubernetes-sigs/cluster-api-provider-azure/blob/main/azure/services/networkinterfaces/spec.go#L201

IPv6 cluster is IPv6 first then IPv4

where do you see this?

lzhecheng · 2023-03-25T00:58:01Z

@CecileRobertMichon it is interesting. According to logs in CNM, IPv6 comes first and then the IPv4. These IPs comes from instance metadata.

cloud-provider-azure/pkg/provider/azure_instances.go

Line 109 in a2758f6

    
           return az.getLocalInstanceNodeAddresses(metadata.Network.Interface, string(name))

# I added some logs in CNM to help debug
I0324 11:46:49.170166       1 nodemanager.go:284] DEBUG old nodeAddresses [{InternalIP 2001:1234:5678:9abc::5} {Hostname zhecheng-324-6-control-plane-gbmtr}]
I0324 11:46:49.170399       1 nodemanager.go:285] DEBUG new nodeAddresses [{Hostname zhecheng-324-6-control-plane-gbmtr} {InternalIP 2001:1234:5678:9abc::5} {InternalIP 10.0.0.5}]

I am not familiar with the CAPZ code, but is it possible s.IPConfigs is an empty slice in your reference?

jackfrancis · 2023-03-25T01:04:51Z

This sounds like a similar symptom that we discovered when using secondary IPs w/ Azure CNI.

CecileRobertMichon · 2023-03-25T01:07:52Z

I am not familiar with the CAPZ code, but is it possible s.IPConfigs is an empty slice in your reference?

No, we always add the ipv4 ip config as the primary IP: https://github.com/kubernetes-sigs/cluster-api-provider-azure/blob/main/azure/services/networkinterfaces/spec.go#L151

lzhecheng · 2023-03-27T08:49:12Z

I am not familiar with the CAPZ code, but is it possible s.IPConfigs is an empty slice in your reference?

No, we always add the ipv4 ip config as the primary IP: https://github.com/kubernetes-sigs/cluster-api-provider-azure/blob/main/azure/services/networkinterfaces/spec.go#L151

If there's the case, I think IMDS is not working properly.

lzhecheng · 2023-03-27T09:07:12Z

From CNM, it gets host IPs with InstanceMetadataService and always returns IPv4 first then IPv6. If there's no such case that CNM gets a single IPv6 first then 2 IPs, this issue won't happen.

aojea · 2023-03-27T16:15:25Z

@lzhecheng is kubelet configuring the --node-ip address in these clusters? because if that is the case, the controller must return the node.status.addresses in the same order as the --node-ip family

CecileRobertMichon · 2023-03-27T22:53:21Z

Here is an example of a dual-stack cluster node:

On the Azure portal the NIC shows the ipv4 address first on the NIC:

And same thing in node addresses:

status:
  addresses:
  - address: dual-stack-25113-md-0-478wl
    type: Hostname
  - address: 10.1.0.4
    type: InternalIP
  - address: 2001:1234:5678:9abd::4
    type: InternalIP

And here is an ipv6 single stack cluster:

But note that this time the ipv6 address shows up first in the node IPs:

status:
  addresses:
  - address: ipv6-4530-md-0-f7sdx
    type: Hostname
  - address: 2001:1234:5678:9abd::4
    type: InternalIP
  - address: 10.1.0.4
    type: InternalIP

CecileRobertMichon · 2023-03-27T23:03:41Z

is kubelet configuring the --node-ip address in these clusters

For the ipv6 cluster, kubelet --node-ip is set to ::

For the dual stack cluster, it is not set (assuming it is getting defaulted)

lzhecheng · 2023-03-28T01:49:38Z

is kubelet configuring the --node-ip address in these clusters

For the ipv6 cluster, kubelet --node-ip is set to ::

For the dual stack cluster, it is not set (assuming it is getting defaulted)

Maybe for IPv6, --node-ip should not be set either since it is actually a dual-stack cluster as for Nodes? When someday the cluster is truly IPv6 only, this option should be set.

CecileRobertMichon · 2023-03-28T16:35:49Z

Let me try without it. cc @aramase

aojea · 2023-03-28T17:23:33Z

@CecileRobertMichon @lzhecheng please check https://github.com/kubernetes/enhancements/tree/master/keps/sig-network/3705-cloud-node-ips , this is WIP and a known and complicated problem, we also have some ammends because of the complexity kubernetes/enhancements#3898

Please check the KEP to see if you are case is contemplated and if not, please let us know

CecileRobertMichon · 2023-03-29T23:22:55Z

Let me try without it

update: the e2e test passed several times in a row when node-ip is not set kubernetes-sigs/cluster-api-provider-azure#3221

lzhecheng · 2023-03-30T01:39:12Z

Let me try without it

update: the e2e test passed several times in a row when node-ip is not set kubernetes-sigs/cluster-api-provider-azure#3221

Glad to know that. I think CNM doesn't need to change because as you said, IPv4 first and then IPv6 is the expected behaviour for CAPZ dualstack and IPv6 only.

CecileRobertMichon · 2023-03-30T17:41:39Z

kubernetes-sigs/cluster-api-provider-azure#3221 is ready for review/merge

lzhecheng · 2023-04-20T05:26:55Z

/close
The remaining k/k PR is an improvement, not the direct fix for this issue. The actual fix is on CAPZ side and has been applied.

k8s-ci-robot · 2023-04-20T05:26:59Z

@lzhecheng: Closing this issue.

In response to this:

/close
The remaining k/k PR is an improvement, not the direct fix for this issue. The actual fix is on CAPZ side and has been applied.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

CecileRobertMichon added the kind/bug Categorizes issue or PR as related to a bug. label Feb 22, 2023

CecileRobertMichon mentioned this issue Feb 22, 2023

Switch flavor and test templates to external cloud-provider kubernetes-sigs/cluster-api-provider-azure#3105

Merged

3 tasks

k8s-ci-robot assigned lzhecheng Feb 23, 2023

lzhecheng mentioned this issue Mar 2, 2023

Cannot create a VMSS dual-stack cluster kubernetes-sigs/cluster-api-provider-azure#3163

Closed

lzhecheng mentioned this issue Mar 24, 2023

[WIP] Fix incorrect order of Node internal address issue #3643

Closed

CecileRobertMichon mentioned this issue Mar 30, 2023

Switch ipv6 template to external cloud-provider kubernetes-sigs/cluster-api-provider-azure#3221

Merged

3 tasks

k8s-ci-robot closed this as completed Apr 20, 2023

ipv6 + Calico + out of tree cloud-provider is broken #3401

ipv6 + Calico + out of tree cloud-provider is broken #3401

Comments

CecileRobertMichon commented Feb 22, 2023

feiskyer commented Feb 23, 2023

feiskyer commented Feb 23, 2023

CecileRobertMichon commented Feb 23, 2023

lzhecheng commented Feb 23, 2023

CecileRobertMichon commented Feb 27, 2023

lzhecheng commented Mar 2, 2023

CecileRobertMichon commented Mar 3, 2023

lzhecheng commented Mar 3, 2023

CecileRobertMichon commented Mar 3, 2023 • edited Loading

lzhecheng commented Mar 6, 2023

CecileRobertMichon commented Mar 6, 2023

lzhecheng commented Mar 8, 2023

CecileRobertMichon commented Mar 8, 2023 • edited Loading

lzhecheng commented Mar 15, 2023

lzhecheng commented Mar 17, 2023

CecileRobertMichon commented Mar 17, 2023

lzhecheng commented Mar 22, 2023

lzhecheng commented Mar 23, 2023 • edited Loading

CecileRobertMichon commented Mar 23, 2023

lzhecheng commented Mar 24, 2023

lzhecheng commented Mar 24, 2023

CecileRobertMichon commented Mar 24, 2023 • edited Loading

lzhecheng commented Mar 25, 2023 • edited Loading

jackfrancis commented Mar 25, 2023

CecileRobertMichon commented Mar 25, 2023

lzhecheng commented Mar 27, 2023

lzhecheng commented Mar 27, 2023

aojea commented Mar 27, 2023 • edited Loading

CecileRobertMichon commented Mar 27, 2023

CecileRobertMichon commented Mar 27, 2023

lzhecheng commented Mar 28, 2023

CecileRobertMichon commented Mar 28, 2023

aojea commented Mar 28, 2023

CecileRobertMichon commented Mar 29, 2023

lzhecheng commented Mar 30, 2023

CecileRobertMichon commented Mar 30, 2023

lzhecheng commented Apr 20, 2023

k8s-ci-robot commented Apr 20, 2023

CecileRobertMichon commented Mar 3, 2023 •

edited

Loading

CecileRobertMichon commented Mar 8, 2023 •

edited

Loading

lzhecheng commented Mar 23, 2023 •

edited

Loading

CecileRobertMichon commented Mar 24, 2023 •

edited

Loading

lzhecheng commented Mar 25, 2023 •

edited

Loading

aojea commented Mar 27, 2023 •

edited

Loading