In Alertmanager, when IP family in k8s Service is configured as "IPv6 only" on dual stack lab, Pod to pod communication happens over IPv4 address.

[AVESH-RAI]

What did you do?

Alertmanager is brought in HA mode on a dual stack K8s cluster. The Service configured is

`
args:

• --cluster.advertise-address=[$(POD_IP)]:9094
• --cluster.peer=alertmanager-pod-0.headless-svc.cluster.local:9094
• --cluster.peer=alertmanager-pod-1.headless-svc.cluster.local:9094
  `

Service.yaml
spec: ipFamilies: - IPv6 ipFamilyPolicy: SingleStack

What did you expect to see?

Pod should communicate over IPv6 address only.

What did you see instead? Under which circumstances?

We see that pod is communicating over IPv4 address, even when headless service is configured in IPv6 (IP family)

System information

Kubernetes

Alertmanager version

alertmanager-v0.28.1

### Alertmanager configuration file

```yaml

Prometheus version

Prometheus configuration file

Logs

[TheMeier]

What is the value of the env-var $POD_IP ?

[AVESH-RAI]

@TheMeier
the value of POD_IP is this.

`

• name: POD_IP
  valueFrom:
  fieldRef:
  apiVersion: v1
  fieldPath: status.podIP
  `

[TheMeier]

@AVESH-RAI that is the definition. I wanted to now if the actual vallue of the environment variable $POD_IP inside the pod, especially if it is an IPv4 address

You can get it with:
kubectl get pod -o jsonpath='{.status.podIP}'

[AVESH-RAI]

Hi @TheMeier
The value of POD_IP is:
POD_IP=IPv4,IPv6
The issue is i want the pod to communicate with each other in ipv6 address only in a dual stack cluster(IPv4,IPV6), since the headless service is in IPv6 address only.

[TheMeier]

That does not sound correct. I did spin up a kind cluster with dual stack if a look a some random pod status I see something likes this:

  hostIP: 10.89.0.3
  hostIPs:
  - ip: 10.89.0.3
  - ip: fc00:f853:ccd:e793::3
  phase: Running
  podIP: 10.89.0.3
  podIPs:
  - ip: 10.89.0.3
  - ip: fc00:f853:ccd:e793::3

So you probably want to try

name: POD_IP
  valueFrom:
    fieldRef:
      apiVersion: v1
      fieldPath: status.podIPs[1].ip

or something similar. Also if it POD_IP would contain IPv4,IPv6 alertmanager would fail to start with a message like this:

time=2026-01-07T11:28:08.081+01:00 level=ERROR source=main.go:262 msg="unable to initialize gossip mesh" err="invalid advertise address: address 10.89.0.3,fc00:f853:ccd:e793::3:9094: too many colons in address"

[AVESH-RAI]

@TheMeier
Yes ,

name: POD_IP
valueFrom:
fieldRef:
apiVersion: v1
fieldPath: status.podIPs[1].ip

I want the podIPs[1].ip to get the 2nd IP.

When we use status.podIP, we get only one IP. So that way alertmanager would work.
But when we set status.podIPs , we get both the IPs and thats where we want to set the 2nd IP for IPv6.

[TheMeier]

So is this resolved then?

[AVESH-RAI]

Hi @TheMeier

The value of status.podIPs.ip is not in the form of an array. It is a comma separated value. So we cannot get the 0th and 1st index from the status.podIPs.ip .
So the issue is still there.