Support for IPV6 in dual stack? #7116

Background for this question/request is related to the US government mandate detailing in this memo:

The Office of Management and Budget (OMB) has mandated that 80% of IP-enabled assets be IPv6-only by the end of fiscal year 2025.

Any updates here? I can see that the DB is listening on an IPv6 interface, however the k8s service is only SingleStack IPv6.

Bumping this also, @sxd. We'd like to set this behaviour in the kind: Cluster CR (or wherever most appropriate)

For example:

Kind: Cluster
...
    template:
      services:
        ipFamilies:
          - IPv6
        ipFamilyPolicy: SingleStack
...

Using OpenShift.

Thank you.

The only current way to do this is to define additional managed services within a cluster and specify the desired IP stack setup in those service templates. I use a managed stanza in my cluster definition similar to the following:

    managed:
      services:
        ## disable the default services
        disabledDefaultServices: ["ro", "r"]
        additional:
          - selectorType: rw
            serviceTemplate:
              metadata:
                name: "db-dbname-ds-rw"
              spec:
                type: ClusterIP
                ipFamilyPolicy: RequireDualStack
          - selectorType: r
            serviceTemplate:
              metadata:
                name: "db-dbname-ds-r"
              spec:
                type: ClusterIP
                ipFamilyPolicy: RequireDualStack
          - selectorType: ro
            serviceTemplate:
              metadata:
                name: "db-dbname-ds-ro"
              spec:
                type: ClusterIP
                ipFamilyPolicy: RequireDualStack

Assuming, for example, that you have proper DualStack support running on whichever nodes your cluster gets scheduled to (aka those Nodes have and know about and assign IPv4+IPv6 PodCIDRs), and ipv6 routing works however you do for your substrate (I use cilium-bgp in conjunction with frr BGP routing core for dumb/external nodes), a cluster with the above provisioned setup would be accessible via the ipv4 and ipv6 advertised addresses allocated on each of the db-dbname-ds-WHATEVER services. CNPG as of v1.28.1 does not allow disabling the default rw service.

Hi everyone, #10425 resolves this. Appreciate your upvotes for the PR.

apiVersion: postgresql.cnpg.io/v1
kind: Cluster
metadata:
  name: cluster-example-dualstack
spec:
  instances: 3

  managed:
    services:
      # This template is applied to all services (rw, ro, r) as a defaults
      # layer. Per-service values in additional managed services take precedence.
      serviceTemplate:
        spec:
          ipFamilyPolicy: RequireDualStack
          ipFamilies:
            - IPv4
            - IPv6

  storage:
    size: 1Gi