Skip to main content
Calico Cloud documentation

Advertise Kubernetes service IP addresses

Big picture

Enable Calico Cloud to advertise Kubernetes service IPs outside a cluster. Calico Cloud supports advertising a service’s cluster IPs and external IPs.

Value

Typically, Kubernetes service cluster IPs are accessible only within the cluster, so external access to the service requires a dedicated load balancer or ingress controller. In cases where a service’s cluster IP is not routable, the service can be accessed using its external IP.

Just as Calico Cloud supports advertising pod IPs over BGP, it also supports advertising Kubernetes service IPs outside a cluster over BGP. This avoids the need for a dedicated load balancer. This feature also supports equal cost multi-path (ECMP) load balancing across nodes in the cluster, as well as source IP address preservation for local services when you need more control.

Concepts

BGP makes it easy

In Kubernetes, all requests for a service are redirected to an appropriate endpoint (pod) backing that service. Because Calico Cloud uses BGP, external traffic can be routed directly to Kubernetes services by advertising Kubernetes service IPs into the BGP network.

If your deployment is configured to peer with BGP routers outside the cluster, those routers (plus any other upstream places the routers propagate to) can send traffic to a Kubernetes service IP for routing to one of the available endpoints for that service.

Advertising service IPs: quick glance

Calico Cloud implements the Kubernetes externalTrafficPolicy using kube-proxy to direct incoming traffic to a correct pod. Advertisement is handled differently based on the service type that you configure for your service.

Service modeCluster IP advertisementTraffic is...Source IP address is...
Cluster (default)All nodes in the cluster statically advertise a route to the service CIDR.Load balanced across nodes in the cluster using ECMP, then forwarded to appropriate pod in the service using SNAT. May incur second hop to another node, but good overall load balancing.Obscured by SNAT
LocalThe nodes with a pod backing the service advertise a specific route (/32 or /128) to the service's IP.Load balanced across nodes with endpoints for the service. Avoids second hop for LoadBalancer and NodePort type services, traffic may be unevenly load balanced. (Other traffic is load balanced across nodes in the cluster.)Preserved

If your Calico Cloud deployment is configured to peer with BGP routers outside the cluster, those routers - plus any further upstream places that those routers propagate to - will be able to send traffic to a Kubernetes service cluster IP, and that traffic is routed to one of the available endpoints for that service.

Tips for success

  • Generally, we recommend using “Local” for the following reasons:
    • If any of your network policy uses rules to match by specific source IP addresses, using Local is the obvious choice because the source IP address is not altered, and the policy will still work.
    • Return traffic is routed directly to the source IP because “Local” services do not require undoing the source NAT (unlike “Cluster” services).
  • Cluster IP advertisement works best with a ToR that supports ECMP. Otherwise, all traffic for a given route is directed to a single node.

Before you begin...

Required

  • Calico CNI
  • Configure BGP peering between Calico Cloud and your network infrastructure
  • For ECMP load balancing to services, the upstream routers must be configured to use BGP multipath.
  • You need at least one external node outside the cluster that acts as a router, route reflector, or ToR that is peered with calico nodes inside the cluster.
  • Services must be configured with the correct service mode (“Cluster” or “Local”) for your implementation. For externalTrafficPolicy: Local, the service must be type LoadBalancer or NodePort.

Limitations

  • Supported in EKS and AWS, but only if you are using Calico CNI

  • OpenShift, versions 4.5 and 4.6
    There is a bug where the source IP is not preserved by NodePort services or traffic via a Service ExternalIP with externalTrafficPolicy:Local.

    OpenShift users on v4.5 or v4.6 can use this workaround to avoid SNAT with ExternalIP:

      oc edit featuregates.config.openshift.io cluster
    spec:
    customNoUpgrade:
    enabled:
    - ExternalPolicyForExternalIP

    Kubernetes users on version v1.18 or v1.19 can enable source IP preservation for NodePort services using the ExternalPolicyForExternalIP feature gate.

    Source IP preservation for NodePort and services and ExternalIPs is enabled by default in OpenShift v4.7+, and Kubernetes v1.20+.

How to

  1. Determine the service cluster IP range. (Or ranges, if your cluster is dual stack.)

    The range(s) for your cluster can be inferred from the --service-cluster-ip-range option passed to the Kubernetes API server. For help, see the Kubernetes API server reference guide.

  2. Check to see if you have a default BGPConfiguration.

    kubectl get bgpconfiguration.projectcalico.org default
  3. Based on above results, update or create a BGPConfiguration.

    Update default BGPConfiguration.

    Patch the BGPConfiguration using the following command, using your own service cluster IP CIDR in place of "10.0.0.0/24":

    kubectl patch bgpconfiguration.projectcalico.org default -p '{"spec":{"serviceClusterIPs": [{"cidr": "10.0.0.0/24"}]}}'

    Create default BGPConfiguration.

    Use the following sample command to create a default BGPConfiguration. Add your CIDR blocks, covering the cluster IPs to be advertised, in the serviceClusterIPs field, for example:

    kubectl create -f - <<EOF
    apiVersion: projectcalico.org/v3
    kind: BGPConfiguration
    metadata:
    name: default
    spec:
    serviceClusterIPs:
    - cidr: 10.96.0.0/16
    - cidr: fd00:1234::/112
    EOF

    For help, see BGP configuration resource.

note

In earlier versions of Calico Cloud, and for IPv4 only, service cluster IP advertisement was configured via the environment variable CALICO_ADVERTISE_CLUSTER_IPS. That environment variable takes precedence over any serviceClusterIPs configured in the default BGPConfiguration. We recommend replacing the deprecated CALICO_ADVERTISE_CLUSTER_IPS with BGPConfiguration.

Service cluster IP advertisement of /32 routes, by default, is enabled only for NodePort and LoadBalancer type services. In some routing infrastructures, you may want to advertise /32 routes of ClusterIP services as well. This is enabled on a per-service basis by adding the annotation projectcalico.org/AdvertiseClusterIP: true to the service. For example:

kubectl annotate service your-service "projectcalico.org/AdvertiseClusterIP=true"
note

You will still need to enable service cluster IP advertisement via BGP configuration as shown above.

  1. Identify the external IP ranges of all services that you want to advertise outside of the Calico Cloud cluster.

  2. Check to see if you have a default BGPConfiguration.

    kubectl get bgpconfiguration.projectcalico.org default
  3. Based on above results, update or create a BGPConfiguration.

    Update default BGPConfiguration

    Patch the BGPConfiguration using the following command, adding your own service external IP CIDRs:

    kubectl patch bgpconfiguration.projectcalico.org default -p '{"spec":{"serviceExternalIPs": [{"cidr": "x.x.x.x"}, {"cidr": "y.y.y.y"}]}}'

    Create default BGPConfiguration

    Use the following sample command to create a default BGPConfiguration. Add your CIDR blocks for external IPs to be advertised in the serviceExternalIPs field.

    kubectl create -f - <<EOF
    apiVersion: projectcalico.org/v3
    kind: BGPConfiguration
    metadata:
    name: default
    spec:
    serviceExternalIPs:
    - cidr: x.x.x.x/16
    - cidr: y.y.y.y/32
    EOF

    For help, see BGP configuration resource.

  4. When configuring a Kubernetes service that you want to be reachable via an external IP, specify that external IP in the service's externalIPs field.

The following steps will configure Calico Cloud to advertise Service status.LoadBalancer.Ingress.IP addresses.

  1. Identify the IP ranges to be used for Service LoadBalancer address allocation.

  2. Check to see if you have a default BGPConfiguration.

    kubectl get bgpconfiguration default
  3. Based on above results, update or create a BGPConfiguration.

    Update default BGPConfiguration Patch the BGPConfiguration using the following command, adding your own service load balancer IP CIDRs:

    kubectl patch bgpconfiguration default --patch '{"spec": {"serviceLoadBalancerIPs": [{"cidr": "x.x.x.x/16"}]}}'

    Create default BGPConfiguration Use the following sample command to create a default BGPConfiguration. Add your CIDR blocks for load balancer IPs to be advertised in the serviceLoadBalancerIPs field.

    kubectl create -f - <<EOF
    apiVersion: projectcalico.org/v3
    kind: BGPConfiguration
    metadata:
    name: default
    spec:
    serviceLoadBalancerIPs:
    - cidr: x.x.x.x/16
    EOF

    For help, see BGP configuration resource.

Service LoadBalancer address allocation is outside the current scope of Calico Cloud, but can be implemented with an external controller. You can build your own, or use a third-party implementation like the MetalLB project.

To install the MetalLB controller for allocating addresses, perform the following steps.

  1. Follow the MetalLB documentation to install the metallb-system/controller resources.

    However, do not install the metallb-system/speaker component. The speaker component also attempts to establish BGP sessions on the node, and will conflict with Calico.

  2. Configure MetalLB to provision addresses by creating the following IPAddressPool, replacing x.x.x.x/16 with the CIDR given to Calico Cloud in the steps above. Please note IPAddressPool requires Metallb 0.13+.

    kubectl create -f - <<EOF
    kind: IPAddressPool
    metadata:
    name: default
    namespace: metallb-system
    spec:
    addresses:
    - x.x.x.x/16
    EOF

Exclude certain nodes from advertisement

In some cases, you may want to exclude certain nodes from advertising service addresses. For example, control plane nodes that do not host any services themselves.

To remove a node from service advertisement, apply the label node.kubernetes.io/exclude-from-external-load-balancers=true.

For example, to exclude the node control-plane-01 from service advertisement, you can run the following command:

kubectl label node control-plane-01 node.kubernetes.io/exclude-from-external-load-balancers=true

Tutorial

For a tutorial on how service advertisement works with Calico Cloud, see the blog Kubernetes Service IP Route Advertisement.

Additional resources

Other topics on creating network policy for Kubernetes services: