Service ClusterIP allocation

In Kubernetes, Services are an abstract way to expose an application running on a set of Pods. Services can have a cluster-scoped virtual IP address (using a Service of type: ClusterIP). Clients can connect using that virtual IP address, and Kubernetes then load-balances traffic to that Service across the different backing Pods.

How Service ClusterIPs are allocated?

When Kubernetes needs to assign a virtual IP address for a Service, that assignment happens one of two ways:

dynamically

the cluster’s control plane automatically picks a free IP address from within the configured IP range for type: ClusterIP Services.

statically

you specify an IP address of your choice, from within the configured IP range for Services.

Across your whole cluster, every Service ClusterIP must be unique. Trying to create a Service with a specific ClusterIP that has already been allocated will return an error.

Why do you need to reserve Service Cluster IPs?

Sometimes you may want to have Services running in well-known IP addresses, so other components and users in the cluster can use them.

The best example is the DNS Service for the cluster. As a soft convention, some Kubernetes installers assign the 10th IP address from the Service IP range to the DNS service. Assuming you configured your cluster with Service IP range 10.96.0.0/16 and you want your DNS Service IP to be 10.96.0.10, you’d have to create a Service like this:

  1. apiVersion: v1
  2. kind: Service
  3. metadata:
  4. labels:
  5. k8s-app: kube-dns
  6. kubernetes.io/cluster-service: "true"
  7. kubernetes.io/name: CoreDNS
  8. name: kube-dns
  9. namespace: kube-system
  10. spec:
  11. clusterIP: 10.96.0.10
  12. ports:
  13. - name: dns
  14. port: 53
  15. protocol: UDP
  16. targetPort: 53
  17. - name: dns-tcp
  18. port: 53
  19. protocol: TCP
  20. targetPort: 53
  21. selector:
  22. k8s-app: kube-dns
  23. type: ClusterIP

but as it was explained before, the IP address 10.96.0.10 has not been reserved; if other Services are created before or in parallel with dynamic allocation, there is a chance they can allocate this IP, hence, you will not be able to create the DNS Service because it will fail with a conflict error.

How can you avoid Service ClusterIP conflicts?

The allocation strategy implemented in Kubernetes to allocate ClusterIPs to Services reduces the risk of collision.

The ClusterIP range is divided, based on the formula min(max(16, cidrSize / 16), 256), described as never less than 16 or more than 256 with a graduated step between them.

Dynamic IP assignment uses the upper band by default, once this has been exhausted it will use the lower range. This will allow users to use static allocations on the lower band with a low risk of collision.

Examples

Example 1

This example uses the IP address range: 10.96.0.0/24 (CIDR notation) for the IP addresses of Services.

Range Size: 28 - 2 = 254
Band Offset: min(max(16, 256/16), 256) = min(16, 256) = 16
Static band start: 10.96.0.1
Static band end: 10.96.0.16
Range end: 10.96.0.254

pie showData title 10.96.0.0/24 “Static” : 16 “Dynamic” : 238

JavaScript must be enabled to view this content

Example 2

This example uses the IP address range: 10.96.0.0/20 (CIDR notation) for the IP addresses of Services.

Range Size: 212 - 2 = 4094
Band Offset: min(max(16, 4096/16), 256) = min(256, 256) = 256
Static band start: 10.96.0.1
Static band end: 10.96.1.0
Range end: 10.96.15.254

pie showData title 10.96.0.0/20 “Static” : 256 “Dynamic” : 3838

JavaScript must be enabled to view this content

Example 3

This example uses the IP address range: 10.96.0.0/16 (CIDR notation) for the IP addresses of Services.

Range Size: 216 - 2 = 65534
Band Offset: min(max(16, 65536/16), 256) = min(4096, 256) = 256
Static band start: 10.96.0.1
Static band ends: 10.96.1.0
Range end: 10.96.255.254

pie showData title 10.96.0.0/16 “Static” : 256 “Dynamic” : 65278

JavaScript must be enabled to view this content

What’s next