Skip to main content
Version: 3.28 (latest)

Migrate from one IP pool to another

Big picture​

Migrate pods from one IP pool to another on a running cluster without network disruption.

Value​

Pods are assigned IP addresses from IP pools that you configure in Calico. As the number of pods increase, you may need to increase the number of addresses available for pods to use. Or, you may need to move pods from a CIDR that was used by mistake. Calico lets you migrate from one IP pool to another one on a running cluster without network disruption.

Concepts​

IP pools and cluster CIDRs​

Calico supports using multiple disjoint IP pool CIDRs within the cluster. However, Kubernetes expects that all pods have addresses within the same cluster CIDR. This means that although it is technically feasible to create an IP pool outside of the cluster CIDR, we do not recommend it. Pods allocated addresses outside of the Kubernetes cluster CIDR will lose network connectivity.

Before you begin...​

Verify that you are using Calico IPAM.

If you are not sure which IPAM your cluster is using, the way to tell depends on install method.

The IPAM plugin can be queried on the default Installation resource.

kubectl get installation default -o go-template --template {{.spec.cni.ipam.type}}

If your cluster is using Calico IPAM, the above command should return a result of Calico.

Verify orchestrator support for changing the pod network CIDR.

Although Kubernetes supports changing the pod network CIDR, not all orchestrators do. Check your orchestrator documentation to verify.

How to​

Migrate from one IP pool to another​

Follow these steps to migrate pods from one IP pool to another pool.

note

If you follow these steps, existing pod connectivity will not be affected. (If you delete the old IP pool before you create and verify the new pool, existing pods will be affected.) When pods are deleted, applications may be temporarily unavailable (depending on the type of application); plan accordingly.

  1. Add a new IP pool.
note

It is highly recommended that your Calico IP pools are within the Kubernetes cluster CIDR. If pods IPs are allocated from outside of the Kubernetes cluster CIDR, some traffic flows may have NAT applied unnecessarily causing unexpected behavior.

  1. Disable the old IP pool.
note

Disabling an IP pool only prevents new IP address allocations; it does not affect the networking of existing pods.

  1. Delete pods from the old IP pool. This includes any new pods that may have been created with the old IP pool prior to disabling the pool.

  2. Verify that new pods get an address from the new IP pool.

  3. Delete the old IP pool.

Tutorial​

In the following example, we created a Kubernetes cluster using kubeadm. But the IP pool CIDR we configured (192.168.0.0/16) doesn't match the Kubernetes cluster CIDR. Let's change the CIDR to 10.0.0.0/16, which for the purposes of this example falls within the cluster CIDR.

Let’s run kubectl get ippools to see the IP pool default-ipv4-ippool.

NAME                  CREATED AT
default-ipv4-ippool   2024-03-28T16:14:28Z

Step 1: Add a new IP pool​

We add a new IPPool resource with the CIDR range 10.0.0.0/16.

Add the following to your default installation, below the existing IP pool.

kubectl edit installation default
- name: new-ipv4-pool
  cidr: 10.0.0.0/16
  encapsulation: IPIP

Let’s verify the new IP pool.

kubectl get ippools
NAME                  CREATED AT
default-ipv4-ippool   2024-03-28T16:14:28Z
test-pool             2024-03-28T18:30:15Z

Step 2: Disable the old IP pool​

Edit the default installation, and modify default-ipv4-ippool so it no longer selects any nodes. This prevents IP allocation from the pool.

kubectl edit installation default
- name: 192.168.0.0-16
  allowedUses:
  - Workload
  - Tunnel
  blockSize: 26
  cidr: 192.168.0.0/16
  disableBGPExport: false
  encapsulation: VXLANCrossSubnet
  natOutgoing: Enabled
- nodeSelector: all()
+ nodeSelector: "!all()"

Apply the changes.

Remember, disabling a pool only affects new IP allocations; networking for existing pods is not affected.

Step 3: Delete pods from the old IP pool​

Next, we delete all of the existing pods from the old IP pool. (In our example, coredns is our only pod; for multiple pods you would trigger a deletion for all pods in the cluster.)

kubectl delete pod -n kube-system coredns-6f4fd4bdf-8q7zp

Step 4: Verify that new pods get an address from the new IP pool​

  1. Create a test namespace and nginx pod.

    kubectl create ns ippool-test

  2. Create an nginx pod.

    kubectl -n ippool-test create deployment nginx --image nginx

  3. Verify that the new pod gets an IP address from the new range.

    kubectl -n ippool-test get pods -l app=nginx -o wide

  4. Clean up the ippool-test namespace.

    kubectl delete ns ippool-test

Step 5: Delete the old IP pool​

Now that you've verified that pods are getting IPs from the new range, you can safely delete the old pool. To do this, remove it from the default installation, leaving only the newly create IP pool.

Additional resources​