/vm

Package and Run Virtual Machines as Docker Containers

Primary LanguageGoApache License 2.0Apache-2.0

RancherVM

Package and run KVM images as Kubernetes pods, run at scale.

How It Works

RancherVM allows you to create VMs that run inside of Kubernetes pods, called VM Pods. A VM pod looks and feels like a regular pod. Inside of each VM pod, however, is a container running a virtual machine instance. You can package any QEMU/KVM image as a Docker image, distribute it using any Docker registry such as DockerHub, and run it on RancherVM.

RancherVM extends the Kubernetes API with Custom Resource Definitions, or CRDs. Users define a VirtualMachine CRD specification detailing what base image, how much compute resources and what keypairs are authorized to open an SSH session. A Kubernetes controller creates VM pods as necessary to achieve the desired specification and reflects this in the VirtualMachine CRD status.

RancherVM comes with a Web UI for managing public keys, compute nodes, virtual machines and accessing the VNC console from a web browser.

How it works

Deployment

Prerequisite

  1. Kubernetes v1.8+
  2. KVM has been installed on all nodes.
    1. Follow the distribution-specific instructions to ensure KVM works. We only require KVM to be enabled in the kernel. We do not need any user space tools like qemu-kvm or libvirt. On Ubuntu 16.04, you can make sure KVM is enabled by checking that both devices /dev/kvm and /dev/net/tun exist.
    2. An easy way to run KVM on your Windows or Mac laptop is to use nested virtualization with VMware Workstation or VMware Fusion. Just enable "Virtualize Intel VT-x/EPT or AMD-V/RVI" in VM settings. Remember to use Ubuntu 18.04 or CentOS with latest kernel in this case, otherwise you will hit bug that VM hanging from Booting from harddisk...
  3. All nodes must be in the same layer 2 network, with an existing DHCP server.
  4. Host bridge network has been prepared for RancherVM. See here for details.
    1. Identify your NIC name (e.g. ens33) on each host, and replace the value here
    2. Identify your bridge name (e.g. br0) on each host, then replace the value here and the line below contains br0.

Once you have Kubernetes and KVM both setup, and correctly setup the nic names, you can deploy the system:

kubectl create -f deploy/ranchervm.yaml

When you see all pods are ready and running as follows, you've deployed RancherVM successfully. Single-node Kubernetes clusters are expected to run fewer pods.

$ kubectl -n ranchervm-system get pods
NAME                             READY     STATUS    RESTARTS   AGE
backend-5f5dd7878-5p6rm          1/1       Running   0          1h
frontend-5b5d47c669-cnlwn        1/1       Running   0          1h
ip-controller-648cdf6854-gkvxj   2/2       Running   0          1h
vm-controller-7c5fdbb68d-d8qq8   1/1       Running   0          1h

RancherVM is ready to use. To access the UI, discover the endpoint as following.

$ kubectl -n ranchervm-system get svc/frontend
NAME       TYPE           CLUSTER-IP    EXTERNAL-IP       PORT(S)        AGE
frontend   LoadBalancer   10.43.55.38   100.200.200.123   80:31520/TCP   42m

If the Kubernetes Cluster supports creating LoadBalancer services, access RancherVM UI using the EXTERNAL-IP (100.200.200.123 in the case above) on port 80. Otherwise, access RancherVM UI using <node_ip>: (port is 31520 in the case above), where node_ip is the public IP address of any node in the Kubernetes cluster.

Access the UI for most of RancherVM functionalities.

Build VM Images

You can find instructions on how to build images, including Windows images, in the RancherVM Images document.

Networking

The details of how RancherVM configures network for the VM Pod is documented in RancherVM Networking.

Build from Source

To build a Docker image, run IMAGE=yes hack/build.sh.

To only build the binary for your local OS & ARCH, run hack/build.sh.