Skip to content
Go to file


Failed to load latest commit information.


Build Status Go Report Card

The hypervisor-based container runtime for Kubernetes

Frakti lets Kubernetes run pods and containers directly inside hypervisors via runV. It is light weighted and portable, but can provide much stronger isolation with independent kernel than linux-namespace-based container runtimes.

Frakti serves as a CRI container runtime server. Its endpoint should be configured while starting kubelet. In the deployment, hyperd is also required as the API wrapper of runV.


Build frakti:

mkdir -p $GOPATH/src/
git clone $GOPATH/src/
cd $GOPATH/src/
make && make install

Install docker and hyperd:

apt-get install qemu libvirt0 -y
curl -sSL | bash

Configure hyperd with gRPC endpoint

cat >/etc/hyper/config <<EOF
# Boot kernel
# Boot initrd
# Storage driver for hyperd, valid value includes devicemapper, overlay, and aufs
# Hypervisor to run containers and pods, valid values are: libvirt, qemu, kvm, xen
# The tcp endpoint of gRPC API

systemctl restart hyperd

Setup CNI networking using bridge plugin

$ sudo mkdir -p /etc/cni/net.d  /opt/cni/bin
$ git clone $GOPATH/src/
$ cd $GOPATH/src/
$ ./
$ sudo cp bin/* /opt/cni/bin/
$ sudo sh -c 'cat >/etc/cni/net.d/10-mynet.conflist <<-EOF
    "cniVersion": "0.3.1",
    "name": "mynet",
    "plugins": [
            "type": "bridge",
            "bridge": "cni0",
            "isGateway": true,
            "ipMasq": true,
            "ipam": {
                "type": "host-local",
                "subnet": "",
                "routes": [
                    { "dst": ""   }
            "type": "portmap",
            "capabilities": {"portMappings": true},
            "snat": true
$ sudo sh -c 'cat >/etc/cni/net.d/99-loopback.conf <<-EOF
    "cniVersion": "0.3.1",
    "type": "loopback"

Then start frakti:

frakti --v=3 --logtostderr --listen=/var/run/frakti.sock --hyper-endpoint= &

Finally, start kubernetes with frakti runtime:

cd $GOPATH/src/
export PATH=$GOPATH/src/${PATH}
export CONTAINER_RUNTIME_ENDPOINT=/var/run/frakti.sock

To start using the cluster, open up another terminal and run:

cd $GOPATH/src/
export KUBECONFIG=/var/run/kubernetes/admin.kubeconfig


Further information could be found at:

The differences between frakti with other Linux container runtimes

  • Better Security and Isolation
    • frakti provides hardware virtualization based Pod sandbox for Kubernetes.
  • No Kernel Sharing
    • Every Pod in frakti has its own kernel (Bring Your Own Kernel), LinuxKit image support is on the way
  • Match k8s QoS Classes.
    • frakti is best to run Pod with resources.limits being set (i.e. all Guaranteed and most Burstable Pods), otherwise, frakti will set default resource limit for Pod.
    • This behavior is configurable by --defaultCPUNum and --defaultMemoryMB of frakti
  • Mixed Runtimes Mode.
    • frakti support mixed runtimes on the same Node (runV and Docker). We recommend user to run BestEffort Pods, daemon Pods in Docker runtime by adding annotation to them.
    • Additionally, special cases like privileged Pods, host network Pods etc will be automatically run in Docker runtime.
  • Persistent Volume
    • All k8s PVs are supported in frakti.
    • Block device as volume directly: in this way, frakti will mount block device directly to VM based Pod, which gives you better performance than first approach, now we support:
      • Cinder RBD: Stackube for an example (k8s + frakti + Cinder + Keystone + Neutron).
      • GCE Persistent Disk
  • Cross-host Networking
    • frakti is fully based on CNI (bridge mode only for now), so there's no big difference here.
    • Network plug-ins like Flannel & Calico can work out-of-box with frakti based Kubernetes.

Besides the lists above, all behaviors of frakti are 100% the same with other Linux container runtimes like Docker, please enjoy it!


The work done has been licensed under Apache License 2.0.The license file can be found here. You can find out more about license at

You can’t perform that action at this time.