Kata Containers, a second layer of isolation is created on top of those provided by traditional namespace-containers. The hardware virtualization interface is the basis of this additional layer. Kata will launch a lightweight virtual machine, and use the guest’s Linux kernel to create a container workload, or workloads in the case of multi-container pods. In Kubernetes and in the Kata implementation, the sandbox is carried out at the pod level. In Kata, this sandbox is created using a virtual machine.
This document describes how Kata Containers maps container technologies to virtual machines technologies, and how this is realized in the multiple hypervisors and virtual machine monitors that Kata supports.
A typical deployment of Kata Containers will be in Kubernetes by way of a Container Runtime Interface (CRI) implementation. On every node, Kubelet will interact with a CRI implementor (such as containerd or CRI-O), which will in turn interface with Kata Containers (an OCI based runtime).
The CRI API, as defined at the Kubernetes CRI-API repo, implies a few constructs being supported by the CRI implementation, and ultimately in Kata Containers. In order to support the full API with the CRI-implementor, Kata must provide the following constructs:
These constructs can then be further mapped to what devices are necessary for interfacing with the virtual machine:
Ultimately, these concepts map to specific para-virtualized devices or virtualization technologies.
Each hypervisor or VMM varies on how or if it handles each of these.
Kata Containers is designed to support multiple virtual machine monitors (VMMs) and hypervisors. Kata Containers supports:
Which configuration to use will depend on the end user's requirements. Details of each solution and a summary are provided below.
Kata Containers with QEMU has complete compatibility with Kubernetes.
Depending on the host architecture, Kata Containers supports various machine types,
for example pc and q35 on x86 systems, virt on ARM systems and pseries on IBM Power systems. The default Kata Containers
machine type is pc. The machine type and its Machine accelerators can
be changed by editing the runtime configuration file.
Devices and features used:
- virtio VSOCK or virtio serial
- virtio block or virtio SCSI
- virtio net
- virtio fs or virtio 9p (recommend: virtio fs)
- VFIO
- hotplug
- machine accelerators
Machine accelerators and hotplug are used in Kata Containers to manage resource constraints, improve boot time and reduce memory footprint. These are documented below.
Machine accelerators are architecture specific and can be used to improve the performance and enable specific features of the machine types. The following machine accelerators are used in Kata Containers:
- NVDIMM: This machine accelerator is x86 specific and only supported by
pcandq35machine types.nvdimmis used to provide the root filesystem as a persistent memory device to the Virtual Machine.
The Kata Containers VM starts with a minimum amount of resources, allowing for faster boot time and a reduction in memory footprint. As the container launch progresses, devices are hotplugged to the VM. For example, when a CPU constraint is specified which includes additional CPUs, they can be hot added. Kata Containers has support for hot-adding the following devices:
- Virtio block
- Virtio SCSI
- VFIO
- CPU
Firecracker, built on many rust crates that are within rust-VMM, has a very limited device model, providing a lighter footprint and attack surface, focusing on function-as-a-service like use cases. As a result, Kata Containers with Firecracker VMM supports a subset of the CRI API. Firecracker does not support file-system sharing, and as a result only block-based storage drivers are supported. Firecracker does not support device hotplug nor does it support VFIO. As a result, Kata Containers with Firecracker VMM does not support updating container resources after boot, nor does it support device passthrough.
Devices used:
- virtio VSOCK
- virtio block
- virtio net
Cloud Hypervisor, based on rust-VMM, is designed to have a lighter footprint and attack surface. For Kata Containers, relative to Firecracker, the Cloud Hypervisor configuration provides better compatibility at the expense of exposing additional devices: file system sharing and direct device assignment. As of the 1.10 release of Kata Containers, Cloud Hypervisor does not support device hotplug, and as a result does not support updating container resources after boot, or utilizing block based volumes. While Cloud Hypervisor does support VFIO, Kata is still adding this support. As of 1.10, Kata does not support block based volumes or direct device assignment. See Cloud Hypervisor device support documentation for more details on Cloud Hypervisor.
Devices used:
- virtio VSOCK
- virtio block
- virtio net
- virtio fs
| Solution | release introduced | brief summary |
|---|---|---|
| QEMU | 1.0 | upstream QEMU, with support for hotplug and filesystem sharing |
| NEMU | 1.4 | Deprecated, removed as of 1.10 release. Slimmed down fork of QEMU, with experimental support of virtio-fs |
| Firecracker | 1.5 | upstream Firecracker, rust-VMM based, no VFIO, no FS sharing, no memory/CPU hotplug |
| QEMU-virtio-fs | 1.7 | upstream QEMU with support for virtio-fs. Will be removed once virtio-fs lands in upstream QEMU |
| Cloud Hypervisor | 1.10 | rust-VMM based, includes VFIO and FS sharing through virtio-fs, no hotplug |