Skip to content

Latest commit

 

History

History
165 lines (105 loc) · 6.76 KB

Limitations.md

File metadata and controls

165 lines (105 loc) · 6.76 KB
Raw

Overview

A Kata Container utilizes a Virtual Machine (VM) to enhance security and isolation of container workloads. As a result, the system has a number of differences and limitations when compared with the default Docker* runtime, runc.

Some of these limitations have potential solutions, whereas others exist due to fundamental architectural differences generally related to the use of VMs.

The Kata Container runtime launches each container within its own hardware isolated VM, and each VM has its own kernel. Due to this higher degree of isolation, certain container capabilities cannot be supported or are implicitly enabled through the VM.

Definition of a limitation

The Open Container Initiative Runtime Specification ("OCI spec") defines the minimum specifications a runtime must support to interoperate with container managers such as Docker. If a runtime does not support some aspect of the OCI spec, it is by definition a limitation.

However, the OCI runtime reference implementation (runc) does not perfectly align with the OCI spec itself.

Further, since the default OCI runtime used by Docker is runc, Docker expects runtimes to behave as runc does. This implies that another form of limitation arises if the behavior of a runtime implementation does not align with that of runc. Having two standards complicates the challenge of supporting a Docker environment since a runtime must support the official OCI spec and the non-standard extensions provided by runc.

Scope

Each known limitation is captured in a separate GitHub issue that contains detailed information about the issue. These issues are tagged with the limitation label. This document is a curated summary of important known limitations and provides links to the relevant GitHub issues.

The following link shows the latest list of limitations:

Contributing

If you would like to work on resolving a limitation, please refer to the contributors guide. If you wish to raise an issue for a new limitation, either raise an issue directly on the runtime or see the project table of contents for advice on which repository to raise the issue against.

Pending items

This section lists items that might be possible to fix.

OCI CLI commands

Docker and Podman support

Currently Kata Containers does not support Docker or Podman.

See issue kata-containers#722 for more information.

Runtime commands

checkpoint and restore

The runtime does not provide checkpoint and restore commands. There are discussions about using VM save and restore to give us a [criu](https://github.com/checkpoint-restore/criu)-like functionality, which might provide a solution.

Note that the OCI standard does not specify checkpoint and restore commands.

See issue kata-containers/runtime#184 for more information.

events command

The runtime does not fully implement the events command. OOM notifications and Intel RDT stats are not fully supported.

Note that the OCI standard does not specify an events command.

See issue kata-containers/runtime#308 and kata-containers/runtime#309 for more information.

update command

Currently, only block I/O weight is not supported. All other configurations are supported and are working properly.

Networking

Resource management

Due to the way VMs differ in their CPU and memory allocation, and sharing across the host system, the implementation of an equivalent method for these commands is potentially challenging.

See issue clearcontainers/runtime#341 and the constraints challenge for more information.

For CPUs resource management see CPU constraints.

Architectural limitations

This section lists items that might not be fixed due to fundamental architectural differences between "soft containers" (i.e. traditional Linux* containers) and those based on VMs.

Storage limitations

Kubernetes volumeMounts.subPaths

Kubernetes volumeMount.subPath is not supported by Kata Containers at the moment.

See this issue for more details. Another issue focuses on the case of emptyDir.

Host resource sharing

Privileged containers

Privileged support in Kata is essentially different from runc containers. The container runs with elevated capabilities within the guest and is granted access to guest devices instead of the host devices. This is also true with using securityContext privileged=true with Kubernetes.

The container may also be granted full access to a subset of host devices (kata-containers/runtime#1568).

See Privileged Kata Containers for how to configure some of this behavior.

Appendices

The constraints challenge

Applying resource constraints such as cgroup, CPU, memory, and storage to a workload is not always straightforward with a VM based system. A Kata Container runs in an isolated environment inside a virtual machine. This, coupled with the architecture of Kata Containers, offers many more possibilities than are available to traditional Linux containers due to the various layers and contexts.

In some cases it might be necessary to apply the constraints to multiple levels. In other cases, the hardware isolated VM provides equivalent functionality to the the requested constraint.

The following examples outline some of the various areas constraints can be applied:

  • Inside the VM

    Constrain the guest kernel. This can be achieved by passing particular values through the kernel command line used to boot the guest kernel. Alternatively, sysctl values can be applied at early boot.

  • Inside the container

    Constrain the container created inside the VM.

  • Outside the VM:

    • Constrain the hypervisor process by applying host-level constraints.

    • Constrain all processes running inside the hypervisor.

      This can be achieved by specifying particular hypervisor configuration options.

Note that in some circumstances it might be necessary to apply particular constraints to more than one of the previous areas to achieve the desired level of isolation and resource control.