system-upgrade-controller

System Upgrade Controller

Introduction

This project aims to provide a general-purpose, Kubernetes-native upgrade controller (for nodes). It introduces a new CRD, the Plan, for defining any and all of your upgrade policies/requirements. A Plan is an outstanding intent to mutate nodes in your cluster. For up-to-date details on defining a plan please review v1/types.go.

Presentations and Recordings

April 14, 2020

CNCF Member Webinar: Declarative Host Upgrades From Within Kubernetes

• Slides
• Video

March 4, 2020

Rancher Online Meetup: Automating K3s Cluster Upgrades

• Video

Considerations

Purporting to support general-purpose node upgrades (essentially, arbitrary mutations) this controller attempts minimal imposition of opinion. Our design constraints, such as they are:

• content delivery via container image a.k.a. container command pattern
• operator-overridable command(s)
• a very privileged job/pod/container:
  • host IPC, NET, and PID
  • CAP_SYS_BOOT
  • host root file-system mounted at /host (read/write)
• optional opt-in/opt-out via node labels
• optional cordon/drain a la kubectl

Additionally, one should take care when defining upgrades by ensuring that such are idempotent–there be dragons.

Deploying

The most up-to-date manifest is usually manifests/system-upgrade-controller.yaml but since release v0.4.0 a manifest specific to the release has been created and uploaded to the release artifacts page. See releases/download/v0.4.0/system-upgrade-controller.yaml

But in the time-honored tradition of curl ${script} | sudo sh - here is a nice one-liner:

# Y.O.L.O.
kubectl apply -k github.com/rancher/system-upgrade-controller

API Documentation

Autogenerated API docs for upgrade.cattle.io/v1 Plan are available at doc/plan.md

Example Plans

• examples/k3s-upgrade.yaml
  • Demonstrates upgrading k3s itself.
• examples/ubuntu/bionic.yaml
  • Demonstrates upgrading, apt-get style, arbitrary packages at pinned versions.
• examples/ubuntu/bionic/linux-kernel-aws.yaml
  • Demonstrates upgrading the kernel on Ubuntu 18.04 EC2 instances on AWS.
• examples/ubuntu/bionic/linux-kernel-virtual-hwe-18.04.yaml
  • Demonstrates upgrading the kernel on Ubuntu 18.04 (to the HWE version) on generic virtual machines.

Below is an example Plan developed for k3OS that implements something like an rsync of content from the container image to the host, preceded by a remount if necessary, immediately followed by a reboot.

---
apiVersion: upgrade.cattle.io/v1
kind: Plan

metadata:
  # This `name` should be short but descriptive.
  name: k3os-latest

  # The same `namespace` as is used for the system-upgrade-controller Deployment.
  namespace: k3os-system

spec:
  # The maximum number of concurrent nodes to apply this update on.
  concurrency: 1

  # The value for `channel` is assumed to be a URL that returns HTTP 302 with the last path element of the value
  # returned in the Location header assumed to be an image tag (after munging "+" to "-").
  channel: https://github.com/rancher/k3os/releases/latest

  # Providing a value for `version` will prevent polling/resolution of the `channel` if specified.
  version: v0.10.0

  # Select which nodes this plan can be applied to.
  nodeSelector:
    matchExpressions:
      # This limits application of this upgrade only to nodes that have opted in by applying this label.
      # Additionally, a value of `disabled` for this label on a node will cause the controller to skip over the node.
      # NOTICE THAT THE NAME PORTION OF THIS LABEL MATCHES THE PLAN NAME. This is related to the fact that the
      # system-upgrade-controller will tag the node with this very label having the value of the applied plan.status.latestHash.
      - {key: plan.upgrade.cattle.io/k3os-latest, operator: Exists}
      # This label is set by k3OS, therefore a node without it should not apply this upgrade.
      - {key: k3os.io/mode, operator: Exists}
      # Additionally, do not attempt to upgrade nodes booted from "live" CDROM.
      - {key: k3os.io/mode, operator: NotIn, values: ["live"]}

  # The service account for the pod to use. As with normal pods, if not specified the `default` service account from the namespace will be assigned.
  # See https://kubernetes.io/docs/tasks/configure-pod-container/configure-service-account/
  serviceAccountName: k3os-upgrade

  # Specify which node taints should be tolerated by pods applying the upgrade.
  # Anything specified here is appended to the default of:
  # - {key: node.kubernetes.io/unschedulable, effect: NoSchedule, operator: Exists}
  tolerations:
  - {key: kubernetes.io/arch, effect: NoSchedule, operator: Equal, value: amd64}
  - {key: kubernetes.io/arch, effect: NoSchedule, operator: Equal, value: arm64}
  - {key: kubernetes.io/arch, effect: NoSchedule, operator: Equal, value: s390x}

  # The prepare init container, if specified, is run before cordon/drain which is run before the upgrade container.
  # Shares the same format as the `upgrade` container.
  prepare:
    # If not present, the tag portion of the image will be the value from `.status.latestVersion` a.k.a. the resolved version for this plan.
    image: alpine:3.18
    command: [sh, -c]
    args: ["echo '### ENV ###'; env | sort; echo '### RUN ###'; find /run/system-upgrade | sort"]

  # If left unspecified, no drain will be performed.
  # See:
  # - https://kubernetes.io/docs/tasks/administer-cluster/safely-drain-node/
  # - https://kubernetes.io/docs/reference/generated/kubectl/kubectl-commands#drain
  drain:
    # deleteLocalData: true  # default
    # ignoreDaemonSets: true # default
    force: true
    # Use `disableEviction == true` and/or `skipWaitForDeleteTimeout > 0` to prevent upgrades from hanging on small clusters.
    # disableEviction: false # default, only available with kubectl >= 1.18
    # skipWaitForDeleteTimeout: 0 # default, only available with kubectl >= 1.18

  # If `drain` is specified, the value for `cordon` is ignored.
  # If neither `drain` nor `cordon` are specified and the node is marked as `schedulable=false` it will not be marked as `schedulable=true` when the apply job completes.
  cordon: true

  upgrade:
    # If not present, the tag portion of the image will be the value from `.status.latestVersion` a.k.a. the resolved version for this plan.
    image: rancher/k3os
    command: [k3os, --debug]
    # It is safe to specify `--kernel` on overlay installations as the destination path will not exist and so the
    # upgrade of the kernel component will be skipped (with a warning in the log).
    args:
      - upgrade
      - --kernel
      - --rootfs
      - --remount
      - --sync
      - --reboot
      - --lock-file=/host/run/k3os/upgrade.lock
      - --source=/k3os/system
      - --destination=/host/k3os/system

Building

Go binaries are built with registry.rancher.com/rancher/appco-go in a multi-stage Dockerfile. The controller runtime image uses scratch to preserve support for all existing release platforms, including 32-bit ARM.

make

Running

Use ./bin/system-upgrade-controller.

Also see manifests/system-upgrade-controller.yaml that spells out what a “typical” deployment might look like with default environment variables that parameterize various operational aspects of the controller and the resources spawned by it.

Testing

Integration tests are bundled as a Sonobuoy plugin that expects to be run within a pod. To verify locally:

make e2e

This will, via Dapper, stand up a local cluster (using docker-compose) and then run the Sonobuoy plugin against/within it. The Sonobuoy results are parsed and a Status: passed results in a clean exit, whereas Status: failed exits non-zero.

Alternatively, if you have a working cluster and Sonobuoy installation, provided you’ve pushed the images (consider building with something like make REPO=dweomer TAG=dev), then you can run the e2e tests thusly:

sonobuoy run --plugin dist/artifacts/system-upgrade-controller-e2e-tests.yaml --wait
sonobuoy results $(sonobuoy retrieve)

Usage in Rancher

System Upgrade Controller (SUC) is packaged as a Helm chart and distributed with Rancher. The chart is maintained in the rancher/charts repository, with branches corresponding to Rancher release lines.

The table below shows the SUC chart and controller versions used by each supported Rancher release line.

Rancher Release Line	SUC Chart Version	SUC Version
v2.15	110.0.0	v0.20.1
v2.14	109.0.2	v0.19.1
v2.13	108.0.0	v0.17.0
v2.12	107.0.1	v0.16.3

Note: The chart and SUC versions shown above are examples and may change over time as Rancher patch releases update their dependencies.

License

Copyright © 2019-2026 Rancher Labs, Inc.

Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.