container-linux-update-operator

Container Linux Update Operator

Container Linux Update Operator is a node reboot controller for Kubernetes on Container Linux Distro. When a reboot is needed after updating the system via update_engine, the operator will drain the node before rebooting it.

Container Linux Update Operator fulfills the same purpose as locksmith, but has better integration with Kubernetes by explicitly marking a node as unschedulable and deleting pods on the node before rebooting.

Design

Original proposal

Container Linux Update Operator is divided into two parts - update-operator and update-agent.

update-agent runs on each node, waiting for a UPDATE_STATUS_UPDATED_NEED_REBOOT signal via dbus from update_engine. It will indicate via node annotations that it needs a reboot.

update-operator will watch changes to node annotations, and reboot the nodes as needed. It coordinates the reboots of multiple nodes in the cluster, ensuring that not too many are rebooting at once.

Currently, update-operator only reboots one node at a time.

Requirements

• Working Kubernetes >= 1.4 on CoreOS
• update-engine.service should be unmasked, enabled and started in systemd
• locksmithd.service should be masked and stopped in systemd

Usage

To start update-operator and update-agent:

# Open examples/components.yaml and edit the image tag.
kubectl create -f examples/components.yaml

To test that it is working, you can simulate that a reboot is needed by sshing to the node and running locksmithctl send-need-reboot.