kured (KUbernetes REboot Daemon)

kured is a Kubernetes DaemonSet designed to perform safe, automatic node reboots. It triggers reboots when the underlying operating system's package management system indicates a reboot is required.

Core Functionality:

• Sentinel Monitoring: Watches for reboot sentinel files (e.g., /var/run/reboot-required) or the successful execution of a sentinel command.
• Concurrency Control: Uses a lock in the Kubernetes API server to ensure that only one node reboots at a time, preventing cluster-wide downtime.
• Node Lifecycle Management: Automatically cordons and drains worker nodes before a reboot and uncordons them once the node is back online.
• Reboot Deferral: Can be configured to defer reboots if active Prometheus alerts are present or if specific selected pods are running.

kured supports configuring all command line flags via environment variables. The environment variable name is constructed by taking the KURED prefix, followed by the flag name in uppercase, with hyphens replaced by underscores.

Example mapping:

• --node-id $\rightarrow$ KURED_NODE_ID
• --reboot-command $\rightarrow$ KURED_REBOOT_COMMAND
• --drain-timeout $\rightarrow$ KURED_DRAIN_TIMEOUT

kured uses the following canonical annotations to manage node state and maintenance visibility:

weave.works/kured-node-lock
weave.works/kured-reboot-in-progress
weave.works/kured-most-recent-reboot-needed