A small application to watch nodes and to shut them down if certain criteria is met.
I wrote this tool because I have a couple of nodes in my private Kubernetes cluster, that are dedicated to on/off workloads (E.g.: Plex Media Server). In order to save some energy costs, I want to shut them down if this kind of workload is unused and therefore the node isn't utilized. Because running proper cluster autoscalers like Karpenter isn't possible, I came up with this solution.
The following environment variables can be used to control the behaviour:
| Key | Description | Default |
|---|---|---|
| MEMORY_THRESHOLD | Threshold in megabytes that will cause a shutdown | None |
| CPU_THRESHOLD | Threshold in mili CPU that will cause a shutdown | None |
| SHUTDOWN_DELAY_MINUTES | Delay in minutes before performing the shutdown | None |
| SSH_USERNAME | Username used to establish the SSH connection | None |
| SSH_PRIVATE_KEY_PATH | Path to the private key that is used for the SSH authentication | None |
| SSH_PORT | Port used to establish the SSH connection | 22 |
| NODE_NAMES | Comma seperated list of nodes to watch | None |
| DRY_RUN_MODE | Watch resource usage of the nodes, but do not initiate a shutdown | false |
| TZ | Set a timezone that is used for logging | UTC |
A Helm Chart is available and will receive regular updates.
kubernetes-node-shutdown is currently maintained by tibuntu.
Anyone that wants to contribute is highly welcome!
GNU General Public License v3.0