node-drainer

Utilise the power of AWS Auto Scaling group (ASG) lifecycle hooks and drain your Kubernetes nodes gracefully. node-drainer reads AWS ASG lifecycle hook messages from an SQS queue. For each item received it contacts Kubernetes, taints the node to be shut down and evicts any pods not tolerant to the taint. Meant to be run in side Kubernetes with a single replica only.

Development Status node-drainer was designed for internal use only. After migrating to AWS EKS we started to use the Node Termination Handler, which is the official tool from AWS. Development for node-drainer was therefore stopped, but feel free to fork and continuing its legacy!

Use cases

node-drainer is useful whenever any of the Kubernetes worker nodes running in AWS must be shut down. Graceful eviction of Kubernetes pods from terminated nodes ensures continuous operation of services when:

• Performing a rolling Kubernetes cluster update
• Changing AWS EC2 instance types for worker nodes
• Updating the AWS EC2 instance image on worker nodes
• Scaling down the number of workers periodically when the cluster load is low

Usage

All of node-drainer's configuration is done using command line arguments, with the intention to be defined inside a Kubernetes deployment yaml file.

For a full list of parameters run:

./node-drainer -h

node-drainer can be configured to run outside of Kubernetes too, for testing purposes or otherwise. Below are two configuration examples.

Running locally

When running locally we have to specify a valid kubeconfig file path as well as any AWS credentials needed. In the following example we are using a pre-configured AWS profile.

node-drainer --kubeconfig /example/kubeconfig/path --profile example_aws_profile --region example_region --queue-name example_queue_name

Running in Kubernetes

When running inside a Kubernetes cluster in a pod, the Kubernetes configuration information is picked up automatically. We still have to configure AWS access as usual.

node-drainer --access-key-id example_id --secret-access-key example_secret --region example_region --queue-name example_queue_name

Installation

• Binaries for node-drainer are provided for each release here.
• Docker containers are are provided here. To obtain the latest docker image run docker pull quay.io/rebuy/node-drainer:main.
• For deploying node-drainer docker image to your Kubernetes cluster you can use the sample manifest files (found here), just remember to fill in your own AWS credentials. I you use RBAC in Kubernetes you can also take advantage of the sample service account configuration.

To compile node-drainer from source you need a working Golang development environment. The sources must be cloned to $GOPATH/src/github.com/rebuy-de/node-drainer.

Also you need to install godep, golint and GNU Make.

Then you just need to run make build to compile a binary into the project directory or make install to install node-drainer into $GOPATH/bin. With make xc you can cross compile node-drainer for other platforms.

Contact channels

Feel free to create a GitHub Issue for any questions, bug reports or feature requests.

How to contribute

You can contribute to node-drainer by forking this repository, making your changes and creating a Pull Request against our repository. If you are unsure how to solve a problem or have other questions about a contributions, please create a GitHub issue.