kubernetes-cluster-manager

README

kubernetes-cluster-manager (kcm)

IMPORTANT NOTE: This project has been discontinued in favor of kubectl-chart.

Inspired by Zalando's CLM. The Kubernetes Cluster Manager project was started because CLM is tightly coupled to AWS Cloudformation for managing the cluster infrastructure. KCM tries to provide an interface for using different infrastructure provisioners. It also tries to provide visibility about changes by providing diffs for things like manifest changes.

Use with caution. This project is currently alpha quality and APIs are likely to change until the first stable release.

Features:

• Make output of infrastructure provisioners available to manifest renderer
• Show diffs of changes in infrastructure output values and manifests
• Render manifests via helm charts
• Minikube integration for local testing
• Dry run, apply and destroy changes (infrastructure + kubernetes manifests)

Currently supported infrastructure provisioners:

• null (default)
• terraform
• minikube for local testing

Design

The current design assumes that cluster configuration is managed in a git repository. kcm will write changes to yaml files which then should be commited by the CI tool of your choice. kcm can manage multiple clusters simply by having separate directories in the repository for the configuration of each cluster. The git approach is a trade-off to avoid introducing a persistence layer for kcm in the early development phase. However, there might be other options in the future. Check out the Roadmap.

Installation

$ git clone https://github.com/martinohmann/kubernetes-cluster-manager
$ cd kubernetes-cluster-manager
$ make install

This will install the kcm binary to $GOPATH/bin/kcm.

Usage

The documentation is still work in progress. For now, refer to godoc, the examples _examples directory and the command line help:

$ kcm help

Quick examples

Provision infrastructure using terraform and render manifests via helm

$ kcm provision \
  --provisioner terraform \
  --working-dir /path/to/terraform/repo \
  --templates-dir /path/to/helm/charts \
  --dry-run

As the bare minimum kcm expects the infrastructure provisioner to create a kubernetes cluster and to either return the path to a generated kubeconfig in its output, or server and token values needed for establishing a connection to the kubernetes api-server. Alternatively you can manually provide kubernetes credentials via the --cluster-* flags. Detailed examples will follow.

Using a config file and skipping manifest rendering/deployment

$ kcm provision --config config.yaml --skip-manifests

Working with manifests

The kcm manifests command will only render and apply/delete manifests and will skip any infrastructure changes:

$ kcm manifests apply \
  --templates-dir templates/ \
  --manifests-dir manifest/ \
  --cluster-kubeconfig ~/.kube/config \
  --cluster-context eks-dev

Apply all manifests, even if unchanged:

$ kcm manifests apply --config config.yaml --all-manifests

Delete manifests:

$ kcm manifests delete --config config.yaml

Destroying a cluster

$ kcm destroy --dry-run

Roadmap

The following features are currently planned, but I'm also happy about other contributions. PRs welcome!

• Add node pool manager (e.g. for managing spotinst elastigroups)
• Triggering of rolling updates of node pools
• Support for more provisioners (e.g. Cloudformation, kubeadm)
• Replace shell-execs with native go-libraries where possible (and sensible)
• Add support for other persistence layers and configuration sources besides the git approach mentioned in the design section

License

The source code of kubernetes-cluster-manager is released under the MIT License. See the bundled LICENSE file for details.