mck8s: Container orchestrator for multi-cluster Kubernetes

mck8s, short for multi-cluster Kubernetes, allows you to automate the deployment of multi-cluster applications on multiple Kubernetes clusters by offering enhanced configuration possibilities. The main aim of mck8s is maximizing resource utilization and supporting elasitcity across multiple Kubenetes clusters by providing multiple placement policies, as well as bursting, cloud resource provisioning, autoscaling and de-provisioning capabilities. mck8s builds upon other open-source software such as Kubernetes, Kubernetes Federation, kopf, serf, Cilium, Cluster API, and Prometheus.

This code is in active development and not stable, and thus not production ready.

All contributions are welcome :)

Architecture

The figure below shows the architecture of mck8s.

Quick start

Pre-requisites

• A Kubernetes cluster to act as the mck8s management cluster (named cluster0). Since mck8s components need access to the Kubernetes control plane, managed Kubernetes offerings such as GKE are not supported at the moment.
• Few Kubernetes clusters to be managed by the management cluster and on which workloads run. We call these workload clusters (named cluster1, cluster2, cluster3, ...). We assume that you have administrative access to all these clusters and the Kubernetes kubeconfig files of all clusters are available.
• If traffic routing between clusters is desired, it is recommended to deploy Cilium Cluster Mesh on the workload clusters with distinct Pod CIDRs.
• If proximity-aware placement is desired, serf should be deployed on at least one node of each workload cluster.
• If cloud provisioning and autoscaling ia desired, cloud credentials are required. For now, we support OpenStack clusters.

Steps

Prepare

1. Clone this repository to your computer.
2. Copy the kubeconfig file of the management cluster as cluster0 in ~/.kube/ directory of your computer .
3. Copy the kubeconfig files from the workload clusters to ~/.kube/ directory of your computer and rename these files as per their cluster names such as cluster1, cluster2, or cluster3, .....
4. Make sure that the cluster, context, and user names is these files are distinct from each other.
5. Before running prepare.sh, which assumes that there are five workload clusters, cluster1 to cluster5, edit the values.yaml file as per the number of your workload clusters and enter the IP addresses of the master nodes of your clusters.
6. Run the prepare.sh script, which sets up Kubernetes Federation, Prometheus Operator, and Cilium. This script assumes that there are five workload clusters, cluster1 to cluster5. If you have a different number of workload clusters, please adjust the script accordingly.
7. Copy the ~/.kube/config file to the ~/.kube/ directory of the master node of the management cluster.

Deploy the CRDs on the management cluster

1. Switch to the cluster0 context to be able to deploy the CRDs on the management cluster.

   kubectl config use-context cluster0

2. Create the RBAC required

   kubectl apply -f manifests/crds/01_rbac_mck8s.yaml

3. Deploy the CRDs

   kubectl apply -f manifests/crds/

Buid Docker images of the controllers

1. Buid the image of the Multi Cluster Scheduler

   cd multi-cluster-scheduler && docker build -t REPO_NAME/IMAGE_NAME .

2. Build the image of the Multi Cluster HPA

   cd multi-cluster-horizontal-pod-autoscaler && docker build -t REPO_NAME/IMAGE_NAME .

3. Build the image of the Cloud Provisioner and Cluster Autoscaler

   cd cloud-cluster-provisioner-autoscaler && docker build -t REPO_NAME/IMAGE_NAME .

4. Build the image of the Deployment Rescheduler

   cd multi-cluster-rescheduler && docker build -t REPO_NAME/IMAGE_NAME .

Deploy the controller Deployments on the management cluster

NOTE: The pods of these controllers are going to be scheduled on the master node of the management cluster. In the manifest files below, Replace the MASTER_NODE_HOST_NAME, REPO_NAME/IMAGE_NAME, and /PATH/TO/HOME/DIRECTORY/ accordingly before running the following.

1. Deploy the Multi Cluster Scheduler Deployment

   kubectl apply -f manifests/controllers/01_deployment_multi_cluster_scheduler.yaml

2. Deploy the Multi Cluster HPA Deployment

   kubectl apply -f manifests/controllers/02_deployment_multi_cluster_hpa.yaml

3. Deploy the Cloud Provisioner and Cluster Autoscaler Deployment

   kubectl apply -f manifests/controllers/03_deployment_cloud_provisioner_cluster_autoscaler.yaml

Demo: Deployment of mck8s CRDs, controllers, multicluster scheduler, and example multicluster deployment

Let's see a demo of how to deploy the multiclusterscheduler, a sample multiclusterdeployment and a multiclusterservice.

Demo: Inter-cluster traffic routing

In this demo, we show how to access a back-end application deployed on cluster3 via a front-end application deployed on cluster2.

Related Publications

[1] Mulugeta Tamiru, Guillaume Pierre, Johan Tordsson, Erik Elmroth. mck8s: An orchestration platform for geo-distributed multi-cluster environments. ICCCN 2021 - 30th International Conference on Computer Communications and Networks, Jul 2021, Athens, Greece. pp.1-12. (pdf)

Talks

[1] I gave a virtual talk titled "McK8s: Container Orchestration in Kubernetes Multi-Clusters" at KubeCon + CloudNativeCon NA 2021 https://www.youtube.com/watch?v=U1iHBZhEWUA&t=824s