KubeZero - Zero Down Time Kubernetes platform

KubeZero is a Kubernetes distribution providing an integrated container platform so you can focus on your applications.

Design philosophy

• Focus on security and simplicity over feature creep
• No vendor lock in, most components are optional and could be easily changed as needed
• No premium services / subscriptions required
• Staying up to date and contributing back to upstream projects, like alpine-cloud-images and others
• Cloud provider agnostic, bare-metal/self-hosted
• Organic Open Source / open and permissive licenses over closed-source solutions
• Corgi approved 🐶

Architecture

Version / Support Matrix

KubeZero releases track the same minor version of Kubernetes. Any 1.30.X-Y release of Kubezero supports any Kubernetes cluster 1.30.X.

KubeZero is distributed as a collection of versioned Helm charts, allowing custom upgrade schedules and module versions as needed.

%%{init: {'theme':'dark'}}%%
gantt
    title KubeZero Support Timeline
    dateFormat  YYYY-MM-DD
    section 1.28
    beta     :128b, 2024-03-01, 2024-04-30
    release  :after 128b, 2024-08-31
    section 1.29
    beta     :129b, 2024-07-01, 2024-07-31
    release  :after 129b, 2024-11-30
    section 1.30
    beta     :130b, 2024-09-01, 2024-10-31
    release  :after 130b, 2025-02-28

Upstream release policy

Components

OS

• all compute nodes are running on Alpine V3.20
• 1 or 2 GB encrypted root file system
• no external dependencies at boot time, apart from container registries
• minimal attack surface
• extremely small memory footprint / overhead
• cri-o container runtime incl. AppArmor support

GitOps

• cli / cmd line install
• optional full ArgoCD support and integration

Featured workloads

• rootless CI/CD build platform to build containers as part of a CI pipeline, using podman / fuse device plugin support
• containerized AI models via integrated out of the box support for Nvidia GPU workers as well as AWS Neuron

Control plane

• all Kubernetes components compiled against Alpine OS using buildmode=pie
• support for single node control plane for small clusters / test environments to reduce costs
• access to control plane from within the VPC only by default ( VPN access required for Admin tasks )
• controller nodes are used for various platform admin controllers / operators to reduce costs and noise on worker nodes

AWS integrations

• IAM roles for service accounts allowing each pod to assume individual IAM roles
• access to meta-data services is blocked all workload containers on all nodes
• all IAM roles are maintained via CloudBender automation
• aws-node-termination handler integrated
• support for spot instances per worker group incl. early draining etc.
• support for Inf1 instances part of AWS Neuron.

Network

• Cilium using Geneve encapsulation, incl. increased MTU allowing flexible / more containers per worker node compared to eg. AWS VPC CNI
• Multus support for multiple network interfaces per pod, eg. additional AWS CNI
• no restrictions on IP space / sizing from the underlying VPC architecture

Storage

• flexible EBS support incl. zone awareness
• EFS support via automated EFS provisioning for worker groups via CloudBender automation
• local storage provider (OpenEBS LVM) for latency sensitive high performance workloads
• CSI Snapshot controller and Gemini snapshot groups and retention

Ingress

• AWS Network Loadbalancer and Istio Ingress controllers
• no additional costs per exposed service
• real client source IP available to workloads via HTTP header and access logs
• ACME SSL Certificate handling via cert-manager incl. renewal etc.
• support for TCP services
• optional rate limiting support
• optional full service mesh

Metrics

• Prometheus support for all components, incl. out of cluster EC2 instances (node_exporter)
• automated service discovery allowing instant access to common workload metrics
• pre-configured Grafana dashboards and alerts
• Alertmanager events via SNSAlertHub to Slack, Google, Matrix, etc.

Logging

• all container logs are enhanced with Kubernetes and AWS metadata to provide context for each message
• flexible ElasticSearch setup, leveraging the ECK operator, for easy maintenance & minimal admin knowledge required, incl. automated backups to S3
• Kibana allowing easy search and dashboards for all logs, incl. pre configured index templates and index management
• fluentd-concerter service providing queuing during highload as well as additional parsing options
• lightweight fluent-bit agents on each node requiring minimal resources forwarding logs secure via TLS to fluentd-concenter