minikube

minikube supports several container runtimes to satisfy Kubernetes requirements. When starting a cluster, you can choose from the following supported runtimes:

• containerd
• cri-o
• docker

Each runtime has specific configuration requirements and behaviors. Refer to the specific documentation for the runtime you intend to use to ensure proper setup.

minikube provides specific commands to enable and access standard Kubernetes networking and management features:

• LoadBalancer access: Use minikube tunnel to provide access to services of type LoadBalancer.
• NodePort access: Use minikube service to access services of type NodePort.
• Dashboard: Use minikube dashboard to launch the web-based Kubernetes user interface.
• Multi-cluster support: Create and manage multiple independent clusters using the -p (profile) flag: minikube start -p <name>.
• Container Runtimes: Specify a runtime during startup using the --container-runtime flag: minikube start --container-runtime <runtime>.

# Example: Start a specific cluster profile
minikube start -p my-cluster

# Example: Access a LoadBalancer service
minikube tunnel

# Example: Open the dashboard
minikube dashboard

Benchmarks are categorized into two modes to simulate different real-world container loading scenarios:

Initial

Simulates the very first time an image is loaded. To replicate a clean environment, all existing images and caches are cleared from both minikube and Docker between every run.

Iterative

Simulates a development workflow (similar to how Skaffold operates) where only the application code changes. In this mode, only the final layer (the Go binary) is updated, while the existing image and Docker cache are preserved between runs to measure the impact of incremental changes.

Once the registry-aliases addon is configured, you can use the following domains as container registry URLs in your deployment manifests (e.g., skaffold.yaml or Kubernetes YAML files). These domains will resolve to the internal minikube registry:

• example.com
• example.org
• test.com
• test.org

To test this locally with Docker, ensure your shell is using the minikube docker daemon context:

eval $(minikube -p demo docker-env)

The none driver performs destructive operations on certain system paths:

Overwritten on start:

• /etc/kubernetes (configuration files)

Erased on minikube delete:

• /data/minikube
• /etc/kubernetes/manifests
• /var/lib/minikube

Because Kubernetes has full access to your filesystem and docker images, unexpected data loss may occur.

To help distinguish between new regressions and existing flaky tests, minikube automatically comments on Pull Requests (PRs) when integration tests fail.

When integration tests complete, a script named report_flakes.sh processes the gopogh summaries and the public gs://minikube-flake-rate/flake_rates.csv dataset. The resulting PR comment includes:

• A list of all failed tests.
• The recorded flake rate for each failed test.
• Direct links to the flake charts for that specific test and environment.

This allows developers and reviewers to quickly determine if a failure is a known unreliable test (a flake) or a potential new bug introduced by the PR.

minikube is designed to simplify running Kubernetes locally for day-to-day development workflows and learning. Its primary focus is providing a user-friendly, cross-platform, and reliable experience with minimal resource overhead and third-party dependencies.

Key features that support these goals include:

• Single command setup and teardown UX.
• Support for local storage, networking, auto-scaling, and load balancing.
• A unified UX across different operating systems.

Note on Non-Goals: minikube is not intended to simplify Kubernetes production deployment experiences, nor does it aim to support every possible Kubernetes deployment configuration (such as all variations of storage or networking types).

When running mkcmp, you can specify the binaries to compare using two methods:

1. Direct Path: Provide the absolute or relative path to a specific minikube binary file.
2. PR Reference: Use the syntax pr://<PR number> to automatically use the minikube binary built during that specific Pull Request.