The Single-Node Kubernetes Showdown: minikube vs. kind vs. k3d
2024-03-21 15:01:42

As a developer in the cloud-native ecosystem, a common challenge is the need to frequently test applications within a Kubernetes environment. In CI, this often extends to quickly trialing configurations across various Kubernetes clusters, including single-node, high-availability, dual-stack, multi-cluster setups, and more. Thus, the ability to swiftly create and manage Kubernetes clusters on a local machine, without breaking the bank, has become a must-have. This post will dive into three popular single-node Kubernetes management tools: minikube, kind, and k3d.Highlighting their unique features, use cases, and potential pitfalls.


If speed is your only concern, k3d is your best bet. If you’re after compatibility and a simulation close to reality, minikube is your safest bet. kind sits comfortably in the middle, offering a balance between the two.

Technical Comparison

At their core, these three tools serve a similar function: managing Kubernetes on a single machine. However, their differing historical backgrounds and technical choices have led to unique nuances and use cases.


minikube is the Kubernetes community’s OG tool for quickly setting up Kubernetes locally, a first love for many Kubernetes novices. Initially, it simulated multi-node clusters via VMs on your local machine, offering a high-fidelity emulation of real-world scenarios, down to the OS and kernel module level. The downside? It’s a resource hog, and if your virtualization environment doesn’t support nested virtualization, you’re out of luck, not to mention it’s slow to start. Recently, the community introduced a Docker Driver to mitigate these issues, though at the cost of losing some VM-level emulation capabilities. On the bright side, minikube comes with a plethora of add-ons, like dashboards and nginx-ingress, for easy community component installation.


kind is a more recent favorite for local Kubernetes deployment, using Docker containers to simulate nodes and focusing purely on Kubernetes standard deployments, with community components requiring manual installation. It’s the go-to for Kubernetes’ own CI processes. The upside? Quick starts and a familiar environment for Docker veterans. The downside? Container simulation lacks OS-level isolation, sharing the host’s kernel, which can complicate OS-specific testing. I once had a kernel module test fail because the host’s netfilter tweaks caused havoc in a kind-managed cluster.


k3d, a featherweight in local Kubernetes deployment, shares a similar approach to kind but opts for deploying a lightweight k3s instead of standard Kubernetes. This means it inherits k3s’s pros and cons, boasting incredibly fast setup times—don’t worry about correctness; just marvel at the speed. The trade-offs include a super-slimmed-down OS (sans glibc), complicating certain OS-level operations, and a unique installation approach that might puzzle those accustomed to kubeadm’s standard deployment features.

Performance Showdown

While the minikube community provides some performance benchmarks, comparing the startup times of our three contenders, I was curious about other aspects like image size, memory footprint, and bare-minimum setup times, prompting another round of tests.


Testing was straightforward, given each tool’s one-liner setup, with a few caveats:

  1. minikube used the Docker Driver to keep the speed test fair.
  2. All tests assumed pre-downloaded images to exclude network delays.
  3. The latest versions were tested, though Kubernetes versions varied, making this more of a qualitative than quantitative analysis.
  4. Tests focused on basic component starts without additional plugins, ensuring essentials like CNI, CoreDNS, and CSI were included.
  5. docker image and docker stat commands were used to measure image sizes and memory usage, respectively.

Commands used:

time minikube start --driver=docker --force

time kind create cluster

time k3d cluster create mycluster --k3s-arg '--disable=traefik,metrics-server

@server:*' --no-lb


Name Software Version Kubernetes Version Image Size Start Time Memory Usage
minikube v1.32.0 v1.28.3 1.2GB 29s 536MiB
kind v0.22 v1.29.2 956MB 20s 463MiB
k3d v5.6.0 v1.27.4 263MB 7s 423MiB

Evidently, k3d takes the crown in startup performance, boasting significant advantages in image size, startup time, and memory usage. It’s a godsend for those running CI on a shoestring budget.


If speed and resource efficiency are your top priorities, k3d is a no-brainer. For OS-level isolation tests, minikube’s VM Driver is unbeatable. For everything in between, kind offers a balanced compromise between compatibility and performance.

2024-03-21 15:01:42