Skip to content

Comparison

Kubernetes vs Docker: do you need Kubernetes?

Docker runs containers on one machine; Kubernetes schedules them across many. Three small apps: $163/mo on EKS, $60 on DigitalOcean, $7 on your own server.

Nick Potts, who builds Ployz. Published , facts checked ·Sources

The short answer

Do you need Kubernetes, or is Docker enough?

Docker runs containers on one machine; Kubernetes decides which of many machines each container runs on and keeps it running there. You need Kubernetes when you have many services, several teams, autoscaling or automatic failover to pay for. For a few apps on a few servers you don't: three small apps cost about $163 a month on Amazon EKS, $60 on DigitalOcean Kubernetes and about $7 on one Hetzner server with Ployz.

Choose Kubernetes if

For many services and teams
  • You run dozens of services with more than one team.
  • You need autoscaling, or a dead machine must fail over without a human.
  • The bill matters: managed Kubernetes on DigitalOcean or GKE can have a free control plane.
  • Your employer or customer requires it.

Choose Docker if

For a few apps on plain servers
  • You run a few apps on one to four servers and one person deploys.
  • A restart policy, a reverse proxy and a private network cover what you need.
  • You're fine deciding by hand which app runs where.

Choose Your own servers + Ployz if

Several servers, no Kubernetes
  • You want one app across several plain servers, without Kubernetes.
  • You want replicas and HTTPS on every server.
  • Your coding agent deploys to them.
Deploy your app

Learn Docker first. Kubernetes runs the images you build with Docker.

What is the difference between Docker and Kubernetes?

Kubernetes, Docker, Your own servers + Ployz compared, as checked on 1 October 2026
TopicKubernetesDockerYour own servers + Ployz
What it doesAn orchestrator that schedules containers across a cluster of machines. You tell it "run three copies of this image", and its control plane picks machines and starts the containers.Builds images and runs containers. It packages an app and everything it needs into an image, then runs that image as a container on one machine.Runs one app across several of your servers, with replicas and HTTPS on every server, without Kubernetes.
Runs onA control plane plus worker nodes.One machine.Linux servers with systemd, joined in an encrypted WireGuard network.
Unit you deployA pod, wrapped in a Deployment, Service and usually an Ingress.A container, or a Compose file of several.A service. Push to GitHub and Ployz builds on your servers and deploys, with no container registry.
ImagesRuns the images you build with Docker, unchanged, through containerd or CRI-O.Builds them from a Dockerfile.No Dockerfile needed for Node, Python, Ruby, PHP, Go, Rust, Java and Elixir.
A crashed container or a dead machineRestarts containers when they crash and moves them when a machine dies.Restarts crashed containers, with a restart policy. Doesn't move work off a dead machine.If a server dies, replicas on other servers keep serving.
ScalingAutoscaling, with the right add-ons.No autoscaling.You pick the replica count: up to 50 per service, spread one per server first.
Designed forUp to 5,000 nodes and 150,000 pods per cluster.Your laptop to a few servers.A few apps across a few servers.
Minimum machines1 to try it; 3 control plane nodes for high availability.1 machine.1 server. You add another with one command.
HTTPS and load balancingAn ingress controller, cert-manager for HTTPS and a load balancer you pay for.A reverse proxy like Caddy in front, which you set up.Every server serves https itself through Caddy, so there's no load balancer to rent.
UpkeepA cluster to upgrade roughly three times a year, and add-ons that each have their own version to track.No cluster. You decide by hand which app runs where.No Kubernetes and no control plane to look after.
Logs and debuggingYou read pod events, node conditions and controller logs.You read one container's output.Metrics, searchable logs and crash emails for every service and server, in place of the metrics-server, logging and monitoring add-ons a cluster needs.
Three small apps, a monthAbout $163 on Amazon EKS and $60 on DigitalOcean Kubernetes. Kubernetes adds a fixed floor on top of the servers.The servers you rent. The plain-server bill grows only with servers.About $7 on one Hetzner server. Ployz is free on any number of servers, custom domains included.

Docker and Kubernetes aren't competitors, and they're usually used together, as the explainers from Microsoft Azure and AWS both say. Docker is the box, and Kubernetes is the warehouse system that decides where the boxes go.

Does Kubernetes still use Docker?

Not as its runtime, but your images still work.

Kubernetes

Kubernetes v1.24 removed dockershim, the piece that let it talk to the Docker engine, so clusters now run containers through containerd or CRI-O. The Kubernetes dockershim FAQ says it directly: "The Kubernetes v1.24 release actually removed the dockershim from Kubernetes."

Docker

An image you build with Docker runs unchanged on Kubernetes. So "Kubernetes dropped Docker" is true of the runtime, not of the image format, and nothing you've built with a Dockerfile goes to waste.

When is Kubernetes worth it?

Kubernetes earns its cost when you have many services, several teams deploying them, traffic that swings enough to need autoscaling, or a requirement to survive a machine or zone going down without a human.

Kubernetes

With dozens of services and more than one team, everyone gets the same deploy model, the same health checks and the same access control. That consistency is worth a lot once ten people touch production. At that size the control plane fee is a rounding error. Some compliance and platform teams standardise on Kubernetes, and then the question is answered for you.

Docker

A fixed set of servers doesn't add or remove nodes for you. With many services and several teams, the alternative to Kubernetes is building the same scheduling and failover yourself, which is worse.

When is Kubernetes not worth it?

When you run a few apps on one or two servers. Kubernetes is built for thousands of machines, and most solo developers run between one and four.

Kubernetes

You pay a control plane fee, a minimum node count and a load balancer for scheduling decisions you could make by hand in a minute: this app goes on this server. You also take on a cluster to upgrade roughly three times a year, and for a solo developer that time is the bigger cost.

Docker

My own setup is the counterexample. I was paying Railway $500+ a month, most of it for a busy database. Today one dedicated server at €80 a month runs more than I ever ran there, with no cluster at all. One machine has a ceiling, and for most solo projects that ceiling is far away.

The work behind a cluster

Kubernetes

New Kubernetes minor versions come out "on average once every four months". Upgrading the control plane doesn't upgrade your nodes: on EKS, managed node groups "aren't automatically upgraded when you or Amazon EKS update your control plane", which is a second job each time. An ingress controller, cert-manager for HTTPS, a CSI driver for volumes and metrics-server for autoscaling each have their own version to track. Running Kubernetes yourself on cheap servers removes the control plane fee but adds to this work.

Docker

A single web app is one docker run or one Compose file, where Kubernetes needs at least a Deployment, a Service and an Ingress.

What do you lose without Kubernetes?

Mostly the automatic parts. Be honest about which of them you need today, not in a year.

Kubernetes

When a node dies, Kubernetes reschedules its pods on the remaining nodes, which matters if an outage costs you real money per minute. If your traffic goes from 2 machines' worth to 20 several times a day, Kubernetes with a cluster autoscaler adds and removes nodes for you. For multi-region or multi-zone setups, regional clusters spread the control plane and nodes across zones.

Docker

Nothing moves your containers when a server dies, and nothing adds servers when traffic spikes. You also lose the shared vocabulary a team gets from Kubernetes. You keep everything Docker does well: the same images, restart policies and health checks.

Your own servers + Ployz

If losing a server for a while would cost you customers, plan a second server and replicas from the start, or pay for a managed cluster. With replicas on more than one server, the others keep serving when one dies.

How do you run a few apps across a few servers without Kubernetes?

Docker

Put Docker on each server, a reverse proxy like Caddy in front for HTTPS, and a private network between the servers so services can reach each other. Then decide by hand which app runs where. With two or three servers that decision rarely changes, so the hard parts are deploys without downtime and adding a server later. Docker itself also ships Compose for several containers on one machine and Swarm mode for a small cluster.

Your own servers + Ployz

Ployz is built for exactly this shape: a few apps across a few servers. Servers join an encrypted WireGuard network, and services reach each other at NAME.internal. Deploys are zero-downtime when you set a health check: a new replica must pass it before an old one stops.

What would your apps cost on Kubernetes vs plain servers?

Set how many apps you run, how much memory each needs, and how many servers you want at minimum. The calculator works out how many 4 GB machines that takes, then prices the same machines three ways.

Memory your apps need
1.5 GB
4 GB servers needed (3 GB usable each)
1
Kubernetes nodes (2 minimum)
2
  • Amazon EKS$163/mo

    $73 control plane, t3.medium nodes ($30.37) with a public IPv4 ($3.65), one shared Application Load Balancer ($22.27). EBS disks not counted.

  • DigitalOcean Kubernetes$60/mo

    Free control plane, 4 GB nodes at $24, one $12 load balancer. High-availability control plane is $40 more.

  • Your own server + Ployz$7/mo

    Hetzner CX23 ($6.49 + about $0.60 IPv4, Europe) per server. Ployz is free on any number of servers. No load balancer to rent.

Prices checked 2026-10-01. Move the sliders to match your app.

The "3 GB usable" figure is my assumption, not a measurement, for a 4 GB machine once the operating system and system containers take their share. I used it for all three rows so the comparison stays fair. Kubernetes nodes also run their own agents, so if anything the real Kubernetes node count is a bit higher.

With the same assumptions (a two-node minimum on Kubernetes, one shared load balancer, AWS at 730 hours a month), here's how the bill grows:

SetupAmazon EKSDigitalOcean KubernetesHetzner + Ployz
3 apps of 512 MB, 1 server (2 nodes on Kubernetes)$163$60$7
10 apps, 3 servers$197$84$21
30 apps, 8 servers$367$204$57

The Hetzner column is CX23 servers (2 vCPU, 4 GB RAM, 40 GB) at $6.49 plus about $0.60 for the IPv4 address, from Hetzner's 15 June 2026 price adjustment. Ployz is free on any number of servers, so the column is the server bill.

How the EKS line for three apps adds up: $73 control plane, plus two t3.medium nodes at $30.37 each, plus $3.65 per node for the public IPv4, plus $16.43 for the ALB's hours and $5.84 for one LCU. EBS disks, data transfer and NAT gateways are extra, so the real bill is higher.

The CX23 is Europe only (Germany and Finland). For a US datacenter, Netcup's VPS 500 (2 vCPU, 4 GB RAM, 64 GB SSD) is €8.26 a month incl. VAT on a 12-month term and has a Manassas location (netcup VPS plans); a $24 DigitalOcean Droplet also works.

Two things stand out. DigitalOcean is the fair managed option for a small team, because its free control plane leaves the gap as mostly the second node and the load balancer. And EKS's $73 control plane alone costs more than ten Hetzner servers. If you're on AWS already and want containers without the cluster fee, AWS Fargate is the other route to price.

What does managed Kubernetes cost?

EKS charges $0.10 per cluster per hour, about $73 a month, before any servers. GKE charges the same rate but gives a monthly credit that covers one cluster. DigitalOcean's control plane is free. On all three you pay for the worker machines on top.

ProviderControl planeNodesLoad balancerSource
Amazon EKS$0.10/cluster/hour (about $73/month); $0.60/hour on extended supportEC2 prices, e.g. t3.medium (2 vCPU, 4 GiB) at $0.0416/hour, about $30.37/monthALB: $0.0225/hour plus $0.008 per LCU-hourEKS pricing, EC2 on-demand, ELB pricing
Google GKE$0.10/cluster/hour; $74.40/month free tier credit per billing account covers one zonal or Autopilot clusterCompute Engine pricesBilled separatelyGKE pricing
DigitalOcean KubernetesFree; high-availability control plane $40/monthFrom $12/month (1 vCPU, 2 GB); 4 GB nodes are $24$12/monthDOKS pricing, Droplet pricing

The lines people miss:

  • Public IPv4 on AWS is $0.005 an hour per address, in use or idle (VPC pricing). That's $3.65 a month for each node with a public address.
  • EKS extended support. Each Kubernetes version gets 14 months of standard support on EKS, then 12 months of extended support, which is on by default (EKS versions). It raises the control plane to $0.60 an hour, about $438 a month for one cluster if you fall behind on upgrades.
  • GKE's free tier covers one cluster. Google's own example bills every cluster hour past the credit at the full rate (GKE pricing). Regional clusters, the highly available kind, aren't covered by the credit at all.
  • DigitalOcean recommends two nodes. Its pricing page says "We recommend using a minimum of two nodes to prevent downtime during upgrades or maintenance." So the real floor is two nodes plus a load balancer, $36 a month with the cheapest node.

GKE isn't in the calculator, because I couldn't confirm the per-hour Compute Engine rate on Google's page. With the free tier its control plane can cost nothing for one cluster, so its bill comes down to nodes and load balancers, like DigitalOcean's.

Moving a few apps to your own servers

You can wire up Docker, the proxy and the private network yourself, or use a self-hosted platform. Coolify and Dokku are the popular ones, and PaaS pricing compares them with hosted platforms on cost. I built Ployz because I didn't want a three-server minimum to get off one server.

ployz up --server root@your-server

That installs Ployz and deploys the current directory. Add a second server and it joins the same private network. Every server serves https itself. There's no load balancer to rent.

Claude Code, Codex or Cursor runs it through ployz mcp or the CLI with --json. Every deploy, restart or delete your agent asks for waits for you to approve.

Run your apps on your own servers with Ployz, free on any number of servers.

Questions

Should I learn Docker or Kubernetes first?

Docker first, because Kubernetes runs container images and you need to know how to build and run one before an orchestrator makes sense. Writing a Dockerfile, running it locally and deploying it to one server teaches you most of what you'll use day to day. Learn Kubernetes when a job or a project actually needs it.

Is Kubernetes free?

The software is free, and DigitalOcean's control plane is free, but a cluster isn't. You pay for at least two nodes and a load balancer, $36 a month at DigitalOcean's cheapest node. EKS adds about $73 a month for the control plane alone.

Can Docker replace Kubernetes?

For a few apps on one to four servers, yes. Docker with a restart policy, a reverse proxy and a private network covers what most small projects need. It won't reschedule containers off a dead machine or autoscale, which is what Kubernetes is for.

What's the cheapest way to run Kubernetes?

DigitalOcean Kubernetes, or GKE's free tier for one zonal or Autopilot cluster. Both can have a free control plane, so you pay for nodes and load balancers. Running it yourself with kubeadm saves the fee but needs three control plane machines for high availability. The kubeadm docs ask for "2 GB or more of RAM per machine" and 2 CPUs for control plane machines.

Sources

Every statement about Kubernetes and Docker comes from the pages below, checked on 1 October 2026. They can change after that date.

Run your apps and databases on your own servers with Ployz.

Deploy your app