A second service and a first outage do not mean you need Kubernetes. Stay on Docker Compose on one VPS until you need independent scaling, rolling deploys across hosts, or a team that already runs it. Managed DOKS control plane is free; HA is $40/month and on by default for API creates on 1.36+.
This does not apply when you already run a multi-node cluster with dedicated platform staff, or when a customer contract already requires a managed Kubernetes control plane.
By Sudipta Dhara. Drafted 2026-10-19. Last updated 2026-10-19. AI disclosure: drafted with AI assistance. No affiliate links in this draft.
Who this is for: tech leads, PMs, and DevOps engineers on small teams (about 1–15 engineers) choosing a platform · Intermediate · You’ll need: one app you can describe as services + data stores, and the provider pricing pages open the day you decide.
Methodology. Docs-plus-our-own-practice. We re-read DigitalOcean Kubernetes pricing, supported releases, and changelog, plus Contabo and Droplet list prices, on 19 Oct 2026 IST. We did not run a timed Compose-versus-k3s lab for this refresh, so this page prints no setup hours, no recovery timings, and no invented cluster invoice totals. Practice note: ProDevs ships hub and CI workloads on Compose / plain VPS shapes rather than a production Kubernetes fleet — see self-hosted GitHub runners cost. Version pins for context: Docker Compose v5.6.0 (2026-10-02), k3s v1.37.1+k3s1 (2026-09-30).
Stay on Compose on one VPS until multi-service scaling, multi-host rolling deploys, or existing Kubernetes skills are real. Price DOKS as separate lines — especially the $40/month HA control plane the create API turns on by default on 1.36+.
Shareable fact
| Rule | Detail | Caveat |
|---|---|---|
| DOKS HA by default | On 1.36.0+, omitting ha enables HA ($40/month, prorated hourly) |
Checked 19 Oct 2026 IST (changelog, pricing) |
| Two-node floor | DigitalOcean recommends ≥2 nodes so upgrades do not take the app down | One node is cheaper and ignores that line |
| k3s server floor on DO | Cheapest compliant Basic Droplet is $18/month (2 GiB / 2 vCPU) | The $12 plan is 2 GiB / 1 vCPU — below k3s’s 2-core minimum |
Sibling guides: Docker or a VPS, VPS sizing for AI coding agents, Docker memory limits.
Who it’s for
A developer, DevOps engineer, or tech lead deciding whether the next app needs managed Kubernetes, a single VPS with Compose or k3s, or a one-app platform. Bangalore (BLR1) and India tax notes appear where the docs print them. This is not an India-only tax guide.
Not for
- Teams already running Kubernetes at scale with dedicated platform staff.
- A single static site — use a static host or App Platform’s free static sites.
- A one-week prototype where a laptop or one VPS is enough.
- An EKS version-upgrade runbook (link only; no upgrade steps here).
Alternatives
One app on App Platform, or on a single VPS. App Platform includes 3 free static-site apps (1 GiB outbound each). A paid app starts at $5/month (1 vCPU, 512 MiB, 50 GiB transfer) — checked 19 Oct 2026 IST on App Platform pricing. Use it when you want the platform to own one process. A single VPS is the same idea when you want the shell.
Compose on one VPS. Pin Compose (v5.6.0 in this draft) and keep a few containers on one machine you patch. If you are still choosing Docker versus renting the box, start with Docker or a VPS.
Managed Kubernetes. DOKS when you already have more than one service, want the provider to run the control plane, and can spare a second node. Amazon EKS is the same shape on AWS — read rates on the EKS pricing page; this article does not print them.
What does Kubernetes give you that Docker Compose doesn’t?
Kubernetes schedules across nodes, restarts failed pods, and supports rolling deploys and declarative scaling. You pay with a control plane, node pools, and a version calendar.
Compose keeps containers on one host. Restart policies help, but one machine remains the failure domain. k3s sits in the middle: Kubernetes APIs on a box you still patch, with no multi-node resilience on one VPS.
When does a small team actually need Kubernetes?
You need Kubernetes when… several services must scale independently; you need zero-downtime deploys across more than one host; someone already operates Kubernetes day to day; a second server is arriving and deploys are already breaking; or a customer path assumes a managed control plane.
Not yet when… one app, one server, no on-call; nobody reads upgrade notes; the pain is “we need a process manager,” not “we need a fleet.”
What does each option cost?
Price lines separately. Do not invent a cluster total. Figures re-checked 19 Oct 2026 IST unless noted.
Managed Kubernetes (DOKS)
| Line | Published price (checked 2026-10-19) |
|---|---|
| Standard control plane | Free |
| High-availability control plane | $40/month, prorated hourly |
| Basic worker node | From $12/month per node |
| Outbound transfer | 2,000 GiB per node, pooled, then $0.01/GiB |
| Load balancer | From $12/month |
| Block storage | From $10/month |
| Container registry | Free up to 500 MiB |
Sources: Kubernetes pricing, DOKS pricing docs.
Worker nodes bill at Droplet rates per second (60-second or $0.01 minimum). Bundled plans cap at 672 hours/month. v5 nodes have no monthly cap and are not in BLR1 (regional availability).
DigitalOcean recommends at least two nodes so upgrades do not cause downtime.
Cost trap: on DOKS 1.36.0+, the create API enables HA by default when ha is unset. Set ha to false for a test without that line. (changelog)
Invoices are USD only. India GST is 18%. A GSTIN removes GST; reverse charge still applies. No INR list. (taxes, India)
VPS column
k3s server minimum: 2 cores / 2 GB RAM (agent: 1 core / 512 MB). Keep TCP 6443 for the API; Flannel VXLAN UDP 8472 should not be public. (k3s requirements)
On DigitalOcean, the cheapest Basic Droplet that meets that server minimum is 2 GiB / 2 vCPU at $18/month. The $12 plan is 2 GiB / 1 vCPU — too small. (Droplet pricing)
Contabo Cloud VPS 4 (4 vCPU / 8 GB / 100 GB SSD) showed schema lowPrice USD 6.6 on 19 Oct 2026 IST. Quote the currency and term you will pay; location surcharges are configurator lines. (Contabo VPS)
Example compose.yaml on one host (replace image tags with yours), then docker compose up -d:
services:
web:
image: nginx:1.27-alpine
ports:
- "8080:80"
worker:
image: busybox:1.36
command: ["sh", "-c", "while true; do echo tick; sleep 30; done"]
k3s pin for context. Disposable host only. The server needs at least 2 cores and 2 GB RAM.
curl -sfL https://get.k3s.io | INSTALL_K3S_VERSION="v1.37.1+k3s1" sh -s - \
--write-kubeconfig-mode 644
sudo k3s kubectl get nodes
Which DOKS version should a new cluster use?
Re-checked 19 Oct 2026 IST on supported releases (page last verified 8 Oct 2026):
| Version | End of support | Upgrades begin |
|---|---|---|
| 1.34 | 27 October 2026 | 26 November 2026 |
| 1.35 | 28 February 2027 | 28 March 2027 |
| 1.36 | 28 June 2027 | 28 July 2027 |
| 1.37 | 28 October 2027 | 27 November 2027 |
Do not start a new cluster on 1.34 (EOS 27 Oct 2026). Prefer 1.35, 1.36, or 1.37. Newest 1.36.x patch seen that day: 1.36.5-do.1 (14 Oct 2026) on the changelog. Pin k3s separately (v1.37.1+k3s1 here); do not write “latest.”
BLR1 has DOKS and bundled nodes; not v5 or GPU nodes.
Is k3s on one VPS a sensible middle step?
Yes when you need kubectl and Deployments but not a managed control plane and a second node. You still own patching, datastore disk (SSD recommended), and the firewall.
Practising shell and kubectl before you touch a cluster? Talk2Term is our tool for that. It does not run your cluster. No usage numbers claimed here.
Managed or self-managed: who looks after it?
Managed moves the control plane off your pager. You still size nodes and upgrade versions — DOKS force-upgrades after end of support. Current worker images are Debian 13-class with containerd and Cilium; build elsewhere and pull from a registry. (changelog)
Self-managed k3s means you patch the server. Compose is smaller still.
Not sure which stage you are at? ProDevs does DevOps and CI setup for small teams — contact or services.
Cons
- HA by default on DOKS 1.36+ adds $40/month unless you set
ha: false. - Second node is the vendor’s own upgrade-safety recommendation.
- USD + 18% GST (GSTIN removes GST; reverse charge still applies).
- v5 not in BLR1; bundled nodes keep the 672-hour cap.
- No timed Compose-vs-k3s hours in this draft — we cut those claims until a measured lab exists.
So what should my team run?
| Stage | Run |
|---|---|
| One app | Compose on a VPS, or App Platform |
| Several services, one team | Compose or single-node k3s + backups/monitoring |
| Multi-service scaling / HA / K8s skills | Managed Kubernetes (≥2 nodes; record HA choice) |
RAM and plan size: Docker memory limits, VPS sizing. CI on a plain box: self-hosted runners cost.
What broke when we almost priced a “cheap” cluster?
We almost treated “control plane free” as “cluster free,” and almost sized k3s from the $12 Droplet row. The 19 Oct 2026 IST docs re-check fixed both: HA is a separate $40/month line on by default for API creates on 1.36+, and the k3s server minimum needs the $18 2-vCPU plan. Cutting invented hour claims kept the page honest when the timed lab was not available.
Raw readings
- Changelog also lists 1.36.5-do.0 (8 Oct 2026); this draft cites 1.36.5-do.1 (14 Oct 2026).
- Contabo Cloud VPS 4 EUR schema
lowPrice5.5 on the same day (USD 6.6 is the shop figure here). - Sept 2026 pins (Compose v5.5.1, k3s v1.37.0+k3s1, DOKS 1.36.3-do.5, “upstream 1.37 not on DOKS”) are superseded — DOKS now lists 1.37.
- EKS per-cluster-hour rates stay linked, not quoted.
- Contabo India location surcharge dollars were not confirmed as a stable public string on this re-check; use Contabo’s configurator.
FAQ
Do I need Kubernetes for a small team or startup?
Usually not at first. Stay on Docker Compose on one VPS until you need independent scaling, rolling deploys across hosts, or a team that already runs Kubernetes. Managed DOKS control plane is free; HA is $40/month and on by default for API creates on 1.36+.
Kubernetes vs Docker Compose: what’s the difference in practice?
Compose runs containers on one host with a short YAML file you patch yourself. Kubernetes schedules across nodes, restarts failed pods, and supports rolling deploys — at the cost of a control plane, node pools, and an upgrade calendar.
When should I move from a VPS to Kubernetes?
Move when several services must scale independently, you need zero-downtime deploys across more than one host, or the team already operates Kubernetes day to day. One app on one server is still a Compose or App Platform job.
Is k3s on one VPS a good middle step?
Yes if you need the Kubernetes API on one box. The k3s server minimum is 2 cores and 2 GB RAM; on DigitalOcean that is the $18 Basic Droplet (2 GiB / 2 vCPU), not the $12 plan. One node still has no multi-node resilience.
How much does managed Kubernetes cost on DigitalOcean?
Price lines separately. Standard control plane is free; HA is $40/month prorated hourly and on by default for API creates on 1.36+. Basic nodes start at $12/month each; DigitalOcean recommends at least two. Load balancer from $12/month; block storage from $10/month.
Managed or self-managed Kubernetes for a small team?
Managed moves the control plane off your pager; you still upgrade the cluster version and pay for nodes. Self-managed k3s means you patch the server, disk, and firewall. Pick managed when nobody wants that pager.
Next step
Open the DOKS pricing page and the Droplet (or Contabo) page side by side today. Write three lines: (1) Compose / k3s / DOKS, (2) whether HA is on, (3) one node or two. Then stay on Compose or create the cluster with those choices recorded.
ProDevs does DevOps and CI setup for small teams — contact us with your stack and timeline.
