Kubernetes or Docker on a VPS: when it’s time for your team to move to a cluster

It’s fashionable to deploy Kubernetes “for the future”. But for three containers on one server it only adds work. Let’s see where the line is.

Docker Compose on a VPS: when it’s enough

One server, a docker-compose.yml file, a few containers — website, database, queue. Deployment is git pull and docker compose up. This covers most projects of small teams: corporate websites, online stores, internal services, bots. The only cost is the VPS, for example 4 vCPU / 8 GB / 100 GB with an IP address at 18 000 UZS per day.

Limitations: during updates the service is unavailable for a few seconds; if the server goes down, everything goes down; you can only scale vertically, by adding resources to a single server.

Kubernetes: what changes

• Zero-downtime updates. The new version starts next to the old one, and traffic switches over only after health checks pass.
• Self-healing. A crashed container is restarted, a failed node is replaced, and the load is redistributed.
• Scaling. More requests mean more replicas of the service; fewer at night. Nodes are added without migration.
• An environment for every branch. CI spins up a test environment for each pull request and removes it after the merge.

The price tag

A managed cluster at MyCloud: 20 000 UZS per day for “Dev” management or 60 000 for “Prod” with a fault-tolerant control plane, plus nodes at VPS prices. A three-node cluster of 4 vCPU / 8 GB / 100 GB comes to 20 000 + 3 × 17 000 = 71 000 UZS per day (cluster nodes don’t need a separate IP address). Compared with a single VPS at 18 000, that’s almost four times more.

When the switch pays off

1. You have more than 5–7 services, and they’re updated several times a week.
2. A minute of downtime during an update already costs money or reputation.
3. Load varies several-fold over a day or a season.
4. Your team has 3+ developers, and each needs their own test environments.

If at least two of these apply to you, the cluster will save more than it costs. If not, stay on Docker Compose and spend the difference on backups and monitoring.

A compromise

Start with a “Dev” cluster of one or two nodes: you get Kubernetes-style deployments and environments at a price close to two VPS. Moving to fault-tolerant “Prod” doesn’t require recreating the cluster.