Docker Hosting Explained: Which of Four Kinds You Need

Docker is a way to package an app and run it the same everywhere. Docker hosting is whatever runs that package for you, and providers use the phrase for very different amounts of help. The difference is best described as who gets the phone call when it breaks.

No kind of docker hosting in the table is better than the rest. Each one moves work between you and someone else, and the price follows the work.

The simplest docker hosting is a server with an operating system and nothing else. You install Docker, decide how the app gets there, put a proxy in front for TLS and remember to patch the box. It is the cheapest per resource by a wide margin. Our Hetzner Cloud pricing guide shows what a small server really costs once the extras are added.

It is also less exotic than it sounds. Docker's own production guidance for Compose starts here.

The same page suggests a separate production Compose file, no bind mounts for application code, and a restart policy so a crash does not mean downtime. That gets you a long way on one machine.

Panel-based docker hosting works like this: Coolify, Dokploy and Dokku install on a VPS and add a dashboard, git-push builds and automatic certificates. You are still the person patching the server, but the deploy experience is close to a managed platform. Coolify's pricing page describes self-hosting as free with full access to all features, and sells an optional hosted control plane at $5 a month for two servers.

Two cautions. The panel is one more thing to keep updated and backed up, and if it fails you fall back to raw Docker commands. Second, check the licence before you build on one. Dokploy now keeps some future features under a source-available licence, which why open-core PaaS tools keep going commercial explains, and our Coolify review covers the free option in detail.

Managed docker hosting builds your image from a repository or pulls one you built, runs it, and looks after the server, TLS, restarts and rollbacks. Pricing takes two shapes. Some charge for resources used: Railway lists $20 per vCPU per month and $10 per GB of memory per month, on a $5 Hobby or $20 Pro subscription that includes the same amount in usage credit. Others, including Helm7, sell a minimum monthly spend with usage drawn against it.

On Helm7 that minimum starts at $12 a month for Builder, with Postgres, auth, storage and custom domains on the same bill. We run two EU regions today, Falkenstein and Helsinki, and have no US region yet. If you need one, another platform is the better choice. Our best Docker hosting providers puts the managed options side by side.

When you have many services that need to scale independently across many machines, and people whose job is to run the cluster. The project describes itself accurately.

That sentence is the whole warning. You get orchestration, and you build the platform around it: builds, TLS, secrets, logging, deploy flow. Our rule of thumb, which is an opinion and not a measurement, is that a team with one to five services is paying for machinery it does not use.

Worked example, so you can check the arithmetic. A Next.js app that needs 2 vCPU and 4 GB of memory, running all month. We price it at published rates, and where a provider bills by usage we show the worst case, which is full use around the clock. Currencies differ, so read each row, not the column.

For docker hosting on price alone the VPS wins by a wide margin, and that is the honest answer if you are happy to be its operator. The €6.49 leaves out your time, your monitoring and your restore drill. Railway's figure is a ceiling, since it bills by the minute and most apps idle. The managed rows include the work you would otherwise do yourself, and the gap between them is the fee for not doing it.

• Set a restart policy. Compose offers no, always, on-failure and unless-stopped. Use unless-stopped for a long-running service so it comes back after a reboot but stays down when you stop it on purpose.
• Add a healthcheck. A container that is running but not answering looks healthy without one.
• Keep database files on a named volume, never in the container layer, and test a restore before you need one.
• Pin image tags to a version, not latest, so a rebuild cannot change what runs.
• Watch the Docker Hub pull limit: 100 pulls per six hours for unauthenticated users, per IPv4 address or IPv6 /64, and 200 for a free authenticated account. Shared CI runners hit it.
• Build for the CPU you will run on. Multi-arch images avoid a surprise if you later move to Arm.

services:
  web:
    image: example.com/app:1.4.2
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://localhost:3000/api/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

If your image is not built yet, how to dockerize a Next.js app walks through it.

1. 1Who is paged at 3am?If the honest answer is you and you do not want it to be, skip the raw VPS and the panel. If you are comfortable with it, they are the cheapest route.
2. 2How many services, really?One to five: a VPS, a panel or a managed platform. Dozens with separate scaling needs: Kubernetes starts to earn its cost.
3. 3Where must the data live?Regions differ. If you need a specific one, that narrows the choice before price does.

Whichever docker hosting you choose, a standard image is portable. Moving between the four kinds is a change of host, not a rewrite, as long as the state lives in Postgres or a volume you can dump. Our pricing page shows how ours works if a managed platform is where you land.