Skip to content
DisrupsoftDisrupsoft
General

Custom software development: where and how to deploy it

Custom software development: where to host it, what Docker adds, how to set up CI/CD with GitHub and what to ask about costs and who owns what.

Custom software development: where and how to deploy it

The quoting system went live in March. Six months later, the developer who set it up left the vendor, and nobody else knew how to get into the server. The password was sitting in an email, the server was a virtual machine paid for with a former manager's credit card, and the only way to ship a change was to SSH in and copy files by hand on a Friday night.

In custom software development, people talk a lot about what will be built and very little about where the system will live and how it will be updated. That part, application deployment, decides how much each change costs you, who can touch production and what happens if the vendor disappears. Settle it before you sign. Once the system is running, it gets much harder.

We covered project stages and what you should receive at handover in software design and development: stages and deliverables. Here we focus on the last stretch: getting the system into production and keeping it there.

Where your system can live

There are three main options, and none of them wins in the abstract. It depends on how much you want to manage yourself, how much traffic you expect and what your industry requires.

OptionWhat you manageWhen it fitsWhat to watch
Managed cloud (PaaS, for example Azure App Service)The application and its configurationInternal systems and portals with variable demand; teams without a server adminCosts that grow with usage, and add-on services nobody turns off
Virtual server (a VPS or a cloud VM)The operating system, patches, firewall, backupsStable workloads, tight budgets, need for fine-grained controlSomeone has to patch the server every month
Your own server (on-premises)Everything, hardware includedIntegrations with systems that only live on the internal network, or requirements to keep data in-housePower, off-site backups, hardware refresh

Managed cloud saves you from maintaining the operating system, but not from maintaining the application. A VPS is cheaper per month and costs more in someone's hours. And your own server makes sense when the integration demands it, say when the system has to talk to an AS/400 that never leaves the company network.

A rule of thumb: pick the option that someone at your company or at your vendor can run without stress. If nobody knows how to patch Linux, the cheap VPS ends up expensive.

What Docker brings to application deployment

Docker packages the application together with everything it needs to run (runtime, libraries, base configuration) into an image. That image runs the same way on the developer's laptop, in the test environment and in production. The classic "works on my machine" problem shrinks a lot.

In practice, for your company:

  • You can switch hosting without redoing the installation. The same image runs on a VPS, on Azure or on your own server.
  • Versions are clearly identified. If an update fails, going back to the previous image takes minutes.
  • The installation recipe is written down in a file (the Dockerfile) that lives in the repository instead of in one person's head.

One licensing point that often gets missed: Docker Engine, the engine that runs containers on Linux servers, is open source. Docker Desktop, the app many developers use on Windows or Mac, is a different thing. Under the Docker Desktop license agreement, it's free for companies with fewer than 250 employees and less than USD 10 million in annual revenue; if your company exceeds either threshold and your team uses it, you need a paid subscription. If a vendor does the development on its own machines, the license is their problem. If your in-house team does it, it's yours.

CI/CD with GitHub: every change takes the same path

CI/CD means every code change automatically goes through the same tests and the same release process. With GitHub Actions, it usually looks like this:

  1. The developer pushes a change to a branch and opens a pull request.
  2. The pipeline builds the code, runs the automated tests and builds the Docker image.
  3. Someone else reviews the code and approves the merge.
  4. Once merged, the system deploys itself to the test environment.
  5. After validation, an authorized person approves the move to production.

The most useful part is the record: it shows what changed, who approved it and when it reached production. If something breaks, you know which version to look at.

On cost: in January 2026 GitHub lowered the price of its hosted runners. According to its billing documentation, each plan includes a monthly amount of compute for private repositories: USD 12 on Free, USD 18 on Team and USD 300 on Enterprise Cloud (for reference, USD 300 covers about 50,000 minutes on a standard Linux runner at USD 0.006 per minute). Standard runners are free for public repositories. Self-hosted runners don't draw from that amount; GitHub had announced a per-minute charge for them starting March 2026, then postponed it to rethink the approach. For a typical internal system, the pipeline is rarely the expensive part of the bill.

One plan detail does matter. In private repositories, mandatory approvals before deploying to an environment ("required reviewers") are only available on GitHub Enterprise; on Free, Pro and Team they only work in public repositories, according to GitHub's environments reference. If your plan doesn't include them, you can handle production approval with branch protection rules or a documented manual step. What you can't have is a production release that depends on someone's memory.

Test and production environments

At a minimum, a custom system should have two separate environments: one for testing (QA or staging) and production. A few things worth agreeing on:

  • The test environment should match production in versions and configuration. If testing runs on a different database, the tests are worth less.
  • Credentials for each environment are stored as pipeline secrets or in a secrets manager, never in the code.
  • Don't copy the production database into testing as is. If it holds personal data about customers or employees, that data ends up exposed in an environment with weaker controls. We explain why in real data in test environments and Law 21.719.
  • Decide who signs off in testing before anything goes to production. Ideally someone from the business side, not just the technical team.

Costs and responsibilities: what to ask before signing

Deployment brings recurring costs and tasks that someone has to own. These questions help keep them from falling through the cracks:

  1. Whose name are the hosting subscription, the domain, the repository and the GitHub organization under? The healthy answer is your company's, with the vendor as an invited user.
  2. What does the monthly infrastructure cost include, and which services are billed separately (database, storage, backups, certificates)?
  3. Who applies operating system and Docker image patches, and how often?
  4. How are backups made, where are they stored, and when did someone last test a restore?
  5. Who gets the alerts if the system goes down after hours?
  6. How long does it take to roll back to the previous version if a deployment fails?
  7. If you switch vendors, what do you get so another team can deploy the next day? At a minimum: code, Dockerfile, pipeline definition and infrastructure documentation.

What happens after go-live (maintenance, version upgrades, the yearly budget) is covered in detail in custom software development: what happens after delivery.

Frequently asked questions

Do I need Docker if my system is small?

It isn't required, but it helps even with small systems: it documents how to install the system and lets you change hosting without redoing the setup. Adding it at the start of a project costs little.

Cloud or on-premises for an internal system?

If the system doesn't need to connect to equipment that only exists on your network, managed cloud is usually the simplest to run. Your own server makes sense for local integrations or specific requirements about where the data must live.

Can the repository stay in the vendor's account?

It can, but it's a bad idea. If the repository and the pipeline belong to your company, switching vendors or bringing in another team doesn't depend on a handover.

Ship without depending on one person

At Disrupsoft we build custom software with Angular, Node.js, Symfony and Flutter, and we work with Docker, GitHub, CI/CD and Azure so every delivery reaches production along a known path. If you're about to start a project, or you have a system that still gets deployed by hand, let's talk about your custom software development.

Desarrollo De SoftwareSoftware A MedidaDevopsDockerCi Cd

← View more articles