If you have ever wondered what the technical team behind a hospital website or booking tool means when it says “we containerised it”, this explainer is for you. No jargon test at the door. We will use an everyday picture, a hospital’s central stockroom, to explain what a container is, why looking after containers has become a job of its own, and what a few common terms mean. The job, as one reference calls it, is ContainerOps. Nothing here is medical; it is all about the software that supports a busy orthopaedic and knee-care service.
The stockroom picture
Imagine a procedure kit packed in a sealed tray. Everything needed is inside, the label says what it is, and wherever the tray goes, it behaves the same. A software container is like that tray. The program, the libraries it needs and its settings are packed together, so it runs the same on a developer’s laptop, a test machine and the live server.
The sealed tray is called an image, and the running tray is the container. One image can start many containers, just as one kit design can be packed many times.
So what is ContainerOps?
It is a handy name for the everyday care of those trays. The write-up in the Ops Atlas entry for this topic uses it to mean building images you can trace, labelling and storing them sensibly, and running them in a predictable way on a machine. The Atlas concedes the term has no official definition anywhere; consider it informal shorthand, nothing more grand.
It also draws a line around the subject. Looking after one machine’s containers belongs here; coordinating thousands across many machines belongs to a different discipline. Our stockroom keeps its own shelves in order here, and leaves the hospital-wide logistics to someone else.
Layers: why the tray has a history
An image is built in layers, like adding items to the tray one at a time. Each layer records what changed. Here is the surprising part: if you add something sensitive in one layer and take it out in a later one, the earlier layer still contains it. That is why the Atlas recommends giving a build temporary access to secrets rather than copying them in. Think of it like photocopying a page with a password on it, then blacking it out on a later copy. The first copy is still in the drawer.
Labels and fingerprints
Every image can carry a human-friendly tag, such as “version 2” or, worse, “latest”. Tags can be moved, so the same word may point to a different image next month. A digest, by contrast, is a lengthy fingerprint worked out from what the image actually contains. If the content changes, the fingerprint changes.
In stockroom terms, a tag is “the blue tray” and a digest is the printed batch number. For an urgent job you want the batch number. The Atlas therefore says production should deploy by digest, with release tags that never change as friendly aliases.
Walls and allowances
Two kernel features keep containers from stepping on one another. Namespaces give each container its own limited view, as if it had its own private room. Control groups set allowances for memory, processor time and the number of processes. The Atlas advises always setting the memory and process allowances, and being gentler with processor limits, since a hard cap can briefly stall a busy program.
A common misunderstanding is that a container is a miniature computer. It is not. All containers on a host share the same core of the operating system, rather like departments sharing one building’s electricity supply. That is why code you cannot fully trust sometimes runs inside an extra protective layer, using tools the Atlas names as sandboxed runtimes.
Where the trays are stored
A registry is the central shelf. Teams push finished images there and servers pull them down. Good housekeeping includes deciding how long to keep old ones, because a busy team can pile up many gibibytes a week without noticing. The Atlas also suggests keeping a nearby copy so a slow or restricted public shelf does not hold up your work, and it cautions that such a copy is not a backup.
A few terms, in plain words
- Base image: the starting tray you build on. Smaller starting trays mean fewer things to keep up to date.
- Root user: the all-powerful account. Containers are safer when they do not run as it.
- Compose: a short text file describing several containers that belong together on one machine.
- Exit code 137: the program was forcibly stopped, often because it ran out of memory or did not respond to a polite stop request.
- Rootless: running containers without administrator rights, which limits the damage if something escapes.
One container’s day, start to finish
It helps to follow a single tray through a day. In the morning a developer writes a recipe describing what goes in. A build machine follows the recipe and produces an image, which is pushed to the shelf and given its fingerprint. In the afternoon the server pulls that exact image and starts a container, applying the allowances set for it. Overnight, the program writes notes in its log. If the host asks the container to stop for maintenance, a well-made program tidies up and exits within a few seconds. A badly made one ignores the request and is switched off abruptly after a short wait, ten seconds by default in Compose.
Each stage offers a chance for a small habit to pay dividends later, which explains why the routine earned a name.
Why a hospital team should care
Booking tools, visitor information pages and staff intranets are not glamorous, but patients and staff lean on them every day. When one breaks because an unattended image went stale, or a public download limit blocked an update, the inconvenience lands on real people. Careful container habits reduce those surprises, and they do it without any clinical knowledge or expensive equipment.
Three everyday mistakes
The first is trusting a floating tag in production, which is like grabbing “the blue tray” and hoping it is the same one as last week. The second is leaving a container exposed to the whole network when only the proxy needs to be seen; the Atlas notes that the engine itself describes this default as insecure by default. The third is nobody owning the container, so it quietly ages like an unattended supply cupboard.
Questions readers often ask
Do we need to learn all this ourselves? Not necessarily. Many hospitals rely on a vendor or an outside team. Even then, it helps to know the vocabulary so you can ask whether images are pinned, whether containers run as ordinary users and who owns each one.
Is a container less safe than a normal server? Neither is automatically safer. A container with a low-privilege account, a locked filesystem and tight allowances can end up stricter than a general-purpose server. One that runs with full privileges and the engine’s control socket mounted inside is far looser.
How often should images be rebuilt? Often enough that a fix published upstream reaches your live machine quickly. As an illustration the Atlas mentions a week for the most serious problems; your own figure should be whatever your team can honestly meet and measure.
What good looks like
A healthy setup is not exciting. Images are rebuilt on a schedule, an update bot proposes new bases, services run as ordinary users with limits, and every container has a named owner and a place where its rebuild steps are written down. Progress is checkable: what share of production is pinned by digest, how old the bases are, how quickly a published fix reaches the live machine.
Further reading on this site
If you want the wider picture of how IT teams are organised, try the article on the unified architecture behind XOps, and the practical piece on how outside engineers can smooth day-to-day cloud and software running. Both sit one step above the container level and explain how it fits.
The next time someone in your hospital IT team mentions a container, you will know the picture: a labelled, sealed tray, a shelf, a set of allowances, and a person whose name is on the label. That is the heart of it.
