Every student pilot learns early that aviation has its own language. Before you can fly a pattern you must be able to say “squawk”, “NOTAM” and “flap setting” without stopping to think. Software teams have the same habit, and one of their newer phrases is ContainerOps. This primer borrows the study methods of the ground-school classroom to explain it, one term at a time, so a curious reader from the flying world can follow the conversation when engineers start talking about images and registries.
One caution about the name. According to the Ops Atlas entry on ContainerOps, this is the Atlas’s own name for a job rather than an industry-standard term, and the entry openly calls it ambiguous. It covers the care of individual containers beneath any orchestration layer: how images are built, how they are named and stored, and how they behave while they run. Think of it as the discipline of the single aircraft rather than the airline schedule.
Why a glossary works for this subject
If you have prepared for an aptitude paper, you know that vocabulary comes before speed. Our own readers who are working through typical aptitude paper questions asked of Indian pilot candidates often find that half the difficulty is simply decoding the wording. Container work is similar. Once the nouns are clear, the rules around them look ordinary, even sensible. So here is the glossary, arranged in the order you would meet the ideas on a real project.
Image
An image is a sealed, read-only bundle: the application plus every file it depends upon before it can start. In aviation terms, it is the fully assembled and inspected aircraft sitting on the ramp, as opposed to a pile of parts on a bench. The Atlas places the open image format under the Open Container Initiative, a body that launched in June 2015, and it lists version 1.1.1 of the image specification, dated March 2025. A standard format matters because many different tools can then read the same package.
Layer
Images are built from stacked layers, and the Atlas stresses that each layer records a set of changes. This has a consequence that surprises newcomers. If one layer adds a password file and a later layer deletes it, the earlier layer still contains the file. Compare it with a maintenance logbook: striking out an entry does not unwrite it, because the page underneath is still there. The recommended remedy is to build with temporary secret mounts so the sensitive material never lands in any layer at all.
Tag and digest
A tag is a readable nickname that someone can re-attach to another build whenever they like. A digest is a lengthy fingerprint derived from the precise bytes of the image, so it only ever names one thing. Pilots will recognise the difference between calling an aircraft “the blue trainer” and quoting its registration and serial number. The Atlas treats deploying by digest, with release tags that are never reused, as the dependable way to know what is flying.
Registry
A registry is the storage hangar where images wait until someone pulls them out. Public ones are convenient but come with usage caps. The Atlas notes that anonymous pulls from one popular public service were limited to 100 per six-hour window when it was checked, with a higher cap for signed-in free accounts. A private registry, or a mirror that caches popular images, works like a local parts depot so you are not queuing at a distant counter every time.
Runtime
The runtime is the engine that actually starts a container from an image. The Atlas names containerd and runc as the common pairing. If an image is the aircraft, the runtime is the crew that preflights it, starts it and hands it a defined slice of resources. It also records the commands and settings under which the thing was launched, which is the container equivalent of a flight plan filed before departure.
Namespace
Namespaces give each container its own view of the system: its own process list, its own network view, its own file tree. The result feels private. Yet the Atlas is explicit that this is isolation from a shared kernel, not a separate machine. Imagine several classrooms in one building that all share a single power supply. Walls hold in everyday use, but a fault in the shared supply would reach everybody. That is why the entry mentions sandboxed runtimes for code you do not trust.
Cgroup
Control groups, usually shortened to cgroups, are the meters and limits. They decide how much memory and how many processes a container may use. Treat them like weight and balance: you would never dispatch an aircraft without checking the load, and the Atlas wants a memory ceiling plus a cap on the number of processes on every container. Processor caps get a gentler recommendation, because they can throttle short bursts and make an application feel sluggish while its average use looks modest.
Exit code 137
This is the number a container leaves behind when it was ended forcefully. It equals 128 plus 9, and signal 9 is the unconditional kill. Often the cause is a memory ceiling being reached, though the Atlas adds that on current kernels the status may also be reported when memory is exhausted across the entire machine. It is the closest thing to an uncommanded shutdown report, and reading it correctly saves hours of guessing.
Signal and grace period
When an orchestration tool or an operator wants a container to stop politely, it sends a request called SIGTERM and then waits. In Compose, the default wait is ten seconds before the forceful kill arrives. Programs started through a shell wrapper may never receive the polite request, so the Atlas recommends the exec form of the entrypoint and a tiny init helper. It is the difference between a crew that hears the briefing and one that was standing outside the room.
Rootless
A rootless container runs without administrator rights on the host. If something escapes, the intruder inherits very little. The Atlas also lists the trade-offs: resource limit flags can be ignored without a modern cgroup version and a service manager, ports below 1024 need a system setting, and the networking is slower. Anyone who has flown with restricted privileges on a student licence will recognise the pattern of safer, but with a few more conditions to remember.
How to practise these words
The best way to retain vocabulary is the same one used for navigation charts: repeat, apply, check. Take a small application you already run, and answer five questions about it. Which base image does it start from? Is it deployed by tag or by digest? Does it run as an administrator? Are memory limits set? Where do its logs go, and are they rotated? Wherever the honest answer is a shrug, that is your first training item. That kind of preparation echoes the habits described in our guide to building a study routine for aptitude assessments in India, where steady practice beats cramming.
Where the discipline stops
It helps to know what is outside the cockpit. The Atlas leaves orchestration, signing and scanning in depth, admission policy, the host operating system and log pipelines to other disciplines. ContainerOps stays with the single container and its image. That boundary keeps the vocabulary manageable, much as a private pilot syllabus stops before airline transport theory.
A short self-test
- Explain tag versus digest in a single sentence.
- Why does deleting a file in a later layer not remove it from the image?
- What might an exit code of 137 tell you?
- Why is a shared kernel relevant to the word “isolation”?
- Which two limits does the Atlas say belong on every container?
If you can answer these aloud, you already speak enough of the language to follow a team discussion, ask a sensible question, and notice when something sounds wrong. Vocabulary is only the first step, but in aviation as in operations, a shared and precise vocabulary is what keeps people from talking past each other when it matters.

