Picture a small consulting firm that renovates old warehouses. On the first day of every job the foreman unrolls a drawing, and by the last day the walls rarely match it exactly. A pipe was moved to dodge a beam, a door was widened for a forklift, and nobody updated the paper. Infrastructure in the cloud behaves the same way, and InfraOps is the habit of noticing the difference before it hurts. Here is the whole idea explained in everyday language, with no tool knowledge needed.
Three pieces of paper, not one
The Ops Atlas, an online reference on operations work, describes infrastructure managed through code as a matter of three copies. Its InfraOps entry calls them code, state and reality. In our warehouse story, code is the architect’s drawing, state is the foreman’s notebook listing what he believes he has built, and reality is the actual building. Most of the time they match. The skill lies in knowing what to do when they do not, since each pairing can fail differently.
Four ways the paper and the building part company
The Atlas lists four typical gaps, and they translate neatly.
- Approved but never built. A change was signed off in the drawing but nobody carried it out on site.
- Drift. A worker altered something on site and told no one, so the building now differs from both drawing and notebook.
- Unlisted extras. A storage shed went up that appears in neither the drawing nor the notebook.
- Demolished behind the notebook’s back. Something was knocked down, yet the notebook still lists it as standing.
Once you can name the four gaps, a messy situation becomes a short checklist. That is the first real benefit of the vocabulary.
Why a “clean report” can still hide trouble
Suppose the foreman compares his notebook with the drawing and finds nothing wrong. Reassuring, but he has not walked the site. The shed that nobody recorded is invisible to that comparison. The Atlas makes the same point about infrastructure previews: a clean result only speaks about things the tool already tracks. To catch the shed you need a separate walk-through of what actually exists, which in cloud terms means comparing the provider’s inventory with your list.
Small buildings are easier to inspect
Few consultants would try to supervise one sprawling mega-site alone. In the same way, the Atlas advises against putting everything into one giant record. It quotes guidance from Google Cloud that a single state should avoid going beyond about a hundred resources, and ideally holds only a few dozen. Smaller records mean quicker checks and, importantly, a smaller zone of damage when someone makes a mistake. For clients with heavy visitor traffic, our piece on handling very large web audiences on modern cloud foundations shows why that kind of separation pays off.
Look before you pour concrete
Before any structural change, a sensible foreman asks for a written description of exactly what will happen. The infrastructure version is a preview of changes, sometimes called a plan. The Atlas calls skipping its review a classic anti-pattern, usually excused with the idea that the code documents itself. Reading the preview is where the “delete the database” surprises get caught. Previews also expire: once one has been carried out, any others drawn against the old notebook are out of date and must be redone.
Keep the notebook safe
The notebook, or state, deserves a lock on the door. Only one person should write in it at a time, which is what a state lock enforces. The Atlas adds a human rule: never break a lock that you did not set, because the owner may be mid-task. It also warns that notebooks and previews can contain secrets, so they should not be passed around casually or published as downloadable artifacts.
What if the job goes wrong halfway?
Unlike a film set, a building project cannot rewind. The Atlas says the same of infrastructure changes: there is no automatic transactional rollback. The remedy is to fix forward, make a new plan from the current situation and carry it out. Planning this way, with small steps, is far less dramatic than hoping an undo button exists.
The rebuild test
A drawing is only trustworthy if you could build from it again. Good renovation firms prove this occasionally by reconstructing a small section from scratch and seeing whether it comes out right. The Atlas has the same idea in its maturity ladder: a team reaching its higher stages is expected to pass a rebuild drill, creating a whole environment from the code and confirming that it works. The first attempt almost always uncovers a forgotten manual step, a hidden dependency or a setting that lived only in somebody’s memory. Finding those in a drill is cheap; finding them during an outage is not.
A few numbers to keep the story honest
Stories can mislead, so the Atlas suggests counting. How much of the site does the drawing cover? How many unreported changes did the weekly walk-through find? How long did each one stay unresolved? It argues that a sentence such as “hundreds of items nobody reviews, plus one change left alone for more than two weeks” moves people more than a bare percentage ever will.
Where to start in a real project
Begin small. Pick one environment, write its drawing, and check the building against it on a regular schedule. Our guide to gathering requirements for a web project properly is useful for the earlier stage of agreeing what the drawing should contain. Then measure how much of the site the drawing truly covers, because an unmeasured plan tends to flatter itself.
That is the whole story. Three copies, four gaps, a habit of inspection and a rule of fixing forward. The rest of the field is mostly detail added to that picture.

