InfraOps for Charter Teams: Fourteen Straight Answers

InfraOps for Charter Teams: Fourteen Straight Answers

Charter operators and the travel platforms around them run on software, and software runs on infrastructure that needs steady care. People in those teams keep asking the same questions, so here they are with straight answers. The vocabulary belongs to InfraOps, a casual term for running infrastructure from code, and each fact is taken from the Ops Atlas, specifically its long InfraOps reference page.

What is InfraOps, in one sentence?

In the Atlas it is shorthand for infrastructure operations, stretched over everything that happens to a code-defined setup: drafting, review, rollout, drift checks, refactoring and retirement. It is neither software nor a qualification.

Is it the same as a product with that name?

No. A commercial offering with an identical name appeared in 2025, which the Atlas treats as a coincidence of naming rather than a definition. When you see the word in a job ad or a sales pitch, ask which meaning is intended.

Why would a charter business care?

Because a quote engine, a booking calendar, a crew-scheduling tool and a payments page all depend on servers and networks. If a flight request fails at the wrong moment, a customer books elsewhere. Reliable infrastructure is less visible than a polished aircraft, but it decides whether the booking arrives.

Pricing is a good illustration. The piece on how repositioning flights are sold at reduced prices depends on live availability, which in turn depends on systems that must be consistent in real time.

Which three records should agree?

Code states what you intend. State is the tool’s saved record of what it created. Reality is what the cloud provider actually runs. Most surprises come from the gaps between them, and the Atlas names four: a change merged but never applied, drift introduced by hand, resources never brought under management, and items deleted outside the tool’s knowledge.

What does a clean plan tell me?

Less than people think. It tells you that managed resources match the code. It says nothing about things the tool does not know. The Atlas shows this with a drift tutorial in which a hand-made firewall group surfaced only as an identifier in a diff.

How big should one configuration be?

Small. The Atlas cites a cloud provider’s advice of no more than a hundred resources per state, and ideally a few dozen. It also mentions a messaging company that grew to roughly 1,400 state files by 2022, arranged per region and per account. Splitting makes each preview faster and each mistake cheaper.

Can I just apply from my laptop?

You can, but the Atlas files it under habits to drop. Nobody else sees what was run, local copies go stale, and strong credentials end up spread across personal machines. A shared pipeline with a reviewed plan is the better route.

What if an apply fails halfway?

There is no transactional rollback. You fix forward with a fresh plan. If a state lock is held, leave it alone unless you placed it; somebody else may still be mid-change.

How do I know if things have drifted?

Schedule a comparison. The Atlas describes running a plan with a special exit status so that a pipeline can tell the difference between no changes and changes found. A refresh-only variant records reality without touching anything. Alert on the result and give each finding an owner.

Cloud-specific drift tools have quirks: one inspects only the properties you set explicitly, so a gigabyte can be flagged against 1024 megabytes. A different preview mode can show properties as removed that still exist. Read results critically before acting.

Should drift be fixed automatically?

Only in narrow cases. The Atlas says reverting everything by reflex amounts to an unreviewed apply pipeline with the strongest credentials. It endorses it solely for components with no lasting data, with no incident open; elsewhere, notify an owner and aim to settle things within about three days.

What is the cost of one wrong command?

Potentially large. The Atlas looks back at a 2017 storage-service outage that began with one command and one wrong input, after which the operator throttled removals and set capacity floors. Guardrails of that kind are cheap compared with a grounded schedule.

Operators who already plan an empty-leg schedule know the logic, as in this traveller-oriented guide to empty-leg options: reposition in small steps and never commit more than the schedule can absorb.

Do secrets end up in state?

They can. The Atlas warns that stored state and saved previews can hold sensitive values, and that plan runs execute provider code, so the credentials used for planning need nearly the same care as those used for applying. Newer tool versions add options including short-lived resource types, arguments that are never stored, and encrypted state, so check what your version supports before you assume secrets are safe.

Should we bring every existing resource under code at once?

Probably not. Imported definitions tend to mirror historical accidents, and one provider recommends avoiding imports when you can. Pick the handful of riskiest items first, bring them in, review the result and only then continue. A steady quarter of careful imports beats a weekend of rushed ones.

Who needs to be in the room?

Anyone who can approve a change to production, plus whoever answers customer calls when production misbehaves. Both groups should see the same weekly summary: how many items sit outside code, how old the oldest drift finding is and which changes bypassed the pipeline. Keep the meeting to fifteen minutes and finish with a named owner for each open item. Charter customers will never see that list, but they feel its effects every time a quote loads quickly at the hour they need it.

What should I measure?

The Atlas, speaking for itself and not for any standard, tracks codified coverage, the rate and lifespan of drift, writes outside the pipeline, failed applies, blast radius and image age. It dismisses vanity figures such as code volume, module tallies and tag totals.

Where do I start?

In the first month, set up a trustworthy backend for state, a pipeline that produces plans and one production change applied through it. In the second, break up large states, put a gate on plans, run drift checks overnight and retire laptop applies. In the third, import the riskiest unmanaged resources, rebuild machine images weekly and run a first full rebuild drill.

If I do one thing first, which one?

Count what is not under code. Write the absolute number on a whiteboard where the team can see it. Everything else follows from agreeing on how big that number is.