Everything in this account is made up: the school, the staff and every number are stand-ins, there so that a handful of CloudOps ideas can be seen in motion. Imagine a small flight-training organisation with a scheduling system, a student records portal and a simulator booking service, all running on a public cloud. Its two-person platform team is about to learn what the word “operations” means when the provider, and not the organisation, controls the layer underneath.

Monday: the schedule that would not load

Early on a Monday, instructors cannot open the dispatch board. Students in the briefing room are looking at spinning wheels. The platform lead, a former line mechanic, does what she was trained to do: she refuses to guess. Her first question is whether the application is down or whether the platform beneath it is. A quick check of the provider’s health page shows nothing wrong, so she widens the search. Her colleague notices that the monitoring dashboards they use to see everything are hosted in the same provider and Region as the system under investigation.

That detail turns out to matter. The Ops Atlas, whose CloudOps entry in the Atlas collects real incidents for this reason, describes a 2025 outage at a major provider in which a faulty quota-policy change spread across the world in seconds, and where customers who hosted their monitoring on the same platform lost sight of their own systems. The lesson the Atlas draws is not that clouds are unreliable. It is that visibility should not depend on the thing being observed.

Tuesday: the team writes down what it owns

By Tuesday the dispatch board is back, caused by an ordinary application bug. But the near-miss has done its work. The lead gathers the team and draws two columns on a whiteboard: what the provider is responsible for, and what they are. It is the shared-responsibility idea in its simplest form. The column on the provider’s side includes data centres and the managed database service. Their own column includes access settings, backups, the accounts themselves and, crucially, how they respond when the provider announces a change.

Aviation people recognise the habit. An operator knows the manufacturer is responsible for the design and the operator for the maintenance programme. The exercise sounds basic, yet it surfaces gaps. Nobody owns the mailbox that receives provider notices. Nobody knows who may raise a service limit. Two accounts were created by hand years ago and carry no standard settings.

The first month: inventory, then routing

Using the Atlas’s thirty-sixty-ninety outline as a loose guide, the team starts small. In the first thirty days they list every account, owner and bill, and they redirect provider notices to a shared group inbox with an on-duty owner. This is the cheapest step in the plan and the one that pays back first, because a missed retirement notice, like a missed airworthiness bulletin, can turn into an urgent problem.

While the list grows, the team compares it with a loose picture of the school’s actual activity: flight bookings, instructor accounts and the simulator scheduler. Anything that cannot be attributed to a person or a purpose goes on a “question” list. The training organisation’s own pages on the instrument-procedures simulator joining the programme show why booking systems for scarce equipment cannot be treated as an afterthought: when a device is limited, a scheduling outage affects every student queuing for it.

Months two and three: a landing zone and gentle guardrails

In the second month the team builds a landing zone, a template that creates a new account with the standard baseline already applied. Previously each new project meant a manual setup and a small gamble. Now a request produces an account with consistent logging, tagging and access rules. They resist the temptation to put the new rules into force straight away. Following the Atlas’s advice, the guardrails begin in audit mode, reporting what would have been blocked.

The audit report is humbling. One legacy tool launches resources in a Region no one remembers choosing. Another team has been tagging by hand, with inconsistent spelling. Neither is a disaster, and both are fixable, because nothing was blocked while the problems surfaced. By day ninety, in the Atlas’s target, more than 80% of accounts carry the baseline. The school is a little short, but trending in the right direction, and the team records the gap instead of hiding it.

The game day

The team now turns to recovery. Their tier-one service is the dispatch board, and the recovery target is short. The Atlas offers a principle that the lead finds almost mechanical in its logic: the shorter the recovery target, the fewer management-layer steps should sit on the recovery path, and for the most critical failover there should be none at all. Steps such as creating new resources, changing policies or requesting more capacity are control-plane actions, and in a regional emergency they may be slow or unavailable at the very moment they are needed.

So the team reviews its runbook with a highlighter. Every step that asks the platform to build something new is marked yellow. There are nine. The redesign pre-provisions the standby capacity so that failover becomes a traffic switch and not a construction project, a pattern the Atlas calls static stability. The team then schedules a game day, a rehearsal in which they pretend the primary Region is gone and count how many highlighted steps remain. On the first attempt they miss two. A pre-signed access role was never tested, and a certificate was only valid in the primary Region. They fix both and run it again a fortnight later with zero yellow steps.

It reminds the lead of simulator sessions in her own training, where the point of an engine-failure scenario is not to pass but to discover what you forgot. For a closer look at flight-deck hardware that makes this style of practice possible, the school’s announcement of an advanced glass-cockpit twin added to its training fleet sets out the kind of equipment that rewards rehearsal.

An aside about ceilings

Midway through the quarter, a new term’s enrolment arrives early and the simulator scheduler needs more compute than usual. The launch fails. Because the team had read about the three failure types, they do not panic: the quota console shows the account at its ceiling, and the cause turns out to be a reservation made for a previous peak and left idle, which still counted against the on-demand allowance. Releasing it clears the problem in minutes. The team then adds an alarm that warns when usage crosses eighty per cent of any ceiling that matters, and an entry in the calendar to review those ceilings before each intake.

What the story shows

Notice how little of this required a new product. The team named owners, routed notices, standardised how accounts begin, tried rules before enforcing them and rehearsed a failure. The Atlas labels its maturity ladder as its own scheme: five rungs, beginning with improvised practice and ending with a resilient stage where rehearsals prove that nothing on the critical recovery path needs the management layer. The fictional school has climbed one or two rungs in a quarter, and has evidence for each.

A few cautions apply even to a tidy tale. A single team’s rehearsal does not prove that a provider will behave during a real event. The Atlas also says plainly that the field is emerging and that its own emphasis is editorial. And the numbers used here, nine steps, two misses, a fortnight, are invented for the story and should not be read as benchmarks.

Questions to take back to your own team

  • Where do provider notices go today, and who answers them?
  • Does your monitoring survive the failure of the platform it watches?
  • How many steps in your most critical recovery plan require the provider’s management layer to be working?
  • When did you last rehearse, and what did the rehearsal teach you?

The lead closes her notebook at the end of the quarter with a short entry: the most useful thing the team did was to stop assuming. In aviation and in operations alike, the quiet work of writing down who owns what, and then testing it, is what keeps ordinary days ordinary.

© 2024 ADVANCED INTERNATIONAL AVIATION ACADEMY. ALL RIGHTS RESERVED.

STAY CONNECTED WITH US: