
Restricting where an organisation may run workloads is one of the first jobs a CloudOps team takes on, and one of the easiest to get subtly wrong. Teams in the Netherlands and elsewhere in Europe often want resources kept in a short list of Regions, for latency, tidiness or data-location expectations. This tutorial walks through the work in eight ordered steps, from the first inventory to the first review of exceptions, and flags the places where the rule does less than people assume.
The sequence follows the Region-guardrail material in the Ops Atlas entry on CloudOps. The Atlas is candid that CloudOps remains a young field with nothing like a formal standard, so the steps below are a defensible working method rather than a certified procedure. Nothing here is legal advice on data residency; if a regulation drives your Region list, confirm the requirement with your own advisers.
Step 1: List what runs where today
Before restricting anything, find out where resources already are. Query each account for the Regions in use and record the result in a simple table: account, Region, number of resources, owner. This is the evidence for every later decision. It also reveals surprises, such as an experiment in a distant Region that nobody remembers. The Atlas gives a reason for starting here: existing resources are one of the blind spots of a Region rule, because once access to a Region is denied, the team that owns what lives there may no longer be able to reach it.
Step 2: Decide the allowed list and its reasons
Write the permitted Regions down together with the reason each one is on the list. Two or three is a common starting point for a mid-sized organisation. Put the justification in the same document, because the list will be challenged and “somebody decided” is a poor defence. Keep the CloudOps framing in mind: the aim is a predictable estate that is easy to monitor, not a maximum number of restrictions.
Step 3: Map the rule to your provider’s mechanism
The Atlas distinguishes several mechanisms. On AWS with Control Tower, a control named AWS-GR_REGION_DENY applies across the whole landing zone, while CT.MULTISERVICE.PV.1 can be applied per organisational unit. That difference matters: the first is blunt, the second lets a research unit differ from a payments unit. On Azure, the Policy setting called “Allowed locations” does the job, but it does not cover resource groups and ignores the special location value global. On Google Cloud, an organisation-policy constraint on locations decides only the place of creation; the place where data is later handled falls outside it. Choose the mechanism before you write a single exception, and record which scope (landing zone, organisational unit, subscription or folder) it will attach to, because that decision determines how many exceptions you will end up maintaining.
Step 4: Account for global services
Some services are not tied to one Region at all. The Atlas names identity services and certain endpoints that live in us-east-1 as examples. A rule that blocks those can break sign-in or administrative tooling in unexpected ways, so test it against the global services your teams actually use. List them explicitly and mark each one as allowed, or allowed with monitoring.
Step 5: Put the rule in audit mode first
Switch the guardrail on in a mode that only reports. For two weeks, collect every action that would have been blocked and ask who attempted it and why. This is the cheapest source of exceptions you will ever get. The Atlas’s wider 60-day plan puts audit-mode guardrails alongside the landing zone and quota alarms for the same reason: enforcement without observation tends to produce an angry queue and a campaign to remove the rule.
A small worked illustration, generic and not drawn from any real company: an audit-mode report after two weeks shows forty blocked attempts. Thirty are one analytics team spinning up test clusters in a Region nobody had approved, six are a backup tool writing to a default Region, and four are people exploring. The fix is different for each: add the analytics Region to the allowed list if the reasons hold, change the backup tool’s setting, and send the four a short note. Without the report, all forty would have arrived as angry tickets on enforcement day.
Step 6: Build the exception path before you enforce
Guardrails without an exception route are listed in the Atlas as an anti-pattern. A workable route has four parts: a request form that records the Region, the reason and the duration; a named approver; an automatic expiry; and a record that is reviewed every quarter. Make it fast enough that people use it instead of finding workarounds. If approval takes a week, the rule will be bypassed through shared credentials or a rogue account, which is far worse than the original problem.
Step 7: Enforce group by group
Turn enforcement on for one organisational unit first, ideally a team that is friendly and well-informed. Watch for failed deployments over several days. Then move to the next group. The Atlas’s 90-day outline expects enforcement by unit and all accounts brought through the standard creation process by that point, together with a first rehearsal of recovery. Enforcement by unit is also why the per-unit mechanism from Step 3 is worth the extra setup.
Step 8: Measure conformance and review exceptions
Region conformance, the share of resources that sit in permitted locations, is one of the metrics the Atlas proposes. Track it monthly and read it with the blind spots in mind: a high figure does not prove that data was processed only where you intended, nor that old resources were moved. Review the exception log at the same meeting. Exceptions that keep being renewed are a sign that the allowed list needs to change, and exceptions that nobody renewed can be closed.
Common failure points
- Treating the rule as a data-processing control. On at least one provider it governs creation only.
- Forgetting the management account. Service control policies on AWS do not affect it, and the Atlas advises against placing workloads there.
- Skipping recovery planning. If your disaster-recovery Region is not on the allowed list, the first real failover will also be the first rule violation.
That last point links to the Atlas’s argument about static stability: the more your recovery depends on creating or changing things through the control plane, the more a restrictive rule or an outage in that layer can slow you down. Pre-provisioned capacity in an approved second Region avoids that trap.
Where to read more
For the primary text, see the Atlas entry on CloudOps, including its guardrail section. If you are mapping this work to a career, our guide to the progression from entry-level DevOps work to seniority shows where cloud-estate skills appear on the ladder, and the piece on new career pathways for software developers explains how application developers pick up operational responsibilities like these.
When you finish the eight steps you should have five artefacts: an inventory table, a documented allowed list, a rule in the right mechanism, an exception process and a monthly conformance figure. Together they turn a vague wish to “keep things in Europe” into something that can be tested and audited.