DEV Community

OneLifeVibe
OneLifeVibe

Posted on

Adding a Cloud Is a Network Project: Onboarding a New Account Without Rewiring the Estate

Most teams treat a new cloud account as an administrative event. A form is filled in, billing is attached, an admin role is handed over, and the account exists. The first workload team then arrives and asks the question that should have been asked on day one: how does anything in here reach anything else, and what is it allowed to reach?

That is when the network work starts, usually under time pressure, and usually by copying whatever the last account did. The copied design carries the last account's exceptions, its temporary routes, and its undocumented shortcuts. Six months later the estate has two accounts that look similar on a diagram and behave differently in an incident.

Onboarding a new account is a network project with a repeatable shape. Done well, the new account joins the estate in days, with connectivity that is boring because it was decided in advance.

Decide the Address Space Before Anyone Deploys

The first decision is addressing, and it is the hardest one to reverse.

Allocate the new account a block from the estate's central address plan, sized for where the account will be in three years, not for its first workload. Accounts grow by accretion: a second region, a separate environment, a data store that needs its own subnet. A block that fits today forces either a painful renumbering later or a second, non-contiguous block that makes routing and firewall rules harder to read forever.

Record the allocation in the same register every other account uses, before the account is created. If the register is a spreadsheet, that is fine. What matters is that one place is authoritative, that overlapping allocations are impossible to miss, and that nobody allocates from memory in a chat message.

Reserve space inside the block in a consistent layout. Put infrastructure subnets, workload subnets, and endpoint subnets in the same relative positions in every account. Consistency here pays off daily: an engineer moving between accounts can read a subnet and know what kind of thing lives there, and a firewall rule written for one account translates cleanly to another.

Connect It the Way Every Account Is Connected

A new account should join the estate through the standard pattern, even when its first workload only needs one connection.

In practice that means routing through the estate's shared transit or hub layer rather than a bespoke direct link to whichever system the first workload talks to. A point to point link built for one project becomes permanent by accident, bypasses the inspection and logging applied to standard paths, and is discovered years later by an auditor or an outage.

Build the standard path first, then let the workload use it. If the standard path is genuinely too slow or too costly for a specific flow, that is a design conversation about the standard, not a reason to quietly build a second standard for one account.

Apply the same rule to DNS. The new account resolves names through the estate's shared resolution pattern from the start, with its own private zones delegated in the usual way. Retrofitting DNS after workloads have hardcoded resolvers and copied records is some of the most tedious work in cloud engineering, and all of it is avoidable in week one.

Set the Default Posture Before the First Workload

An empty account is the cheapest place to set security defaults, because nothing depends on the permissive version yet.

Start with egress. Decide which destinations workloads in this account may reach on the internet, if any, and route the rest through the estate's standard egress controls. An account that begins life with open egress accumulates dependencies on it within weeks, and closing it later becomes a breaking change nobody wants to own.

Do the same for account to account traffic. The default between accounts should be deny, with named flows opened deliberately: this service in the new account may reach that database in the data account, on this port, for this reason. Writing those flows down as they are opened builds the connectivity record as a byproduct of the work, instead of as an archaeology project before an audit.

Turn on flow logging and send it to the estate's central collection from the first day. Logs that start when the account starts are complete evidence. Logs enabled later always have a gap exactly where an investigator wants to look, and for SOC 2 readiness that gap is a finding waiting to be written.

Give the Account an Owner and an Exit

Every account needs a named team that owns it, recorded where the address allocation is recorded. Unowned accounts are where forgotten test systems, stale access, and unpatched workloads accumulate. When something in the account behaves strangely at 2 AM, the owner field is what turns a mystery into a phone call.

Ownership includes an exit plan, even a one line one. If this account had to be closed in a year, what would need to move, and where would it go? Teams that can answer that question keep accounts clean almost automatically, because they deploy as if portability matters. Teams that cannot answer it discover, during a migration or a divestiture, that one small account is load bearing for half the estate.

A Short Checklist That Prevents Most Regret

Before the first production workload lands, the account should be able to answer yes to a small set of questions. Is its address block registered and non-overlapping? Does it route through the standard hub, with no side links? Does it resolve DNS the standard way? Is egress controlled and logged? Are cross-account flows named and defaulted to deny? Do flow logs reach central collection? Is an owner recorded?

None of this is advanced engineering. It is sequencing. Each item is quick in an empty account and expensive in a full one, which is why onboarding deserves to be treated as a project with a checklist instead of a ticket that closes when the account appears in the console.

This kind of standardised account onboarding is a regular part of the private connectivity design work at CloudVerg, where multi-cloud estates built on VergMesh and reviewed through VergShield treat every new account as joining an existing network with known patterns for addressing, routing, and evidence, so growth adds capacity without adding surprises.

A new cloud account is never just an account. It is a new neighbourhood in a city that already has roads, rules, and residents. Build the roads first, and everything that moves in afterwards is simpler.

Top comments (1)

Collapse
 
robertpaschal profile image
Odinaka Robert Nnamani •

address space first. everything else you can rewire later, that one you cannot.