5 cloud architecture mistakes

Quick reference

1. ❌ Everything in one subnet — One route table means nothing is private. ✅ Split public and private — Each subnet gets only the route it needs.

2. ❌ A single availability zone — One zone outage takes the whole app down. ✅ Two zones from day one — A zone failure becomes a failover, not an outage.

3. ❌ Long-lived access keys — A leaked key outlives the person who made it. ✅ Roles and OIDC federation — Credentials expire in minutes, not years.

4. ❌ No tags, no budgets — The bill arrives with no way to split it. ✅ Owner and cost tags — Every resource names who pays for it.

5. ❌ Backups never tested — An untested backup is a hope, not a plan. ✅ A restore drill each quarter — You learn the restore time before an incident.

Explained

The five

1. Everything in one subnet. One subnet means one route table, and a route that reaches the internet makes the subnet public. Split public from private and route the private side through a NAT gateway.

1. A single availability zone. One zone outage takes the whole app with it. Two zones turn that outage into a failover; cross-zone traffic costs a little, downtime costs more.

1. Long-lived access keys. A key in a CI variable outlives the person who made it. Roles and OIDC federation give workloads credentials that expire in minutes.

1. No tags, no budgets. The bill arrives with no way to split it and nobody can name the team behind the expensive queue. Owner and cost-centre tags cost nothing now and hurt to retrofit.

1. Backups never tested. An untested backup is a hope. A quarterly restore drill gives you the restore time before an incident needs it, and proves the backup exists.

What they share

Each one is a shortcut in the architecture, not in the code, which is why a code review never catches it.

In one line

Check your subnets

$aws ec2 describe-subnets