Infrastructure Signals

Infrastructure is derived, not chosen from a catalogue. The first question is what the application needs, not which product to buy.

Managed is a boundary shift, not an abdication. Naming the remaining half is the sign they understand it.

Public, private and internal are the vocabulary of every infrastructure security conversation.

"What does an attacker get if this identity is compromised" is the question a policy is actually answering.

Backups are a storage job; recovery is an operational capability. Only one of them is tested by accident.

Knowing that the health check stays green through an egress outage is worth more than knowing it exists.

Redundancy that shares a failure domain is cost without benefit, and it looks identical on a diagram.

Egress, idle capacity and per-zone resources change architectures. Ignoring them is a design error, not a finance one.

Refusing complexity under pressure is harder and more valuable than adopting it.

A change nobody could review before it happened is a change nobody can explain afterwards.

Most production incidents are argued about in the gap between those two, and it should not be a gap.

Production is not a workload description. Kubernetes solves placement, restarts, rollout and discovery at a scale many systems never reach, and it adds a distributed system to operate.

It is easier, and it exposes every component to the internet directly. Public reachability should be a decision per component with a reason.

It makes the error go away and turns any compromise of that workload into a compromise of the account.

The provider protects against hardware failure. It does not protect against a bad migration, a wrong DELETE, or credentials that can drop the database.

Multi-region compute in front of one primary database in one region is single-region availability with a larger bill and more moving parts.

It is cheap at low and spiky utilization and expensive at sustained high throughput. The cost shape differs; the ranking does not hold in general.

They share the host kernel. That is why they start fast, and it is also why the isolation boundary is weaker.

State is how the tool knows what exists. It holds sensitive values, it must be locked, and losing it means reconciling reality by hand.

By the time the decision is made, capacity is still minutes away — metric window, provisioning, image pull and warm-up. You scale on the approach, not the arrival.

They secure the infrastructure. Your network boundaries, identities, policies, secrets, patches and data are yours, and that is where breaches happen.