practice

Blast Radius Reduction

The set of deliberate partitions — accounts, regions, zones, cells, tenants, deployment stages — that bound how far any single failure or compromise can reach.

isolationresiliencesecuritydeployment

Worth treating as a checklist, because the partitions are independent and most estates have some and not others:

Identity and account — separate cloud accounts per environment and per major workload, so a credential compromise or a runaway process is contained. The strongest boundary a cloud provider offers.

Network — segmentation by tier, with egress restricted as well as ingress.

Availability zone — the unit of correlated physical failure.

Region — for regulatory or catastrophic scope, at substantial cost.

Cell — a complete copy serving a slice of customers, bounding logical failures that zones and regions do not.

Tenant — quotas and shuffle sharding so one customer cannot consume shared capacity.

Deployment — canary and staged rollout so a bad release reaches a fraction before it reaches everyone; this is the partition that catches the most incidents, because most incidents are changes.

Operation — limits on what a single command or automation run can affect, which is what turns an operator error into a contained event.

The exercise worth doing explicitly: name the largest blast radius in the system out loud. It is usually a shared database, a shared identity, a global configuration push, or a pipeline with production credentials — and then decide deliberately whether to shrink it or accept it.