How do you design multi-region disaster recovery for a production ML system?

Updated Aug 1, 2026

Short answer

Replicate model artefacts and container images cross-region, define RTO/RPO per component, keep infrastructure as code so a region can be rebuilt, and choose warm standby or active-active based on tolerance — testing failover regularly rather than assuming it works.

Deep explanation

Start by stating requirements, because the architecture follows from them: RTO (how long may it be down) and RPO (how much data may be lost). "Zero downtime" is a budget conversation, not an architecture.

What must be replicated

Component	Mechanism
Model artefacts	S3 Cross-Region Replication
Container images	ECR cross-region replication
Feature data (online)	DynamoDB Global Tables or app-level dual write
Training data	S3 CRR (often lower priority — training can wait)
Model registry	Register packages in both, or replicate on approval

|…

Unlock with a Pro subscription to view this section.

View pricing