How Do You Design a Multi-Region Deployment?
Learn how to design a multi-region deployment: active-passive vs active-active, replication, conflict resolution, and CAP trade-offs.
Expected Interview Answer
A multi-region deployment runs a system's services and data across geographically separate cloud regions, typically to cut latency for global users and to survive a full region failure, and its core challenge is deciding how data is replicated and where writes are allowed — active-passive (one region writes, others stand by), active-active with partitioned ownership (each region owns a data subset), or active-active with conflict resolution (any region can write anything, conflicts resolved after the fact).
The starting point is routing: a global DNS or anycast layer (like Route 53 latency-based routing or a global load balancer) sends users to their nearest healthy region. The harder problem is data. Active-passive keeps one region as the sole writer and replicates asynchronously to standbys, which is simple and strongly consistent but wastes standby capacity and has a nonzero failover RTO. Active-active with data partitioning (e.g., sharding users by region) lets every region serve local writes at low latency with no conflicts, but requires cross-region calls when a user's data lives in another region. Full active-active with the same data writable anywhere needs a conflict resolution strategy (last-write-wins, CRDTs, or application-level merge logic) because network partitions between regions are inevitable, and by the CAP theorem you must choose to sacrifice either consistency or availability during a partition. Beyond data, teams must also handle region-aware service discovery, config drift between regions, and — critically — test actual region failover regularly, since it is a rarely-exercised path prone to silent breakage.
- Reduces latency for globally distributed users by serving them from their nearest region
- Survives a full region outage (natural disaster, cloud provider failure) without total downtime
- Enables data residency compliance by keeping regional user data within required jurisdictions
- Spreads load across regions, reducing the blast radius of a single region's capacity issues
AI Mentor Explanation
A multi-region deployment is like a cricket board running fully equipped stadiums in several cities instead of forcing every match to be played in one home ground. Active-passive is like keeping one main stadium hosting all matches while a backup stadium sits ready but idle, ready to take over if the main ground floods. Active-active is like multiple stadiums simultaneously hosting different matches, each managing its own local scoring, with a central body reconciling the overall league table afterward — and if two stadiums somehow report conflicting scores for a shared record, someone has to decide which one wins. The trade-off between the two mirrors the choice between simple consistency and maximum regional throughput.
Step-by-Step Explanation
Step 1
Add global routing
Use latency-based DNS or an anycast/global load balancer to route each user to their nearest healthy region.
Step 2
Decide the data ownership model
Choose active-passive, partitioned active-active, or fully conflict-resolved active-active based on consistency needs and CAP trade-offs.
Step 3
Implement replication and conflict handling
Set up cross-region replication (sync or async) and, if fully active-active, define a conflict resolution strategy like last-write-wins or CRDTs.
Step 4
Test region failover regularly
Run scheduled failover drills, since a rarely-exercised failover path is prone to silently breaking between the times it is actually needed.
What Interviewer Expects
- Distinguishes active-passive from active-active and explains the trade-offs of each
- Connects the design to the CAP theorem — you cannot have full consistency and availability during a network partition
- Names a concrete conflict resolution technique (last-write-wins, CRDTs, application merge)
- Mentions global routing (latency-based DNS, anycast) as the entry point to the design
Common Mistakes
- Proposing full active-active writable-anywhere without addressing conflict resolution
- Ignoring the CAP theorem trade-off during a network partition between regions
- Forgetting that failover is a rarely-tested path and needs regular drills
- Assuming multi-region automatically solves disaster recovery without also considering data residency and compliance
Best Answer (HR Friendly)
“A multi-region deployment means running the same system in more than one geographic location so users get served from somewhere close to them and the whole system survives even if one location goes down. The hard part is deciding how data stays in sync across those locations, since letting every location write at once risks conflicts that need a clear rule for resolving them.”
Code Example
routing:
type: latency-based-dns
regions:
- name: us-east-1
endpoint: api-use1.example.com
healthCheck: /healthz
- name: eu-west-1
endpoint: api-euw1.example.com
healthCheck: /healthz
dataStrategy:
mode: active-active-partitioned
partitionKey: userRegion
crossRegionReads: allowed
crossRegionWrites: redirected-to-owning-region
conflictResolution:
# used only for shared/global data replicated to all regions
strategy: last-write-wins
clock: hybrid-logical-clock
failoverDrills:
frequency: monthly
simulates: "full region outage with automated DNS re-route"Follow-up Questions
- How would you handle a user whose data is partitioned to a region that just went down?
- What is the difference between last-write-wins and using CRDTs for conflict resolution?
- How does the CAP theorem apply when two regions lose connectivity to each other?
- How would data residency requirements (e.g. GDPR) constrain your multi-region data placement?
MCQ Practice
1. In an active-passive multi-region setup, what happens during normal operation?
Active-passive designates one region as the sole writer, with other regions receiving replicated data and standing ready to take over on failover.
2. Why does fully active-active (writable in every region) require conflict resolution?
When any region can write anything and partitions are inevitable, conflicting concurrent writes must be reconciled after the fact using a defined strategy.
3. What does the CAP theorem imply for a multi-region system during a network partition?
The CAP theorem states that during a network partition, a distributed system must trade off between strict consistency and availability.
Flash Cards
Active-passive multi-region? — One region accepts writes; others replicate and stand by for failover.
Active-active partitioned? — Each region owns and writes its own data subset, avoiding conflicts by design.
Full active-active challenge? — Requires conflict resolution (last-write-wins, CRDTs) since any region can write anything.
CAP theorem relevance here? — During a cross-region network partition, you must choose consistency or availability, not both.