Active-Active

The Topology

 [users EU]──►[region-EU]◄────┐
                              │ replication (all directions)
 [users US]──►[region-US]◄────┤
                              │
 [users APAC]─►[region-APAC]◄─┘

 ALL regions active: serve local traffic, replicate state
 continuously among themselves. no promotion dance —
 a region dying = traffic reroutes, done.

Why Teams Pay For It

 □ RTO ≈ SECONDS: no failover event at all; routing shift only
 □ LATENCY WINS: users served from nearby region —
   performance benefit paying for DR complexity daily
 □ NO IDLE STANDBY: all capacity does useful work
 □ REGIONAL MAINTENANCE: drain one region, others absorb;
   upgrades without capacity cliffs

The Consistency Bill

 two regions accept writes to the SAME record simultaneously:

 t=0  region-EU: set price = 10
 t=0  region-US: set price = 12     (both valid locally!)

 replication delivers both → conflict. now what?

 resolution strategies:
 LAST-WRITE-WINS:   timestamps decide; clock skew lies possible
                    simple; silent lost updates
 APPLICATION LOGIC: domain rules ("lowest price wins"? 
                    highest bid?) — must be deterministic everywhere
 CONFLICT-FREE TYPES (CRDTs): merges mathematically
                    (counters, sets); limited shapes
 AVOIDANCE:         partition data so only ONE region writes
                    each record (user-home-region model) ← 
                    the pragmatic dominant pattern!

 avoidance deserves emphasis: most systems don't need
 any-record-anywhere writes. route each user's WRITES home,
 replicate globally for reads. conflicts vanish by construction.

The Write-Routing Pattern (pragmatic AA)

 USER-HOME-REGION model:

 user profile:    home = EU → EU owns its writes
                  reads anywhere (local replicas)
 cross-region actions (transfer EU→US user):
                  saga/event through the bus; eventual per usual

 [EU]──owns: EU-users' writes──►replicate──►[US][APAC read-only]
 [US]──owns: US-users' writes──────────────►...

 benefits: zero write-conflicts + locality latency +
 true AA resilience. costs: cross-region interactions
 become async workflows (sagas lessons apply).

Requirements Checklist

Requirement	Consequence if skipped
Deterministic conflict resolution	divergent regional state
Idempotent replication application	duplicate effects
Global uniqueness coordination (IDs)	collisions across regions
Replication-lag-aware UX	confused users seeing old data
Per-region independence drills	hidden shared dependencies

 global IDs matter early: UUIDs or region-prefixed keys
 from day one; retrofitting uniqueness across regions hurts.

Interview Framing

“Global social platform: design for region loss with no meaningful downtime” scored shape: active-active topology drawn, WRITE-HOME-REGION pattern chosen explicitly (conflict avoidance over resolution!), replication + lag-aware UX noted, global-ID/uniqueness checklist item named, RTO-statement (“routing shift, seconds”). Choosing avoidance-and-explaining-why beats heroically solving conflicts you engineered away — that’s the seniority signal here.