Unaundaje multi-region DB kwa HA na DR?

Question

IT Interview Practice · Accepted Answer

Replicate database kuvuka regions na chagua topology yako kwa mahitaji: **active-passive** (writer mmoja, warm standby inayochukua kwenye failover) kwa urahisi na strong consistency, au **active-active** (writers wengi) kwa low write latency kwa gharama ya conflict resolution — kisha chagua **sync vs async replication** kuweka maelewano kati ya durability na latency, na fafanua **RPO/RTO** targets za wazi zinazoendesha failover.

```text
        ┌──────────── Region A (primary) ────────────┐
writes ─▶ app ─▶ DB primary ──┐
                              │ replication
        ┌─────────────────────┼── async (RPO>0) ─────┐
        │                     └── sync (RPO=0, slow)  │
        ▼                                             ▼
   Region B standby                            (failover promotes standby)
        │
   health check ──▶ on primary loss: promote + repoint DNS/traffic
```

## Active-passive vs active-active

**Active-passive** ina primary moja inayoshughulikia writes na standbys zinazosoma tu; kwenye failure **unapromote** standby. Ni rahisi na huepuka write conflicts, lakini passive capacity hukaa bila kazi na failover ina pengo. **Active-active** huruhusu kila region kukubali writes kwa local latency na hakuna failover promotion — lakini regions mbili zinaweza kuedit row ile ile kwa pamoja, hivyo *lazima* utatue conflicts (last-write-wins, CRDTs, au app-level merge). Timu nyingi huanza active-passive; active-active ni kwa global write latency usiyoweza kufikia vinginevyo kweli.

## Sync vs async replication

Hii ni durability/latency trade ya msingi. **Synchronous:** primary husubiri remote replica i-acknowledge kabla ya commit — **RPO = 0** (hakuna data loss kwenye failover) lakini kila write hulipa cross-region round-trip latency (~makumi ya ms), na ikiwa replica haifikiki, writes husimama. **Asynchronous:** primary hucommit locally na kusafirisha mabadiliko baadaye — writes za haraka, lakini kupoteza region kunaweza kupoteza sekunde chache za mwisho (**RPO > 0**). Cross-region sync mara nyingi ni polepole mno, hivyo wengi huendesha async na kukubali RPO ndogo.

## Failover, RPO na RTO

- **RPO (Recovery Point Objective):** kiasi cha data unachoweza kupoteza — kimewekwa na sync vs async.
- **RTO (Recovery Time Objective):** muda hadi urudi — umewekwa na jinsi unavyogundua failure na kupromote haraka.
- Failover = gundua (health checks) → promote standby → elekeza traffic tena (DNS/global LB). Iautomate, lakini jilinde dhidi ya **split-brain** (regions zote mbili zinadhani ni primary) kwa quorum/fencing token.

## Mtazamo wa mahojiano

Uchunguzi: je unaelewa kuwa HA/DR ni **maelewano kati ya consistency, latency, na data loss** — si "ongeza replica tu". Nzuri: active-passive na standby. Bora sana: unataja RPO/RTO, athari za sync-vs-async, na hatari ya split-brain. Mtego: kudai async hutoa zero data loss, au kudhani failover ni papo hapo na haina hasara.

## Maelewano na aina za hitilafu

- **Split-brain:** network partition hupromote primaries mbili → data zinazotofautiana. Tumia quorum, fencing, au witness region.
- **Failback ni ngumu:** baada ya primary ya zamani kurudi, kulinganisha writes zinazotofautiana ni kwa mkono na kwa makosa.
- **Gharama:** multi-region halisi huongeza infra takribani mara mbili na huongeza bili za cross-region egress bandwidth.
- **Failover isiyojaribiwa hushindwa:** DR ambayo haijawahi kufanyiwa drill haitafanya kazi kwenye outage halisi — endesha game-days.

Active-passive vs active-active

Sync vs async replication

Failover, RPO na RTO

Mtazamo wa mahojiano

Maelewano na aina za hitilafu

Maswali yanayohusiana