Block storage availability across AZ failure

WHAT THIS TESTS The interviewer wants HA reasoning grounded in the fact that block volumes are zonal, plus a clear articulation of RTO and RPO trade-offs.

A GOOD ANSWER COVERS Start from the constraint: a block volume is bound to a single availability zone, so if that AZ fails the volume and any instance using it become unreachable, and the data is not automatically available elsewhere. Lay out the spectrum. Periodic snapshots to object storage are cheap and durable across zones, but recovery means restoring a volume in another AZ, giving high RTO and an RPO equal to the snapshot interval, so you can lose recent writes. Synchronous block-level replication, or a regional persistent disk that mirrors across two zones, gives near-zero RPO and faster failover at higher cost and some latency. The strongest option is application-level replication, such as a database with a synchronous standby in another AZ that promotes on failure, delivering near-zero RTO and RPO because the application, not the storage layer, owns failover. Match the choice to how much data loss and downtime the business tolerates and what it will pay.

COMMON WRONG ANSWERS Assuming a single block volume is inherently multi-AZ durable. Relying only on infrequent snapshots for a system that cannot lose recent data. Conflating RTO and RPO. Choosing the most expensive synchronous option when the workload tolerates minutes of loss, or vice versa.

LIKELY FOLLOW-UPS Define RTO versus RPO precisely. How does a synchronous standby achieve low RPO. What latency cost does cross-AZ synchronous replication add. How do you test failover regularly so it works when needed.

ONE CONCRETE EXAMPLE A payments database cannot lose committed transactions, so it runs a primary in one AZ with a synchronous standby in a second AZ; on AZ failure the standby promotes in under a minute, giving near-zero RTO and RPO. A separate analytics store, where losing an hour is acceptable, relies on hourly snapshots restored into another AZ on failure, accepting higher RTO and an hour of RPO in exchange for far lower cost, deliberately matching each system's recovery targets to its business value.