Interview prompt
Problem context
Skills being evaluated
Use the sequence below to surface constraints, choose boundaries, test failure behavior, and defend trade-offs. Concrete numbers are interview assumptions, not claims about a real production system.
Clarify the decision
- Classify content, identity, metadata, telemetry, and derived features separately; residency obligations rarely apply identically. Ask whether guest access allows remote processing, what support access is permitted, and which countries form approved zones.
Establish scale assumptions
- Assume tens of thousands of tenants, a small regulated subset, and cross-zone latency above 100 milliseconds. Measure guest collaboration and search patterns before building global copies to optimize rare access.
Functional and non-functional requirements
- Policy determines storage, processing, replication, backup, support, and deletion locations for every data class. The product must disclose degraded cross-zone features and produce evidence of placement and access.
High-level architecture
- Assign each tenant a policy and home zone in a strongly controlled directory. Route writes and regulated processing to zonal data planes; expose cross-zone collaboration through scoped gateways that return authorized views without creating ungoverned replicas.
Data model and flow
- Every record and event carries tenant, data classification, policy version, and allowed zones. Downstream consumers enforce a policy decision before persistence; denied derivations produce a visible capability status rather than a silent missing feature.
Consistency and transaction boundaries
- The tenant home zone owns writes. Remote guests read through that authority or approved caches with bounded metadata; cross-tenant objects need an explicit ownership and copy contract instead of inheriting one collaborator’s policy accidentally.
Failure modes and recovery
- If policy lookup is unavailable, continue only with a cached signed policy within its validity window and fail closed on new exports. Regional failover must target an approved zone, which may mean lower availability for tightly constrained tenants.
Security and privacy
- Use zonal encryption keys, workload identity, tenant-scoped access, and purpose-bound support elevation. Prevent sensitive payloads from leaking through logs, traces, queue dead letters, model prompts, or globally aggregated analytics.
Observability and SLOs
- Continuously inventory where each data class exists, log policy decisions and access, alert on forbidden-zone writes, and test deletion propagation. Provide tenant-facing audit evidence without revealing internal topology.
Capacity and cost
- Residency reduces pooling and increases standby and operational cost. Price or constrain advanced locality requirements explicitly, and keep shared global services payload-blind where possible.
Alternatives and trade-offs
- Remote access to the home zone preserves policy but adds latency; approved replicated views improve experience but expand compliance scope. Choose per feature and make the trade visible in product commitments.
Evolution and migration
- Start with data inventory and policy tagging, block new unclassified sinks, then migrate regulated tenants by domain. Run policy in audit mode before enforcement and maintain a tenant-specific rollback that never moves data to a forbidden zone.
What Staff and Principal candidates should emphasize
- Principal answers follow data beyond the primary database into backups, logs, support, and ML. They connect policy versions to runtime enforcement and admit when residency constraints reduce failover options or product functionality.
Decision trade-offs
Cross-zone collaboration
Option A
Remote reads from tenant home zone
Option B
Replicated approved collaboration views
Recommendation:Default to remote access for strict data classes; replicate only fields and zones explicitly covered by tenant policy and product consent.
Failover
Option A
Any healthy global region
Option B
Only preapproved residency zones
Recommendation:Honor approved zones even when it lowers theoretical availability; market and engineer the correct region-specific SLA.
Follow-up interview questions
- 01Where may encrypted backups live if keys remain in the tenant’s zone?
- 02How do you handle a shared object owned by two tenants with incompatible policies?
- 03Can global fraud detection consume regulated event data?
- 04What happens when a tenant changes residency policy after years of use?
Common weak answers and mistakes
- 01Applying residency only to the primary database and ignoring logs, backups, indexes, and ML.
- 02Allowing global failover that violates the contract during the exact event auditors examine.
- 03Treating tenant home region as an unversioned string without policy history.
- 04Promising identical features despite prohibited cross-zone processing.
Interviewer evaluation rubric
Places the database in Europe but leaves derived data, support access, and disaster recovery unspecified.
Uses tenant home zones, policy-aware routing, zonal keys, and approved failover with a data inventory.
Carries classification and policy version through all sinks, governs collaboration, and continuously proves placement and deletion.
Turns legal constraints into explicit product semantics, cost tiers, and auditable runtime decisions across shared and tenant-owned data.