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
- Break egress into cross-zone, cross-region, internet, customer export, provider, and observability paths. Identify which flows serve resilience, residency, portability, or accidental topology.
Establish scale assumptions
- Build a byte-flow map by source, destination, purpose, and business unit. Model peak, compression, cache hit, replay, and migration volume rather than invoice totals alone.
Functional and non-functional requirements
- Reduce unnecessary transfer, retain customer export, meet recovery and residency, avoid hidden correctness changes, and keep strategic exit feasible. Owners see cost near architecture decisions.
High-level architecture
- Use locality-aware routing, data-affinity placement, batch and compression, CDN or regional caches, payload-thin events, and explicit replication policies. Maintain portable archive exports and tested incremental transfer.
Data model and flow
- Tag large flows with tenant, purpose, source and destination region, dataset, and operation. A cost ledger links bytes to journeys and detects unexpected topology changes.
Consistency and transaction boundaries
- Reducing replication changes staleness and failover guarantees; document per-data-class topology. Caches and projections have invalidation and recovery semantics before traffic is localized.
Failure modes and recovery
- Do not eliminate cross-zone redundancy that supports the SLO without a risk decision. Recovery plans stage incremental data continuously rather than assuming emergency bulk export.
Security and privacy
- Encrypt exports, authorize destination and purpose, enforce residency, and prevent cost optimization from routing sensitive data through unapproved intermediaries.
Observability and SLOs
- Track bytes, cost, cache and compression savings, replication lag, flow purpose, customer export SLO, and unknown transfer. Alert on topology-driven step changes.
Capacity and cost
- Prioritize chatty architecture and redundant copies before negotiating rates. Compare storage duplication against repeated egress and consider provider private pricing without losing portability evidence.
Alternatives and trade-offs
- Local copies reduce repeated egress and latency but increase storage and deletion complexity. Choose using access frequency, consistency, and lifecycle, not a blanket centralization rule.
Evolution and migration
- Instrument flows, fix the top few causes, add locality to service placement, validate recovery, and maintain periodic export drills. Renegotiate after usage architecture stabilizes.
What Staff and Principal candidates should emphasize
- Principal candidates treat the bill as a graph of architectural choices. They preserve customer rights and disaster options while challenging theatrical multi-cloud recovery.
Decision trade-offs
Data placement
Option A
Central authoritative dataset with remote access
Option B
Regional copies near repeated consumers
Recommendation:Replicate when repeated transfer and latency exceed storage and lifecycle cost and the consistency contract is acceptable.
Portability
Option A
Emergency full export during provider failure
Option B
Continuous portable archives and incremental drills
Recommendation:Maintain incremental portable state and tested restore; petabyte emergency export is not a credible recovery plan.
Follow-up interview questions
- 01When is cross-zone traffic worth its cost?
- 02How does caching affect deletion and consistency?
- 03What proves the company can exit the provider?
- 04Should customer exports be priced separately?
Common weak answers and mistakes
- 01Treating all egress as waste and removing resilience traffic.
- 02Buying another copy in every region without lifecycle cost.
- 03Claiming multi-cloud portability with no continuous data path.
- 04Optimizing flows without tenant, purpose, and residency context.
Interviewer evaluation rubric
Negotiates rates or compresses traffic without mapping purpose, topology, consistency, and recovery.
Builds a byte-flow map, localizes chatty paths, uses caching and compression, and protects replication guarantees.
Adds portable archives, export drills, deletion semantics, unit ownership, and data-class placement contracts.
Uses egress economics to redesign data gravity and vendor strategy while improving rather than weakening recoverability.