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 data by access pattern, durability, recovery priority, retention, legal hold, residency, sensitivity, and recomputability. Age alone is not a sufficient lifecycle signal.
Establish scale assumptions
- Measure bytes by class, growth, retrieval frequency and size, restore deadlines, minimum billing periods, and egress. Model rare but large legal and recovery retrievals.
Functional and non-functional requirements
- Automate movement and expiry, preserve legal and recovery objectives, make retrieval time visible, prevent resurrection, and prove policy application. Holds override normal deletion narrowly.
High-level architecture
- Use a policy catalog, immutable lifecycle metadata, tiered object storage, searchable manifests, key management, legal-hold service, and restore workflows. Separate index retention from payload retention.
Data model and flow
- Objects carry class, tenant, region, creation, access summary, policy version, hold state, key version, and integrity hash. Movers perform conditional transitions and record evidence.
Consistency and transaction boundaries
- Policy version at placement explains why an object exists; later policy changes trigger explicit reevaluation. Deletion tombstones and key erasure prevent restored catalogs from republishing expired data.
Failure modes and recovery
- Failed moves are resumable and leave one authoritative copy, integrity is verified after transition, and retrieval exercises catch inaccessible archives. Control-plane outage preserves last valid policy.
Security and privacy
- Encrypt tiers, isolate archive access, audit bulk retrieval, and keep residency during movement. Cold does not mean less sensitive or broadly accessible.
Observability and SLOs
- Track bytes and cost by class, transition backlog, retrieval time, restore success, integrity errors, holds, policy exceptions, and expired-but-present data.
Capacity and cost
- Include request, retrieval, minimum-duration, egress, index, and operational cost. Recompute derived data only if source and compute recovery meet the required RTO.
Alternatives and trade-offs
- Deep archive minimizes steady storage but can violate restore and investigation deadlines. Use recovery-tiered copies and retain compact indexes to find what needs retrieval.
Evolution and migration
- Inventory and classify, apply policy to new data, tier low-risk historical cohorts, exercise restore, then enforce deletion. Review classes with legal, security, and SRE owners.
What Staff and Principal candidates should emphasize
- Staff candidates connect lifecycle to recovery and deletion, not just price per gigabyte. They include manifests, holds, key policy, retrieval exercises, and full cost.
Decision trade-offs
Archive depth
Option A
Cheapest deep archive
Option B
Warmer tier that meets retrieval RTO
Recommendation:Choose the coldest tier whose measured restore and investigation time satisfies the data class objective.
Derived data
Option A
Retain every materialized copy
Option B
Delete and recompute from durable sources
Recommendation:Recompute when source retention, code version, capacity, and recovery time are all proven; otherwise keep the critical projection.
Follow-up interview questions
- 01How do legal holds interact with automated deletion?
- 02What if an archive restore exceeds RTO?
- 03Can derived indexes be deleted before payloads?
- 04How do old backups honor new retention policy?
Common weak answers and mistakes
- 01Tiering solely by object age.
- 02Ignoring retrieval, minimum-duration, and egress charges.
- 03Treating archived data as lower security risk.
- 04Deleting derived state without proving source and recompute recovery.
Interviewer evaluation rubric
Moves old data to cold storage without classification, restore, hold, deletion, or total-cost analysis.
Defines classes, lifecycle metadata, tiers, holds, integrity, retrieval workflows, and policy evidence.
Adds restore exercises, tombstone safety, residency, full pricing, recompute proof, and exception ownership.
Unifies finance, recovery, legal, privacy, and security into an auditable lifecycle architecture with measured trade-offs.