# Common Interview and Exam Questions — Multi-Tenant SaaS Architecture

Source: https://www.skillbyai.com/en/multi-tenant-saas/mt-interview-questions

> Practise concise, structured answers to the multi-tenancy questions that appear most often in system design interviews and exams.

## Design and isolation questions

**"Design a multi-tenant SaaS for invoicing."** Start with tenants, users and memberships; choose pool with `tenant_id` everywhere for small businesses, and a catalogue allowing silo placement for large ones; resolve tenant from subdomain plus token; describe RLS, scoped repositories and isolation tests; finish with sharding by tenant and noisy-neighbour limits. **"Silo vs pool?"** Compare isolation, cost per tenant, operational effort, blast radius and compliance, then say most products mix them per layer and per tier. **"How do you stop data leaks between tenants?"** Layered: edge resolution and membership checks, request context, scoped data access, database policies, tenant-prefixed caches and storage, tests that assert cross-tenant reads fail, and audit logs. **"How do you shard?"** Tenant as shard key, directory placement, single-shard queries, a separate warehouse for cross-tenant analytics, and an online move procedure for rebalancing.

## Draw the request path

In a whiteboard interview, sketch one request end to end: host and token, tenant resolution, context, repository, RLS-protected table, cache key, log line. Pointing at where the tenant is checked at each hop is the clearest way to show you understand isolation.

## Operations and business questions

**"How do you migrate schemas for 2,000 tenant databases?"** Expand-and-contract so old and new code both work, versioned migrations recorded per tenant in the catalogue, canary then waves, halt on failure, resume where it stopped. **"One customer is slowing everyone down. What do you do?"** Confirm with per-tenant metrics, apply rate limits or fair queuing immediately, investigate the workload, then consider quotas, a higher plan or moving them to a dedicated shard. **"How do you delete a customer's data?"** Suspension and grace period, export, deletion across every store from a data map, backups handled by retention or crypto-shredding, and a record of completion. **"How do you bill on usage?"** Durable usage events with idempotency keys, per-tenant aggregation per period, raw events kept for disputes, reconciliation against independent measures.

## Like a viva

A system design interview is closer to a college viva than a written exam. The examiner cares less about one perfect answer and more about whether you can explain why you chose it and what you would do when they change a requirement.

**Quiz:** An interviewer asks: "Your pooled product just signed a bank that requires dedicated storage. What changes?" Which answer is strongest?

- [ ] Fork the codebase for the bank
- [ ] Refuse, because pooled products cannot host such customers
- [ ] Move every tenant to dedicated databases
- [x] Provision a dedicated database for that tenant, record it in the catalogue, and keep the same application code routing by lookup

*Answer:* Provision a dedicated database for that tenant, record it in the catalogue, and keep the same application code routing by lookup. Tier-based placement through a catalogue lets one tenant be siloed without a fork and without changing everyone else's placement.