System Design Interviews in Canada: Method and Local Rules

A timed system design method for Canada, with in-Canada replication sized on the board, OSFI B-13 recovery, PIPEDA breach records and Quebec Law 25.

By · · 7 min read

Illustration for the System Design Interview Framework guide

Canada edition. Written for Canadian interviews: OSFI Guideline B-13 for banks and insurers, PIPEDA breach reporting and two-year breach records, Quebec Law 25 transfers outside the province, English and French content, and the two Canadian cloud regions. US edition

System design interviews in Toronto, Montreal, Vancouver or Waterloo follow the familiar four steps: scope and estimate, sketch the main path, go deep, then scale and trade-offs. The Canadian difference shows up in the requirements you should raise before drawing. A bank or insurer works under OSFI's technology and cyber risk guideline; a private-sector product with personal information generally comes under PIPEDA's breach rules; anything serving Quebec brings in Law 25 and, usually, French. This edition turns those into numbers and design choices you can defend, using the prompts in the System Design Interview Framework (Canada).

The method, timed

StepMinutesOutput
Requirements and estimates8 to 10Features, non-goals, QPS, storage, languages, where users and data are
High-level design10 to 15Main flow, APIs, data model
Detailed design15 to 20One or two components: schema, partitioning, failure handling
Scale, recovery and trade-offs8 to 10Bottlenecks, recovery time and recovery point, what you chose not to do

Three questions to ask in the first five minutes of a Canadian prompt: Is the employer federally regulated? Where are the users, and does that include Quebec? Must data stay in Canada?

OSFI B-13: recovery is a requirement

OSFI's Guideline B-13 on technology and cyber risk management applies to all federally regulated financial institutions, and OSFI's release letter set its effective date as January 1, 2024. Two expectations translate straight into a design:

  • institutions should have plans and capabilities to recover technology services to an acceptable level within an acceptable timeframe, which the institution itself defines (section 2.9.1);
  • they should regularly validate and report on their disaster recovery strategies, plans and capabilities against severe but plausible scenarios (section 2.9.3).

Because B-13 leaves the timeframe to the institution, the recovery time and recovery point you propose are part of your answer. Give numbers and say what they cost.

Worked example: a ledger replicated inside Canada

Prompt. Design the core ledger for a Canadian bank's chequing accounts, with disaster recovery.

Requirements. Post debits and credits, show balances, never lose a committed transaction in a single-zone failure, recover from a regional failure within an agreed time. Data stays in Canada.

Regions. The AWS regions list shows two Canadian regions, Canada (Central) ca-central-1 and Canada West (Calgary) ca-west-1, each with three availability zones; Calgary is opt-in. A design can replicate between them without leaving the country.

Estimates (premises to agree):

ledger change volume:   2 TB a day
average replication:    2 TB x 8 bits / 86,400 s ≈ 185 Mbit/s
on a 1 Gbit/s link:     16,000 s of transfer, about 4.4 hours a day busy

Design. Commit synchronously across the three zones of Canada (Central), so a zone failure loses nothing. Replicate asynchronously to Calgary. Then put the B-13 questions on the board:

  • Recovery point. Asynchronous replication means the Calgary copy lags. If replication falls 10 minutes behind at the busiest hour, a regional failover loses up to 10 minutes of posted transactions unless you can recover them from the payment rails or an upstream log. Size the link for the busiest hour, not the 185 Mbit/s average, and alert on lag.
  • Recovery time. How long does failover take, who decides to trigger it, and how do you stop both regions accepting writes at once?
  • Testing. B-13 asks for validation against severe but plausible scenarios, so describe a failover drill, not just the architecture.

Personal information: PIPEDA breach records

The Office of the Privacy Commissioner's guidance on what you need to know about mandatory reporting of breaches says organizations must report breaches of security safeguards involving personal information that pose a real risk of significant harm, notify the individuals concerned as soon as feasible, and keep records of all breaches for two years, whether or not they meet the harm threshold.

That last point is a data-model item. Add a breach record store with retention of at least two years, and make audit logs queryable by person so you can work out quickly whose data was involved.

Quebec: Law 25 and French

Quebec's Commission d'accès à l'information summarizes the main changes under Law 25: since September 2023, an organization must carry out a privacy impact assessment before communicating personal information outside Quebec; the communication may go ahead only if the assessment shows the information would be adequately protected, and it must be covered by a written agreement. Organizations must also notify the Commission and the people concerned of a confidentiality incident that presents a risk of serious harm, and keep a register of incidents.

So "put the Quebec customers in a US region to save money" is not a free choice. Nor is a Canadian region automatically the answer. Law 25 is about leaving Quebec, not leaving Canada: the rule covers communication outside Quebec, so replicating Quebec customers' data to Calgary also needs the privacy impact assessment and the written agreement. Ask where Quebec data may go before choosing the recovery region.

Language is the other Quebec requirement. A bilingual product affects the content model (two fields or two records per item), search (an analyzer per language for stemming, accents and stop words), templates and testing. Size it: 50 million documents x 2 languages x 2 KB is about 200 GB of index, double a single-language build. It costs little to raise in requirements and a lot to retrofit.

Building blocks to explain in two sentences

  • Synchronous within a region, asynchronous between regions. The usual compromise between durability and the latency of a cross-country round trip.
  • Idempotency. Payments and ledger postings are retried; effects must happen once.
  • Outbox pattern. Commit the business change and the outgoing message together, publish later.
  • Read replicas and replication lag. Read-your-own-writes needs routing to the primary or a session token.
  • CAP and PACELC. In a partition you choose consistency or availability; otherwise latency against consistency.

Common mistakes in Canadian rounds

  • Treating disaster recovery as "add a US region" without asking whether data may leave Canada.
  • Forgetting that PIPEDA requires records of every breach, not only the reportable ones.
  • Designing English-only and adding French later.
  • Giving no recovery time or recovery point when B-13 expects the institution to set them.

Tools for this job

Frequently asked questions

What is OSFI Guideline B-13?

OSFI's guideline on technology and cyber risk management for federally regulated financial institutions, effective January 1, 2024. It expects technology services to be recoverable within a timeframe the institution defines, and disaster recovery to be validated against severe but plausible scenarios.

Can a Canadian system keep primary and backup data in Canada?

Yes. AWS lists Canada (Central) and Canada West (Calgary), each with three availability zones, so primary and disaster recovery copies can both stay in Canada.

How does Quebec Law 25 affect a design?

Before personal information leaves Quebec, a privacy impact assessment and a written agreement are required, and the transfer may proceed only if the information would be adequately protected. The rule is about leaving Quebec, not leaving Canada, so a recovery copy in Calgary needs the same assessment and agreement as one in a US region. Ask where Quebec data may go before choosing regions, and treat vendor location as a requirement to ask about.

When must a breach be reported under PIPEDA?

When it is reasonable to believe it poses a real risk of significant harm. The organization reports it to the Privacy Commissioner, notifies affected individuals as soon as feasible, and keeps records of all breaches for two years.

Sources and further reading

Open the System Design Interview Framework

More Interview Prep

Every tool in the Interview Prep section, free and in your browser.