System Design interviews

Work in the tool while you explain your decisions.

Choose a role and level. Each interview has a new task, follow-up questions and a report that checks your work and your explanation.

Take the assessment Choose an interview

Speak or type. Sign in to save your results across devices.

Timed assessment

12 practical questions
35 minutes
Mid level

Questions are drawn from a bank of 36. Your report includes a skill breakdown and lessons to practice next.

Start assessment

Practice interviews

Junior tasks cover the fundamentals. Mid and senior tasks add debugging, changing requirements and review decisions.

Position

10 interviews

Backend engineer
Junior
50 min

An expiring link service

Harbor Links lets event organizers create short links that stop redirecting at an expiry time. Design creation and redirects for 8.64 million redirects and 86,400 new links per day, a peak ten times average, and 180-day retention. Custom domains and detailed analytics are out of scope.

Start interview

Skills and rubricHarbor redirect path

Trace creation and a cache miss to a durable mapping keyed by code.
Calculate 100 average and 1,000 peak redirects per second from the supplied traffic.
Reject an expired mapping before redirecting and preserve an existing code on collision.

Backend engineer
Junior
25 min

A cursor-based staff directory

Larch Cooperative needs an internal directory. People can share a display name and new records arrive while someone pages through results. There may be no matches. Define a stable read contract without exposing a list offset as an identity.

Start interview

Skills and rubricDirectory contract

State the contract and trace one operation.
Use stable identity and a deterministic cursor; explain duplicates, ties and empty pages.
Test an empty, duplicate or failed request.

Backend engineer
Junior
25 min

A retried delivery webhook

Sable Dispatch receives delivery webhooks from a partner. A lost response makes the partner send the same event twice. Route accepted work to a worker without holding the incoming request open while the downstream partner is slow.

Start interview

Skills and rubricWebhook processing boundary

State the contract and trace one operation.
Durable acceptance precedes acknowledgment; deduplication and downstream idempotency address the ambiguous outcome.
Test an empty, duplicate or failed request.

Backend engineer
Mid
55 min

A ticket hold service

Cedar Hall sells 12,000 assigned seats to as many as 4,000 simultaneous visitors. Design five-minute seat holds followed by payment confirmation, with one confirmed owner per seat. Payment callbacks can arrive twice or out of order, and a successful late charge must have a clear resolution.

Start interview

Skills and rubricConflicting holds

Commit idempotency identities with seat transitions and use the database clock for expiry.
Keep provider calls outside the seat transaction and make callbacks safe under concurrent duplicate delivery.
Trace reconciliation and refund after a late charge without taking another buyer's seat.

Site reliability engineer
Mid
25 min

A stale inventory cache

Birch Outfitters has a finished inventory read design, but sold-out products remain available in cached reads. Review the starting board. Its ttl-only policy allows stale inventory for the entire TTL. Purchases still use the durable stock transaction.

Start interview

Skills and rubricInventory cache repair

Invalidation narrows stale reads; it never authorizes a purchase. Bound fallback and protect the source.
Explain a concrete alternative and why the stated constraint rules it out.
A cache miss races a stock update and writes old data back. How would you close that race?

Backend engineer
Mid
25 min

A preview job with a changed deadline

Fern Studio initially asks for an uploaded document preview in the request response. During the interview, document conversion becomes slow and variable. Revise the contract so the caller can track a durable job while a worker builds the preview.

Start interview

Skills and rubricAsynchronous preview contract

Separate durable job acceptance from execution, and preserve a queryable status for retries and failures.
Explain a concrete alternative and why the stated constraint rules it out.
The client times out after the job is accepted. How does it retry without two conversions?

Data engineer
Mid
25 min

An import with missing business keys

Reed Fieldwork imports partner exports. Several rows have NULL event IDs, some IDs repeat across files, and export timestamps mix time zones. The analytics owner needs accepted rows preserved and bad records available for investigation, without silently counting them as new events.

Start interview

Skills and rubricImport and quarantine

Retain raw inputs, define source identity, quarantine invalid keys and reconcile accepted, rejected and duplicate rows.
Explain a concrete alternative and why the stated constraint rules it out.
The same event ID appears with a corrected timestamp. What determines replacement versus rejection?

Data engineer
Senior
55 min

A reliable event pipeline

Morrow Learning needs hourly lesson-completion metrics from browser events, with a dashboard less than ten minutes behind and corrected daily totals. Design for 24 million events a day, peaks five times average, and events arriving late or twice. Keep 30 days of replayable input and support deletion from retained and derived data.

Start interview

Skills and rubricReplayable completion pipeline

Define the durable acknowledgement, checkpoint and publication boundaries under a crash at each handoff.
Coordinate live data and replay by cutoff or business key and reapply deletions during replay.
Separate producer silence, event-time lag and processing lag in alerts and name the condition that requires a wider rebuild.

Site reliability engineer
Senior
35 min

A noisy tenant and a global quota

Quill Metrics serves independent tenants through two request frontends. A single tenant can exhaust the service. A local counter on each frontend doubles the intended quota. Review the shared enforcement boundary, then explain what happens when that boundary fails.

Start interview

Skills and rubricTenant quota boundary

Use an atomic shared boundary, preserve tenant isolation and explicitly choose behavior when the quota dependency is unavailable.
A large tenant sends a burst while a small tenant is idle. Explain isolation and a measured overload limit.
Define an operational owner and a condition that changes your recommendation.

Site reliability engineer
Senior
35 min

A recovery plan with an untested restore

Aspen Records must not lose an acknowledged write after a single storage-node failure. Its cold restore drill needs eight minutes to fetch the snapshot and four minutes to replay the log. Review the replication and archive paths, calculate cold recovery and distinguish it from the normal replica failover.

Start interview

Skills and rubricRecovery contract review

Distinguish recovery time from data-loss tolerance, test application consistency and state the failure scope of replication.
A regional incident also removes the replica. Separate that failure promise from the single-node guarantee.
Define an operational owner and a condition that changes your recommendation.

Choose your character

System Design interviews

Timed assessment

Practice interviews

An expiring link service

A cursor-based staff directory

A retried delivery webhook

A ticket hold service

A stale inventory cache

A preview job with a changed deadline

An import with missing business keys

A reliable event pipeline

A noisy tenant and a global quota

A recovery plan with an untested restore

Interview Hub access

Your Work Gym account