Interview practice

Home › Interview practice

Mock interviews

Distributed rate limiter: Limit every API key to its plan across 50 servers, at 200 000 checks a second, without slowing the API or taking it down.
Ticket booking: Sell 50 000 seats to a million people in a few minutes, never selling one seat twice, with holds that expire.
Search suggestions: Suggest the top five completions on every keystroke, from what readers searched for, in under 100 ms.
Nearby places: Find the places near a reader among 200 million, in dense cities and empty country alike, within 200 ms.

Problems to solve in the simulator

A Fast Home Page: The home page waits on a recommendations store that's sometimes very slow. Keep the page fast without dropping the box.
Attachments: Writers attach files to their notes. The database's network card fills up, and plain notes slow down with it. Give it the bandwidth it needs.
Cache the Hot Pages: A product catalog where a few pages get most views. Add a cache without serving old prices.
Charge Once: Some tips take longer to save than the app waits, and the retry charges the reader twice. Make a retry safe.
Chat: Design Inkwell Chat: held connections that survive a crash, and messages that are delivered in order, on time and never lost.
Circuit breakers: After a rush the servers' deep queues keep callers retrying into a backlog nobody is waiting for. Add a circuit breaker so the queues can drain.
Even Out the Exports: A few slow PDF exports drag every request down. Same servers, smarter balancing.
Event Stream: Everything readers do on Inkwell is an event. Three services read the stream, one of them joins late, and one writer makes a lot of noise.
Export in the Background: Turning a notebook into a PDF takes the better part of a second. Accept exports at once and render them on workers.
Fair Share: Keep one partner's runaway integration from taking Inkwell's API away from everyone else.
Feed for Everyone: Every new post lands in the home feed of each of its writer's followers. Keep the feeds fresh without paying for idle workers.
File Sharing: Design Inkwell Drive: share big files with readers on three continents, with uploads that never touch the app servers.
Find the Fault: The pager went off: one request in four fails, and the errors say different things. Read what the system tells you, then fix each cause.
Fresh Enough Prices: Prices change behind the cache's back. Pick a TTL that keeps them fresh without drowning the database.
Hello, Backend: Wire your first request path: client, app server, database.
The morning after the rush: Retries pile up behind requests nobody is waiting for.
Too many connections: A slow query holds connections until the database runs out.
Everything times out: A deep queue means almost every request waits past the timeout.
Blogs falls over at noon: A read-heavy blog outgrows one app server and one database node.
The 8:00 rush won't end: Callers retry straight into the rush they are part of.
Wrong prices in the shop: A cache with no expiry shows prices that changed long ago.
Keep Every Note: Make a saved note survive the database's primary dying, even the ones saved a moment before.
Ledger: Credit tips to writers' balances without losing any, even when two arrive together for the same writer.
Mint the Codes: Hand out short, unique codes from many app servers at once, fast enough for a bulk import.
News Feed: Design Inkwell Home: a fast, fresh feed for every reader, when most writers have a few followers and a few have thousands.
Notifications: Design Inkwell Notify: password resets that arrive in seconds, digests that can wait, and no one sent twice or never.
Out of Connections: Blog pages run a slow query, and the database turns them away with half its CPU idle. Give it the connections it needs.
Payments: Design Inkwell Pay: record a reader's tip once, never lose it, and ride out the database's primary dying.
Pictures Everywhere: A megabyte per picture, a thousand pictures a second, and a database that won't get bigger. Take the pictures off the servers.
Publish and Refresh: Make sure writers always see the posts they just published, without paying for strong reads nobody needs.
React to a Post: Every new post must be indexed for search, announced to followers and counted for analytics. Each subscriber has its own deadline, so size each one for its own.
Reactions: Design Inkwell Reactions: likes and counts that survive a post going viral.
Read the Archive: Years of old blog posts, each read about as often as the next. Scale the reads, and spend the money where it helps.
Ride Out the Rush: A daily rush overloads the servers for a few seconds. Make the app retry so the rush passes without errors or long waits.
Right-Size the Cache: A big catalog behind a slow database. Find the cache that makes pages fast for the least money.
Scale Reads with Replicas: A blog read far more than it is written. One database node can't keep up, and it can't get bigger.
Serve the World: Profile pictures for users on three continents. Bring them close with a CDN.
Session Store: Keep sessions readable through a replica outage, without ever returning an old one, by choosing replicas and quorums.
Shed the Load: All day there's more traffic than two servers can take. Make the ones who get in fast, and tell the rest right away.
Size Inkwell Groups: Turn the product team's numbers into servers: fast at the peak, without paying for idle machines.
Social Media Post Likes: Track likes for billions of posts under hot-key fan-out and bursty traffic.
Spread the Feed: Thousands of new posts a second and one viral post, on a store with small partitions.
Stay Connected: Keep thousands of readers connected to live comments through a gateway crash, without a stampede when they all come back.
Stay Up: Keep Inkwell Blogs serving when an app server dies and then the database's primary does.
Survive the Launch: Traffic quadruples at launch. Scale out, then find the next bottleneck.
URL Shortener: Mint short codes for long URLs and serve hot redirects under read-heavy load.