Spotify System Design Interview Questions
Spotify system design rounds ask you to design parts of a music service. The most common question is some version of "design Spotify". After that come recommendations, search, and playback across devices.
Spotify serves a very large audio catalog in many countries. Almost every screen is personalized. Your answer should show that you know this shape.
The Questions Candidates Report
Spotify does not publish a question list. Candidates report the questions below. Each one maps to a system Spotify really runs.
| Question | What it tests |
|---|---|
| Design a music streaming service | Audio delivery, caching, catalog, licensing rules |
| Design the recommendation home feed | Offline training, online serving, ranking |
| Design playlist storage and sharing | Write patterns, ordering, share links |
| Design search over songs and artists | Indexing, typos, ranking |
| Design playback across devices | Session state, device handoff, low latency |
| Design the play and skip event pipeline | Ingest, ordering, correct counting |
| Design podcast hosting and ad insertion | Large files, transcoding, ad stitching |
| Design the yearly listening summary | Batch aggregation over very large logs |
Spotify Systems You Can Name
Spotify engineers have described several of these systems in public talks and blog posts. Naming them shows real interest.
- Audio delivery. A track is encoded once, cached close to listeners, then sent in small pieces.
- Event delivery. Spotify has written about the pipeline that carries play events into storage for analytics. It has publicly described running that pipeline on Google Cloud Pub/Sub and Dataflow.
- Recommendations. Spotify has published work on personalized playlists and on ranking models for the home page.
- Similarity search. Spotify open sourced Annoy, a library for approximate nearest neighbor lookup.
- Developer portal. Spotify open sourced Backstage, the portal its own teams use to manage services.
How to Answer "Design Spotify"
Use the same order every time. It keeps a 45 minute round under control.
- Agree on scope. Play a track, browse a catalog, build playlists. Say what you are leaving out.
- Write the read and write paths. Reads are huge. Writes are small and rare.
- Size it with your own numbers. Spotify publishes user and catalog figures in its investor reports. Do not quote a figure you are unsure of. State an assumption and say it is an assumption.
- Design audio delivery first. Encode once into a few bitrates. Store the files in object storage. Cache them at the edge. Stream in small pieces so seeking is cheap.
- Split metadata from media. Track, album and artist records are small and read constantly. Put them behind a cache.
- Add the event pipeline. Every play, skip and like becomes an event. Send it to a queue, then to a stream job, then to a warehouse.
- Close with failure cases. A dead cache region, a slow database, a duplicate event.
Recommendations, the Part People Skip
Most candidates design streaming well and then rush recommendations. Spotify cares about both.
Split the work in two. An offline job trains models and writes candidate lists for each user. An online service reads those lists and ranks them at request time.
Keep the online path fast. It should read precomputed data, not train anything.
Say how you refresh. A weekly personalized playlist can be a batch job. A home page that reacts to the last song needs a fresh signal.
Mistakes That Cost Points
- Designing a video service. Audio files are far smaller than video. Do not copy a Netflix answer.
- Ignoring licensing. A track may be playable in one country and blocked in another. Rights checks belong in the catalog layer.
- Counting plays with one database counter. Millions of writes per second need a stream job, not a row update.
- Forgetting offline mode. Downloaded tracks need encrypted local storage and a later sync.
- Skipping the skip. Skips are the strongest feedback signal Spotify has. Show where they land.
How to Prepare
- Learn the building blocks first. Caching, sharding, queues and replication appear in every answer. Grokking System Design Fundamentals covers them in order.
- Practice full designs out loud. Grokking the System Design Interview walks through large services step by step, including media delivery and search.
- Compare across companies. The same core pieces appear in the Netflix system design questions and the Uber system design questions.
- Read how Spotify is put together. Our page on Spotify system architecture gives you names you can use in the round.
- Prepare the story rounds too. Spotify weighs collaboration heavily. See Spotify behavioral interview questions.
- Rehearse with a timer. Book a mock interview and finish a full design in 45 minutes.
GET YOUR FREE
Coding Questions Catalog
$99
$197
$72
