Glosarium
Contents

Architecture & System Design › System Design Fundamentals

Designing a News Feed

Fan-out on write vs on read, a classic design problem.

Backend Engineer · Senior

Also known as: news feed design, feed ranking, social feed

Designing a news feed assembles each user’s personalised stream from followed sources: fan-out on write (push posts into followers’ feeds) versus fan-in on read (pull and merge at request time), plus ranking, pagination and freshness. The choice hinges on the follow graph’s shape — and celebrities break naive designs.

push:  post → write to every follower's feed (fast reads, heavy writes)
pull:  request → gather followees' posts → merge+rank (slow reads, cheap writes)
hybrid: push to normal followers, pull celebrities at read time

Ranking (recency × affinity × quality) usually runs over candidate generation (recent posts from follows) with pagination by cursor. Freshness, dedup (edits/deletes propagate), and abuse (spam ranking) complete the picture.

The classic mistakes:

  • Pure push with celebrities. A 100M-follower post means 100M writes per post — infeasible. Hybrid fan-out (push the many, pull the few huge) is the standard answer.
  • Pure pull at scale. Following thousands means thousands of lookups per refresh; latency collapses. Precompute for the common case.
  • Offset pagination. Page 500 of a shifting feed duplicates and skips; cursor-based pagination stays stable as new posts arrive.
  • Ranking as an afterthought. Chronological is a valid choice, but “we’ll add ranking later” ignores that ranking shapes storage (features, precomputation). Decide the feed’s intelligence early.
  • Ignoring deletes/edits. Removed posts lingering in precomputed feeds need invalidation or tombstones; plan the propagation path.
  • No abuse story. Spam, reshares and engagement bait dominate unguarded feeds. Ranking must include integrity signals from day one.
  • Cold start emptiness. New users with no follows see nothing; onboarding (suggested follows, global highlights) is part of the design.

Why it teaches: fan-out asymmetry, push-vs-pull, ranking pipelines and celebrity hotspots — the feed is distributed systems wearing a social mask. Hybridise around the graph’s skew.