Fan-out on Write vs. Read

What problem are we solving?

Construct timelines from followed authors under a predictable read-latency budget.

How the options work

Option	Mechanics	Best fit
Fan-out on write	Publishing a post inserts its reference into many follower timeline indexes.	Ordinary authors with active followers and frequent feed reads.
Fan-out on read	Opening a timeline retrieves recent posts from followed authors and merges/ranks them.	Huge audiences, many inactive followers or low read frequency.

When is each option better?

Write fan-out improves common read latency when precomputation is reused. Read fan-out avoids enormous or unused write amplification. A hybrid pushes ordinary authors and pulls exceptional high-fan-out authors; its threshold should reflect active audience, post/read rates and measured cost.

Concrete example

Push a normal author to 200 active followers. Pull a celebrity’s last few posts into the reader’s candidate merge. Rank the merged set and paginate with a defined session/cursor policy.

Failure modes

Push workers may crash mid-batch or duplicate references; checkpoint ranges and enforce unique viewer/post IDs. Pull requests can fan out to too many authors and exceed deadlines; bound candidate retrieval. Neither mode removes read-time authorization or deletion filtering.

Scale implications

At 1,000 posts/second × 500 active followers, push adds 500,000 references/second. One 100-million-follower author can dominate a burst. Read cost grows with followed authors and refresh rate. Store bounded recent timelines and rebuild inactive users on return where the product allows.

Interview rule of thumb

“What is the follower distribution, not just the average?” Explain the celebrity tail before choosing a universal strategy.

Apply it

Feed.

Before choosing, name the required guarantee, one failure window and the metric that would force you to revisit this decision.

Source: content/comparisons/fanout-write-vs-read.md · Edit the Markdown to make this book your own.