**💻📰 When imperfect systems are good: Bluesky's lossy timelines**

Bluesky's social media platform prioritized performance improvements in its "Following Feed/Timeline" system. The system previously struggled with high latency, particularly impacting the 99th percentile of response times (P99). To address this, engineers made a deliberate trade-off: sacrificing some data consistency for significantly improved write performance. This involved changes to how posts are indexed and delivered to followers. Instead of ensuring every follower sees a post instantly and perfectly, the system now focuses on efficient distribution, accepting that some minor inconsistencies might occur.

The result was a dramatic reduction in P99 latency—over 96%—without noticeably affecting the user experience. This demonstrates a successful approach to system design where achieving "perfection" across all metrics (consistency, availability, latency, throughput) is impractical. Prioritizing one aspect (write performance in this case) based on its impact on users proved a more effective strategy than striving for perfection in all areas simultaneously. This approach highlights a key principle in system design: finding the optimal balance among competing properties to best serve the needs of the application.

[Read More](https://jazco.dev/2025/02/19/imperfection/)

💬 [HN Comments](https://news.ycombinator.com/item?id=43105028) (165)