Designing the library powering Netflix's QoE with Scala 3

on YouTube ↗46m

" summary

Centralized Measures Library defines 750+ concrete measures, implemented in Scala 3 using Opaque Types, given wiring and the boundary API, runs on Spark and Flink, and reuses the AC simulator to replay ~1M sessions in 30–60 minutes; heavy parsing is done once per session. A compile‑time macro called summonCached eliminates runtime hashmaps and cuts CPU by ~20–25%. The pipeline produces a JSON dependency graph for an interactive UI. The library is direct‑style, composable, and deployed across multiple apps.

AI-written summary · about

" chapters (9)

1. 0:00Netflix video streaming QOE
2. 2:23Data collection and analysis
3. 7:35Challenges in metrics design
4. 9:51The measures library design
5. 20:26Direct style and API ergonomics
6. 22:03Inside the boundary API
7. 25:50Optimization and caching
8. 30:50Future work and conclusions
9. 32:20Q&A session

" related talks

• 54mWhy Netflix ❤'s Scala for Machine LearningJeremy Smith, Aish · Scala Days Lausanne 2019
• 38mScaling ScalaJulien Tournay, Filipe Regadas · Scala Love in the City 2021
• 48mHigh performance Scala systems without tearsZahari Dichev · Scala Days Berlin 2018
• 28mStreams on top of ScalaQuentin Adam · Scalar Conference 2015
• 01h 02mSimply ScalaMartin Odersky · ScalaCon 2022
• 43mBuild a Modern, End-to-End, Big Data Scala Reference ApplicationEdgar Orendain · Scale By the Bay 2017
• 46mFast & FunctionalMichał Płachta · GeeCON 2019
• 48mMassively Parallel Distributed Scala Compilation And YouStu Hood · Scala Days Lausanne 2019