Two Hard Things: Naming Things & Cache Invalidation, Valkey Edition
summary
Valkey is a Linux Foundation key‑value in‑memory datastore forked from Redis 7.2.4 and created by the community within eight days after the license change; it adds IO threads and other AWS contributions to raise throughput from ~380k to 1.2M requests/sec and supports clusters up to 2,000 nodes and 1 billion RPS. Talk details practical fixes: a strict key taxonomy, curly‑brace hashing to colocate shards, jittered TTLs, event‑driven invalidation (Debezium), write‑through/behind and tag graphs to avoid stale or duplicate cache entries. Valkey is a Redis 7.2.4 fork IO threads boost: 380k → 1.2M rps memory reduction 47% for small payloads
AI-written summary · about
chapters (17)
- 0:00Introduction and welcome
- 1:07What is caching and why it matters
- 2:52Introduction to Valkey and its origins
- 5:02Valkey 8.0 features and improvements
- 7:11Five common caching mistakes
- 13:33Key naming taxonomy for caching
- 16:18Valkey cluster and sharding explained
- 18:43Avoiding hot shards in clusters
- 20:24Anti-patterns in key naming
- 21:51Working with large keys efficiently
- 23:15Cache invalidation strategies
- 24:39Solving the thundering herd problem
- 27:05Event-driven cache invalidation
- 28:08Write-through vs write-behind caching
- 30:23Performance metrics and optimization
- 31:03Summary and key takeaways
- 31:25Q&A session
related talks
48mWhat's New in Valkey 9?
33mValkey in Telecom: Leveraging Open Source for Unique Needs and G...
25mYou Never Know When You Need a Fork
33mIntroduction to Redis
32mYou Never Know When You Need a Fork: Lessons Learned Building...
23mSystems @Scale Summer Live Q&A Session #1
05mKeynote: Your Core Infrastructure Should Be Vendor Neutral and Open Source
40mSponsored Session: How We Built GLIDE