Catalogue
- Cards
- 165
- Records
- 486
- Workloads
- 19
- Corpus version
- 74847648e25a
7 of 486 records match.
| Kind | Record | Organisation | Year | Setup |
|---|---|---|---|---|
| Failure | Users who follow hundreds of thousands of accounts made hot shards that slowed the timelines sharing them | Bluesky | 2025 | Sharded |
| Decision | Drop timeline writes with a probability set by how many accounts a user follows, accepting imperfect timelines | Bluesky | 2025 | Sharded |
| Decision | Cache high-follow accounts in a Redis sorted set loaded into memory every 30 seconds instead of reading follow counts per write | Bluesky | 2025 | Sharded |
| Decision | Introduce a dedicated BFF microservice returning view models instead of putting UI fields on platform APIs | Lyft | ||
| Decision | Combine Declarative Components with client-side Semantic Components and decoupled chainable Actions | Lyft | ||
| Failure | Vertical scaling from 16 to 48 vCPUs hit bimodal CPU spikes and latency degradation from HotSpot superclass-cache sharing | Netflix | ||
| Decision | Patch HotSpot JVM to pad and then flag-gate bypassing secondary superclass cache writes | Netflix |