Systems @Scale Summer Live Q&A Session #3
summary
Built a 100% accurate mapping from configs to consuming services so commit authors get targeted test plans; many tests still run at 15-minute land-time tests to catch mismerges that review-time can miss. Schema-on-read encodes data per-block and falls back to heterogeneous mode when later records conflict. Privacy rules run via a batched executor that evaluates cheap allow/deny checks first. Storage uses tiered storage with LSM blocks for compression and fewer on-disk copies. Framework supports 30+ back-end storages via generic APIs and adapters.
AI-written summary · about
chapters (11)
- 0:00Welcome to Q&A
- 0:26Testing configs at Meta
- 2:12Schema changes in RoxDB
- 3:14Privacy rule execution at Meta
- 4:16Config reliability improvements
- 5:21Nested indexing at Meta
- 6:47APIs for storage backends
- 7:51Tests at land vs review time
- 10:42Data integrity in schema-less ingest
- 12:05JIT engine optimizations
- 13:07Closing remarks
