Lightning Talk: Towards supporting backward compatibility
summary
Implemented a flatbuffers drop-in replacement for FoundationDB’s serialization that preserves existing serialization idioms and enables rolling upgrades so single storage processes can be switched without triggering cluster recovery, letting teams detect a CPU regression one node at a time. It fails for key-ordering and streaming framing — flatbuffers breaks bytewise ordering and message-consumption guarantees — and is tested by building the FDB server as a shared-library dlopen in the simulator, with caveats about globals.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:20Availability of FDB
- 0:32Schema evolution
- 2:33Applications
- 4:05Flat Buffers
- 4:56Availability
related talks
48mA Deep Dive Into Database Internals
12mLightning Talk: A Stable Limited LibTorch ABI? How?! (and Why?)
19mFlow: An Introduction
01h 12mLightning Talks by Simons Institute Fellows
01h 02mPolyglot Programming through Foreign Function Interfaces
06mLightning Talk: A Quick Intro to Falcon
11mLightning Talk: What If the Runtime Was Portable Too? Self-Hosted Runtime Capabilitie...
41mTowards language support for distributed systems