The InfoQ Podcast Somtochi Onyekwere on Distributed Data Systems, Eventual Consistency and CRDTs
10 snips
Jan 12, 2026 Somtochi Onyekwere is a software engineer at Fly.io, specializing in distributed systems and networking. He discusses the importance of fast, eventually consistent replication in data systems and explains how Conflict-free Replicated Data Types (CRDTs) enhance conflict resolution. Somtochi shares insights on the balance between speed and consistency, use cases for eventual consistency, and the challenges of stale reads. He also dives into Corrosion's architecture and its implementations, including CRSQL, which extends SQLite for collaborative environments.
AI Snips
Chapters
Transcript
Episode notes
Favor Speed Over Strict Consistence
- Fly.io prioritized low-latency reads by favoring speed over strong consistency to serve fast machine creation and routing.
- Somtochi Onyekwere explains eventual consistency allowed quicker decisions with acceptable staleness for most web apps.
Design For Stale Reads And Ownership
- Design the system to handle stale reads by letting authoritative nodes correct or retry actions at request-time.
- Partition ownership (each node writes its own rows) to reduce conflicts and simplify eventual consistency handling.
CRDTs Enable Coordination-Free Convergence
- CRDTs let independent replicas accept updates and later converge without coordination by merging states or operations deterministically.
- State-based CRDTs merge full states; operation-based CRDTs propagate operations and need duplicate-handling mechanisms.