Causality is simple #bbuzz
summary
Argues that causality is simple by reducing it to sets of unique event IDs (a causal histories model) and showing how those sets compress into vector clocks and version vectors; gives the concrete merge rule (pointwise max) and an optimization of storing the last event (the dot) to make causality checks constant time; demonstrates practical failures in Cassandra/Dynamo-style systems, explains Riak’s sibling approach for concurrent writes and deferred reconciliation, and recommends dotted version vectors for large-scale concurrency.
AI-written summary · about
chapters (9)
- 0:00The importance of causality
- 1:32Anomalies in distributed systems
- 3:50Limitations of physical time
- 7:38Space-time diagrams and events
- 11:07Causal histories explained
- 14:51Vector clocks and optimization
- 18:55Version vectors in practice
- 23:50Summary and takeaways
- 28:05Q&A on practical applications
related talks
20mInductive diagrams for causal reasoning
42mCausal Consistency in MongoDB Explained in Detail
47mAdventures in Spacetime
18mDistributed Causal Memory for Higher-Order Distributed Separation Logic
47mComplex event flows in distributed systems
39mEventually Consistent Data Structures
32mCan local-first really scale at the edge?
34mUsing CRDTs Beyond Text Editors