The Accord Consensus Protocol, how to do ACID transactions at scale
summary
Research-backed consensus called Accord delivers single-network-round-trip leaderless consensus that makes global ACID, strict-serialisable transactions across partitions and tables feasible with commodity clocks. It uses fast-path electorates to deterministically shrink quorums and a timestamp reorder buffer to handle clock skew and latency, falling back to a slow path if needed. Implementation targets Cassandra as an open, pluggable protocol (separate codebase, feature branch, large-scale fuzz/soak testing) and guarantees no rollbacks for accepted transactions.
AI-written summary · about
chapters (11)
- 0:00Start
- 6:33Apache
- 13:01Eventual consistency
- 17:41Multitable/partition transactions
- 19:32Isolation & Consistency
- 23:09Consensus Protocols
- 24:41Stable Leaders Downsides
- 30:03No Leaders
- 34:51Stable performance under failure
- 38:16Stable performance under contention
- 40:38Introducing Accord
related talks
34mThe Accord Consensus Protocol: ACID transactions globally at any scale
46mData Replication with Apache Accord
34mCassandra on ACID: This Changes Everything
13mThe ACID BASE Conundrum
54mEvolving Transactions in Apache Cassandra
47mBeing transactional in 2019 - what does this actually mean?
01mDon't Drop ACID - Transactions in Distributed NoSQL
52mTransactions: Myths, Surprises and Opportunities