Architectural Patterns for Distributed and Decentralized Data
summary
Argues that CRDTs converge without a coordinator so replicated state can achieve availability and partition-tolerance at the cost of linearizability; shows CQRS = append-only event log with materialized views where the log is the source of truth and snapshots speed rehydration; demonstrates Kafka delivery tradeoffs (Kafka: at-least-once vs effectively-once) and warns that Saga often fails for non-idempotent side effects (rollbacks and coordination bugs). Recommends Cloudstate/“AA serverless” and even Blockchain as higher-level ways to build these patterns.
AI-written summary · about
chapters (17)
- 0:00Start
- 9:58Consistency Challenges with Global Scale
- 14:50The Compute-Reality Paradox
- 15:35(D)evolution of Consistency
- 16:36Challenges with Understanding Consistency
- 20:51Consistency Models
- 21:42Linearizability
- 22:13CAP Consistency is Linearizable Consistency
- 24:53Strict Serializability
- 34:34Monotonic Sequence
- 35:18Akka Cluster State is Monotonic
- 36:52Convergent Operations
- 37:18Example Operations
- 37:52CRDT Counters
- 38:13CRDT Registers
- 46:14Advantages of Application Tier Consistency
- 48:10Overview of Saga Pattern
related talks
30mData Consistency Patterns in Cloud Native Applications
29mFollowing the Data: A Deeper Look into Distributed Applications
58mElements of Distributed Architectures
43mFunctional and Reactive Event Sourced Applications
36mConflict-Free Replicated Datatypes (CRDT)
56mDeveloping Event-driven Microservices w/Event Sourcing and CQRS
01h 01mLost in transaction Strategies to manage consistency in distributed systems
25mCloudstate—Towards Stateful Serverless