Keynote - Without Resilience, Nothing Else Matters
summary
Claims that resilience is by design and failure is inevitable, so architecture choices matter: use actor-style supervision, let it crash semantics, compartmentalized error kernels, and decentralised gossip rings (Dynamo style) across at least three nodes. Favor asynchronous, message-driven protocols, event sourcing and event log backed state, CRDTs or eventual/causal consistency, and backpressure/Reactive Streams. Expect to build self-healing systems, exercise them with chaos tools, and accept compensating actions instead of global transactions.
AI-written summary · about
chapters (19)
- 0:00Intro
- 4:24Complicated System
- 4:52Complex System
- 7:34Resilience is by Design
- 14:33Dealing in Security UNDERSTANDING VITAL SERVICES, AND HOW THEY KEEP YOU SAFE
- 19:36Recursive Restartability TURNING THE CRASH-ONLY SLEDGEHAMMER INTO A SCALPEL
- 20:03Traditional State Management
- 26:44Enter Supervision
- 27:20WE NEED A WAY OUT OF THE State Tar Pit
- 35:58Towards Resilient Distributed Systems
- 37:19Microservices
- 39:56Apply Promise Theory
- 41:18We need to decompose the system using Consistency Boundaries
- 44:08WITHIN the Consistency Boundary we can have STRONG CONSISTENCY
- 44:57WE NEED SYSTEMS THAT ARE DECOUPLED IN Time and Space
- 47:21Location Transparency ONE COMMUNICATION ABSTRACTION ACROSS ALL DIMENSIONS OF SCALE
- 49:19Resilient Protocols
- 55:59CRUD is DEAD
- 59:11Flow Control ALWAYS APPLY BACKPRESSURE
related talks
51mWithout Resilience, Nothing Else Matters
56mKeynote: Defining and Embracing The Software Fragility
22mResilience Is Not Enough
44mArchitectural patterns of resilient distributed systems
16mResilient by Design: Architectures for Crisis Response & Recovery
02m[ Keynote preview ] Resiliency in Distributed Systems: Beyond Just Two Copies Behind a Load Balancer
16mEngineering Failure-Resilient Systems
50mKEYNOTE: Residuality Theory in Software Architecture