Self-Healing Patterns for Cloud-Native Streaming
summary
A concrete pattern set that actually worked: a replayable append-only log with partition keys and sequence numbers for deterministic replay and snapshots to restore state, idempotent handlers for exactly-once semantics, event-driven autoscaling that spawns consumers from consumer-lag signals to handle spikes (billing workloads hit millions of requests per day) and Linkerd-style mTLS plus circuit breakers to prevent cascading failure. Observability uses Prometheus metrics, distributed traces and heat maps, and automation reduces mean time to recovery dramatically while preserving throughput and ordering.
AI-written summary · about
chapters (13)
- 0:00Introduction and Speaker Background
- 0:35Evolution of Cloud Architecture
- 1:21Key Learning Objectives
- 2:01Architecting Resilient Streaming Systems
- 3:38Modernizing Legacy Systems
- 4:08Circuit Breakers and Failure Mitigation
- 4:47Reliable Event Replay
- 5:42Kubernetes Microservices and Data Scaling
- 9:24Automated Failure Detection
- 11:16Operationalizing Observability
- 12:28Pattern Catalog for Resilience
- 14:29Future of SRE and Kubernetes
- 15:10Conclusion and Closing Remarks
related talks
26mRethinking Cloud-Native Patterns on Kubernetes
41mCloud Native resiliency patterns from the ground up
35mStreams Must Flow: Fault-tolerant streams processing apps on Kubernetes
44mDesign Patterns for Distributed Systems in #AKS
42mPrinciples and Patterns for Streaming Data Analysis
46mService mesh patterns
48mLeveraging the power of Kubernetes with patterns
38mExploring Stateful Microservices in the Cloud Native World