Reliability Is Not Just About Balancing Loads
summary
Asserts that load balancers aren't enough and walks through concrete fixes: right‑sizing to avoid IO exhaustion, guardrails like locked nodes or Tails OS, QoS and circuit breakers, consolidating observability to cut MTTR from hours (three hours) to minutes (five), and operational recipes for Kubernetes, EKS, PVC/PV replication and Kafka upgrades where a team shortens a 15‑hour migration to 45 minutes by applying giops principles. Mentions tools like service mesh (retries/sidecars), DNS/global LB, Argo CD, Lens and Commodore as parts of the stack.
AI-written summary · about
chapters (15)
- 0:00Reliability beyond load balancing
- 2:12What load balancing means
- 4:22Right-sizing workloads
- 6:30Guardrails and security
- 7:55Service latency and observability
- 9:40Modern troubleshooting and MTTR
- 11:28Kubernetes upgrades and dependencies
- 13:13Disaster recovery strategies
- 15:41Reducing human dependency
- 16:45Infrastructure deployment speed
- 17:50Service mesh and failure handling
- 19:15Consolidated reliability platforms
- 20:45DNS reliability challenges
- 21:26Kubernetes multi-version complexity
- 22:08Learning from outages
related talks
13mBuilding Reliable Infrastructure for LLM Workloads on K8s
29mLoad balancing: an in-depth study to scale @ 80K TPS
38mThe Untapped Power of Services - L7 Load Balancing Without a Service Mesh
37mBuilding the Multi-Cluster L7 Load Balancer
29mKubernetes Workload Resiliency in Action: Beyond Basics
33mMany Layers of Availability
35mIntelligent Load Balancing in Kubernetes
35mKubernetes, Lessons Learnt