You’re Monitoring Kubernetes Wrong
summary
Claims that container monitoring just needs Docker stats are wrong; two concrete approaches are compared: sidecar/monitoring containers per pod versus kernel-level capture using the open source sysdig engine that hooks sys_enter/exit to record processes, network, disk and decode SQL. Metrics must be fully tagged and stored raw. A commercial Sysdig Cloud prototype auto-creates annotation-driven teams and dashboards, wires alerts (example: CPU >80% per pod), and in the demo sends registration emails successfully.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:30Unicorns
- 1:02Introduction
- 2:22Context
- 2:46Unit vs Service
- 4:13Services
- 4:36Monitoring Kubernetes
- 4:49Revolutionary
- 5:43Four Things
- 6:18Getting the Data
- 7:09Pulse
- 7:57Instrumenting
- 8:20Kernel Module
- 10:17Guidelines
- 16:50Troubleshooting
- 22:47Microservices
- 23:56Demo
related talks
28mYou’re monitoring containers wrong
29mKubernetes-Defined Monitoring [I]
44mThe Dark Art of Container Monitoring
50mMonitoring and metrology for containers
46mMonitoring and metrology for containers
26mMonitoring Kubernetes: Follow the Data [B]
34mOptimizing operations & container management
30mMix Kubernetes with Linux Kernel Tracing to Boost the Observability of Your Cluster