o11y will not fix your broken monitoring
summary
Claim: buying observability tools does not fix broken monitoring. Concrete fixes are automate monitoring via infrastructure as code, recreate a single source of truth so alerts don’t drift, and reduce noisy alerts — teams hit less than 1% actionable alerts across thousands of services. Examples: a government shop ended with two parallel stacks after a Prometheus migration, Grafana default retention dropped metrics to 30 days, logs got truncated to ~3 days, and a cross‑platform tool like coroot with eBPF actually surfaced actionable traces and profiling.
AI-written summary · about
chapters (16)
- 0:00Observability misconceptions
- 2:09Monitoring hype vs reality
- 6:27Traditional monitoring problems
- 8:18Case study: large government org
- 10:47Case study: misaligned environment
- 12:55Case study: performance alerts
- 14:43Observability regression
- 16:55Why observability?
- 17:59Fix monitoring first
- 21:19When observability needed
- 22:42Identifying real needs
- 24:28Choosing observability stack
- 27:22Modern tooling beyond K8s
- 29:10Pixie and Coroot solutions
- 30:35Observability is not a silver bullet
- 31:22Q&A
related talks
05mObservability will not fix your broken monitoring
40mDemystifying Observability and Monitoring
32mEssential Software Observability for Every Role: Look No Further
52mHolistic Monitoring
26mYou Can’t Fix What You Can’t See: Tales from the Observability Frontier
58mObserving Applications in Production
36mThe road to world-class observability
21mJourney from Broken Dashboards to O11y Maturity