Diagnosing Production Outages — Why Applications Stop Working
summary
The talk analyzes production JVM outages caused by both infrastructure (CPU, RAM, network saturation) and application bugs. It describes two concrete incidents: a replicated-cache change that sent full objects and multiplied memory usage across nodes, and a PDF generator leak where an XMLManager class retained gigabytes. Diagnosis techniques included heap dumps, retained-size heap analysis tools, and controlled reproductions (200 sequential requests versus multi-node load) to prove the bugs and validate fixes.
AI-written summary · about
chapters (8)
- 0:00Diagnosing outages
- 1:08Infrastructure issues
- 6:55Config errors
- 10:07Memory problems
- 22:11Memory leaks
- 32:41Memory causes
- 40:03Summary
- 44:46Q&A
related talks
47mManaging and monitoring a JVM
40mProduction Detectives: Solve the Case Before the Crash
45mDebugging Production: No Re-deploys, No Downtime, No Risk
30mA Well-Maintained Production Application
41mDetective Hat: Investigating Production Issues
58mObserving Applications in Production
39m30 minutes of Failure
47mHow I Solved Production Issues with OpenTelemetry (And How You Can Too)