Your application has died and that's OK
Download as pptx, pdf0 likes13 views
This document summarizes a presentation on crash analysis. It discusses the history of crash analysis from magnetic cores in the 1950s to modern techniques. It outlines different types of failures and crash dumps. It also discusses challenges with crash analysis in locked down production environments and how Kubernetes facilitates crash analysis. The presentation advocates for on-demand crash analysis integrated into development workflows to better diagnose issues.
Your application has died and that's OK
- 2. Your application has died and thats OK!
- 3. Hello! I am Anton Whalley Partner Technical Specialist Node.js Diagnostics Member llnode maintainer IBM Strategic Open Source Committer Rust Dublin Co-Organiser Certified Kubernetes Administrator 3
- 4. A (very) short history of crash analysis Magnetic Cores created Mid 1950s First Deployed by MIT/US Navy Commercialised IBM 770 4
- 5. Modern Crash Analysis 5 Executable Running Process Terminated Process Load Abort Snapshot E.G gcore Core Dump *nix Mini Dump Windows Runtime Specific Reporting WinDbg gcc, lldb Custom Tool Files Joyee Cheung Bring Javascript Back To Life https://www.youtube.com/watch?v=XQIo9knnb2s
- 6. Why Crash Analysis Locked Down Production Environment (Should be) Restricted Access Debug tools not available Heisenbugs Unoptimized Builds Issues with logs Only captures known/knowns Requires adhoc updates Stacktraces Dont capture the full state 6 David Pacheco - https://dl.acm.org/doi/10.1145/2039359.2039361
- 7. Failure Types 7 Implicit Explicit Type Error Uncaught Exception Segmentation Fault Panic Hardware Error Assertion Failure Process Abort Error Code Exit Incorrect Result Leaks Resources Stops Doing Work Pathological Fatal Non-fatal Error Message Returns Error Code Bryan Cantrill Docker in Production https://www.youtube.com/watch?v=AdMqCUhvRz8
- 9. The birth place of site reliability engineering? HMS Sailsbury 1747 First Controlled Experiment Identified a cure for scurvy 40yrs to be adopted In 1780 1457 admissions; in 1806 there were 2. 9
- 10. The return of scurvy! HMS Alert 1875 85% of crew succumbed Lemons changed for limes - 1860s Misunderstood causes The rise of steam engines meant shorter trips 10
- 11. How We Forget Improvements in Adjacent Technology Questionable Refinements in Approach Solutions are not Accessible Concepts with Unbalanced weighting The Generation Gap 11
- 12. Crash Analysis in K8s Core Dump Handler Open Source Cloud Agnostic - xKS From 12 Separate organisations contribute 12 https://github.com/IBM/core-dump-handler/
- 13. 13 On Demand Crash Analysis Augmented Pods with IDE Integrated Into Git workflow DX Infrastructure gitpod.io https://venshare.com/blog/gitops-coredump/
- 14. Future Work 14 - Automated Analysis - Remove Sensitive Data - Cloud Events
- 15. 15 Thanks! For Listening You can find me at @dhight9 or @No9 GitHub https://openjsf.org/blog/
Editor's Notes
- #5: Crash analysis is how we actually used to debug Extracting information from a malfunctioning application used to be how we debugged.
- #12: Improved Ship Technology == Improved development, practices code debuggers Focus on Backtraces lemons swapped for limes Lemons no longer accessible make tooling easy to find and use Biological Causes == The concept of uptime lead to starved connections/DNS failures The generational gap muscle memory of the organisaton fails
- #13: https://github.com/IBM/core-dump-handler/