大数据场景下应用性能排查的智能根源分析
0 likes•462 views
The document outlines the role of a Principal Data Scientist at AppDynamics, focusing on application performance management (APM) and the integration of various tools and methodologies such as RCA, metrics, and log analysis. It discusses techniques for monitoring different layers of application architecture, including user interactions, server performance, and transaction classifications to identify normal, slow, or erroneous behaviors. Furthermore, it emphasizes the importance of data-driven insights for root cause analysis and performance optimization in production environments.
大数据场景下应用性能排查的智能根源分析
- 1. Principal Data Scientist, AppDynamics
- 2. : + • Principal Data Scientist AppDynamics • APM • 2016 Gartner Forrester • / • KDD,ICDM,IJCAI • APM Log • Java/Python/Scala production code • Sumo Logic, Linkedin, eBay, IBM Watson
- 6. App
- 7. How? APM
- 9. 1 2
- 10. app
- 11. ...
- 12. ...
- 14. Monitor Everything: Application Performance Management Database Monitoring Synthetic Monitoring Browser Real-User Monitoring Mobile Real-User Monitoring Server Monitoring Database API Server ApplicationWeb Server Business Transaction Browser Mobile Unified Monitoring
- 15. • Metrics, KPI • CPU Java GC • • IP • • cluster —> machine —> process —> thread • • • stacktraces,
- 17. 1 Big Data
- 18. 2 Big Data
- 19. = +
- 23. • • • •
- 24. • • • •
- 28. (Filtering)
- 30. “airline:AA” transactions : • 2% • 82% “AA”. • 41
- 32. take actions!
- 39. transaction online classifier • Normal • Slow • Stall • Error
- 41. Now we know when it gets wrong, but can we know why? online classifier •Normal •Slow •Stall •Error
- 43. all transactions (baseline) / / transactions (query) baseline v.s. query transactions (baseline)
- 44. all transactions (baseline) / / transactions (100 ) baseline v.s. query transactions (1000 ) from = ‘Beijing’: Baseline Query 32 3 3.2% 3% attribute
- 45. all transactions (baseline) / / transactions (100 ) baseline v.s. query transactions (1000 ) from = ‘Beijing’: Baseline Query 32 3 3.2% 3% attribute
- 46. all transactions (baseline) / / transactions (100 ) baseline v.s. query transactions (1000 ) airline = ‘the Bus’: Baseline Query 150 80 15% 80% attribute
- 47. all transactions (baseline) / / transactions (100 ) baseline v.s. query transactions (1000 ) airline = ‘the Bus’: Baseline Query 150 80 15% 80% attribute
- 48. all transactions (baseline) / / transactions (100 ) measurements transactions (1000 )
- 49. / / transactions entity baseline transactions of entity 1 transactions of entity 2 transactions of entity n entity 1 entity 2 entity n ….
- 50. baseline: • cluster tier node • app •
- 52. normalization
- 59. errors/exceptions ≠ root cause significant errors/exceptions root cause
- 64. • • production • index • ElasticSearch Nested Documents • ES count overcount undercount
- 65. • • + • RCA • • • • transaction • log
- 66. • Metrics, KPI • • • • unified RCA
- 67. Next Time: RCA Framework > KPI production