Build Failure Prediction in Continuous Integration Workflows
3 likes1,479 views
The thesis by Thomas Rausch investigates build failure prediction in continuous integration (CI) workflows through an empirical study of 14 open-source projects. It identifies key errors causing build failures, factors influencing build outcomes, and demonstrates the feasibility of predicting failures using machine learning algorithms. The study reveals that certain process and change metrics significantly affect build outcomes, and suggests future improvements in prediction methods and CI tool integration.
![Thomas Rausch 15
Build Error
Build log output
validatefetch compile inspect test
“[ERROR] Compilation Error”](https://image.slidesharecdn.com/presentation-online-161002092315/85/Build-Failure-Prediction-in-Continuous-Integration-Workflows-15-320.jpg)
![Thomas Rausch 36
Multi-class classification results
Multi-class classification
●
●
RMSE
●
Error between
actual and predicted
Average RMSE
p=[ p1 ... pn]
Algorithm 0-R NB C4.5 RF
All CM PM
0.0
0.3
0.1
0.2](https://image.slidesharecdn.com/presentation-online-161002092315/85/Build-Failure-Prediction-in-Continuous-Integration-Workflows-35-320.jpg)
Build Failure Prediction in Continuous Integration Workflows
- 1. Build Failure Prediction in Continuous Integration Workflows Master’s Thesis Presentation Software Engineering & Internet Computing Thomas Rausch Advisor: Stefan Schulte Co-Advisor: Waldemar Hummer
- 3. Thomas Rausch 3 Build t Build Code Product
- 4. Thomas Rausch 4 Continuous Integration (CI) ● Build at every change ● Identify problems early ● Allow frequent releases VCS CI Server Feedback Build P. M. Duvall, S. Matyas, and A. Glover, Continuous integration: improving software quality and reducing risk. Pearson Education, 2007.
- 5. Thomas Rausch 5 Build Failure Build Wasted resources t Examine Determine Fix ErrorBuild log
- 6. Thomas Rausch 6 Software Defect Prediction Complex code Bad practices Unstable environment Software defect M. D’Ambros, M. Lanza, and R. Robbes, Evaluating defect prediction approaches: a benchmark and an extensive comparison, vol. 17, no. 4–5. 2012.
- 7. Thomas Rausch 7 Build Failure Prediction? Complex code Bad practices Unstable environment Build failure ? Unstable environment ?
- 8. Thomas Rausch 8 Research Questions Which errors cause CI build failures? What factors influence build outcomes? Can we predict build failures?
- 10. Thomas Rausch 10 Empirical Study Research Setting ● 14 open source projects that employ CI Topology mapping a b c d VCS Builds Logs Change history CI build history
- 11. Thomas Rausch 11 Outline I Introduction II Solution Approach Study III Systematic study of build errors IV Factors influencing build failures V Build failure prediction VI Summary & Conclusion
- 12. III Systematic Study of Build Errors
- 13. Thomas Rausch 13 Build Steps validatefetch compile inspect test
- 14. Thomas Rausch 14 Build Failure validatefetch compile inspect test
- 15. Thomas Rausch 15 Build Error Build log output validatefetch compile inspect test “[ERROR] Compilation Error”
- 16. Thomas Rausch 16 Build Error Frequency Faulty VCS interaction Faulty build configuration Dependency error Compilation error Coding-rule violation Failing test Crash 40% 30% 20% 10% 0% 62%
- 17. Thomas Rausch 17 Build Error t validatefetch compile inspect test
- 18. Thomas Rausch 18 Frequency Seconds Error Probability Distribution t
- 20. Thomas Rausch 20 Causes for Build Failures Complex code Bad practices Unstable environment Build failure ?
- 21. Thomas Rausch 21 Measurable Properties Complex code Bad practices Unstable environment ) = ?fn(
- 22. Thomas Rausch 22 Measurable Properties Complex code Bad practices Unstable environment = fn(?)
- 23. Thomas Rausch 24 Measurable Properties Change Metrics ● What was done to the software? Process Metrics ● How were the changes applied?
- 24. Thomas Rausch 25 Measurable Properties Change Metrics ● Complexity ● File types ● Date and time ● Author Changes .java .txt
- 25. Thomas Rausch 26 Measurable Properties Process Metrics ● Build history ● Build type ● Integration scenario b1 b2 b4 t b3 a b e c f d g VCS commit graph CI build information Topology mapping
- 26. Thomas Rausch 27 Statistical Correlation Analyses Categorical Variables ● {c1, …,cn} ~ { , } Pearson’s chi-square test Numerical Variables ● R ~ { , } Mann–Whitney U test D. J. Sheskin, Handbook of Parametric and Nonparametric Statistical Procedures. crc Press, 2003.
- 27. Thomas Rausch 28 Results Metric Meta Build history +++++ Build type ++ Author ++ Change complexity + Date and time + File types ~ Integration Scenario ~ Meta statistics ● Using Fisher’s method ● Relative relation strength Van Zwet, W. R., & Oosterhoff, J. (1967). On the combination of independent test statistics. The Annals of Mathematical Statistics, 38(3), 659–680. article.
- 28. Thomas Rausch 29 PassedBuild outcome Failed Failed Passed Previous build result Percentageofbuilds Results Metric Meta Build history +++++ Build type ++ Author ++ Change complexity + Date and time + File types ~ Integration Scenario ~ b b’
- 29. Thomas Rausch 30 PassedBuild outcome Failed Failed Passed Previous build result Percentageofbuilds Results Metric Meta Build history +++++ Build type ++ Author ++ Change complexity + Date and time + File types ~ Integration Scenario ~ b b’
- 30. Thomas Rausch 31 Results Metric Meta Build history +++++ Build type ++ Author ++ Change complexity + Date and time + File types ~ Integration Scenario ~ Example ● Documentation file (changelog) ● 577 builds – doc. change only ● 14% original failures ● 52% test failures ● 45% environment crash ● 3% dependency error Similar behavior in all projects Noise skews statistics
- 32. Thomas Rausch 33 Machine Learning ? Classifier Observations ? ?
- 33. Thomas Rausch 34 Experiment Design Well-known algorithms ● Naive Bayes ● C4.5 Decision Trees ● Random Forest Feature sets ● Process metrics ● Change metrics ● Combined Prediction for ● Binary (failed/passed) ● Multi-class (error type) Baseline ● 0-R classifier frequency table predicts the average
- 34. Thomas Rausch 35 Binary classification result Average F1 -scores Binary classification ● F1-score, 1.0 = perfect ● Ranging from ● 0.71 – 0.91 Algorithm 0-R NB C4.5 RF All CM PM 0.3 0.4 0.5 0.6 0.7 0.8 0.9
- 35. Thomas Rausch 36 Multi-class classification results Multi-class classification ● ● RMSE ● Error between actual and predicted Average RMSE p=[ p1 ... pn] Algorithm 0-R NB C4.5 RF All CM PM 0.0 0.3 0.1 0.2
- 36. Thomas Rausch 37 Update Predictions False positive elimination ● Likelihood for error declines ● Plausibility of prediction declines tmax t50 t75
- 38. Thomas Rausch 39 Research Questions - Answered Which errors cause CI build failures? ● Failing tests (41%) ● Coding-rule violations (11%) ● Compilation errors (10%) ● A number of errors involving the CI environment
- 39. Thomas Rausch 40 Research Questions - Answered What factors influence build outcomes? ● Process Metrics ● Build history failures persist ● Build type merges vs. forward engineering ● Change Metrics ● Actually, not so much
- 40. Thomas Rausch 41 Research Questions - Answered Can we predict build failures? ● Yes! ● We can update predictions during the execution State Rate min max Passed 87% 45% 99% Failed 66% 27% 96%
- 41. Thomas Rausch 42 Contributions In-depth analysis of CI workflow ● Multiplicity of build errors ● Factors influencing build failures Topology mapping Baseline for CI build failure prediction
- 42. Thomas Rausch 43 Future Work Examine more influence factors ● Developer behavior ● Project management workflows Improve learning methods Incorporate prediction into CI tools ● Improve feedback mechanism ● Improve development productivity
- 43. The end? Questions & Answers Thomas Rausch [email protected]
- 44. Thomas Rausch 45 Research Subjects Project Name Description Apache Storm Distributed Computation Framework Butterknife Android Dependency Injection Library Crate.IO Scalable SQL database JabRef Graphical Java application for managing BibTeX databases jcabi-github Object Oriented Wrapper of GitHub API Hystrix Latency and fault tolerance library for distributed systems Presto Distributed SQL query engine for big data Openmicroscopy Microscopy data environment RxAndroid RxJava bindings for Android Sponge API Minecraft plugin API Spring Boot Java Application Framework Square OkHttp HTTP+HTTP/2 client for Android and Java Square Retofit HTTP client for Android and Java Wordpress-Android WordPress for Android
- 45. Thomas Rausch 46 Integration Scenarios a b c f e d g a b c e d f a b c e d a b c d Initialization Update
- 46. Thomas Rausch 47 Related Work Build Failure Prediction ● Hassan and Zhang 2006 ● Outdated assumptions ● No CI workflow considerations ● Kerzazi et al. 2014 ● Only statistical analysis, no prediction ● Only binary build outcome ● Wolf 2009, Schröter 2010 ● Socio-technical factors ● Closed source enterprise software
- 47. Thomas Rausch 48 Data Source { "id": 22555277, "commit_id": 6534711, "number": "784", "pull_request_number": "1912", "pull_request_title": "Example PR", "started_at": "2014-04-08T19:37:44Z", "finished_at": "2014-04-08T19:52:56Z", "duration": 2648, "state": "failed" "commit": { ... } } Travis-CI ● Hosted CI service ● RESTful API ● Integrated with GitHub
- 48. Thomas Rausch 49 Runtime Evolution