Designing an unobtrusive analytics framework for monitoring java applications sampo suonsyrjä
0 likes421 views
This document discusses designing an unobtrusive analytics framework for monitoring Java applications. It proposes using aspect-oriented programming with AspectJ to monitor usage without altering the target application's code. Event data would be collected via Fluentd and stored in ElasticSearch for analysis with Kibana. This allows usage data to be gathered and compared across versions while avoiding complications from changes to the target application.
Designing an unobtrusive analytics framework for monitoring java applications sampo suonsyrjä
- 1. Designing an Unobtrusive Analytics Framework for Monitoring Java Applications Sampo Suonsyrjä and Tommi Mikkonen Dept. of Pervasive Computing Tampere University of Technology
- 2. Contents • Context – Paradigm change in software development – Web apps vs. installable programs • Motivation – Why the unobtrusiveness? • Case – Vaadin framework as the platform of a target program – The designed analytics framework • Discussion • Conclusions 17.10.15 2
- 3. Paradigm Change in Software Development • Waterfall method and such – User involvement only before development – Requirements defined before any use – Systems remained the same throughout their life – Changes in user requirements? No more use! Throw the cd away • Agile methods – Requirements defined as late as possible – Users involved iteratively – More knowledge of what is possible and needed – Developing less, but with a greater accuracy – Changes in user requirements? Order an extension to the sw. dev. project 17.10.15 3
- 4. Paradigm Change and the First Frontier • Feedback with agile methods – User surveys, thinking aloud, observations – Manually collected qualitative data – Stops when development stops • Continuous Delivery & Deployment systems – New versions are delivered faster than ever – Developing experiments, not just reacting to requirements – Usage Data = results from these experiments – Automatic collecting needed • Web as a platform for analytics – Channel is clear – Tools are available – A/B testing is easier than ever – Continuous experimentation 17.10.15 4
- 5. Why the Unobtrusiveness? • Collecting usage data leads to changes in the target program • Target program is continuously under change • Target program can be built by others • Data should remain – Collectable – Comparable between different versions • If data collecting is independent of the target.. • ..then data collecting mechanism could be reused 17.10.15 5
- 6. Vaadin Framework • Open source framework for developing Rich Internet Applications • Applications are written in Java.. • ..and then transformed into AJAX applications 17.10.15 6
- 7. Target Program • Demo app of the Vaadin Framework • QuickTickets Dashboard Demo • Source code available • Built by someone else than the researchers 17.10.15 7
- 8. Target Program public DashboardView() { HorizontalLayout top = new HorizontalLayout(); addComponent(top); Button notify = new Button('2'); Notify.addClickListener( new ClickListener(){ ... }); top.addComponent(notify); }; 17.10.15 8
- 9. The Designed Framework for Unobtrusive Analytics • AspectJ for usage monitoring • Fluentd for collecting • ElasticSearch for storing • Kibana for visualizing 17.10.15 9
- 10. AspectJ public aspect AddComponentListener { pointcut addComponentCall(Button b): call(* *.addComponent(*))&& args(b); after(final Button b):addComponentCall(b) { b.addClickListener( new Button.ClickListener() { public void click(ClickEvent e) { dataCollector.logEvent(b, e); } }); }} public DashboardView() { HorizontalLayout top = new HorizontalLayout(); addComponent(top); Button notify = new Button('2'); Notify.addClickListener( new ClickListener(){ ... }); top.addComponent(notify); }; 17.10.15 10
- 11. AspectJ • Aspect-oriented programming is unobtrusive by nature • No alteration of the original source code • Tooling is affected however – AspectJ dependency inserted – Aspect class file inserted – Normal compiling + aspect weaving 17.10.15 11
- 12. Fluentd • Implemented similarly as aspects – Fluentd dependency added – A logging class added • Parses data into JSON • Sends data for storing • In this case, Fluentd was installed and run on the same machine as the target program 17.10.15 12
- 13. Fluentd public class DataLogger { private static FluentLogger LOG = FluentLogger.getLogger("button.click"); public void logButtonClick(Button b, ClickEvent event){ Map<String, Object> data = new HashMap<String, Object>(); data.put("Button Caption", b.getCaption()); data.put("Button ID", b.getId()); ... LOG.log("click", data); }} 17.10.15 13
- 14. ElasticSearch and Kibana • ElasticSearch is document oriented • Data was already in JSON àName fields automatically • Real-time access with Kibana • Different visualizations readily available 17.10.15 14
- 15. Discussion • RQ1: To what extent can a data collecting feature be implemented without compromising the evolution of the target program? • Usage monitoring inserted without changing the source code • Aspect and logging class files and dependencies inserted • Target application’s evolution was not compromised – No additional effort needed if new version has different buttons – However, if the new version introduces new ways of implementing buttons aspects need changing as well 17.10.15 15
- 16. Discussion • RQ2: What types of data can be collected with the given approach? • Aspect-oriented monitoring is flexible – Pointcuts could be made on a vast variety of different points – Advices can include almost arbitrary code – Access to the target program in source code level • Vaadin Framework provided an abundant platform for collecting data • Usage data collected in this case • Perhaps end-user feedback in the future? 17.10.15 16
- 17. Discussion • RQ3: How to connect the data collecting feature with an analysis framework? • Aspect-code can be put to produce data in various formats àUnified logging layer (Fluentd: JSON) àPossibility to use standardized storing and visualization tools àUseful when combining different kinds of data (access, error, application logs etc.) 17.10.15 17
- 18. Conclusions • New means of getting feedback from end-users are needed à Analytics with usage data • New versions continuously à Need for unobtrusiveness à Aspect-oriented collecting Thank you! 17.10.15 18