How Class Data Sharing Can Speed up Your Jakarta EE Application Startup
1 like251 views
The document discusses how Class Data Sharing (CDS) can significantly enhance the startup speed of Jakarta EE applications by loading, verifying, and storing class information in a memory-mapped read-only file. It highlights the importance of fast startup for handling large surges in requests and provides a comparison demonstrating that CDS can reduce startup time and memory usage. The document also covers potential pitfalls of using CDS and information on community support for adopting this technology.
How Class Data Sharing Can Speed up Your Jakarta EE Application Startup
- 1. How Class Data Sharing Can Speed up Your Jakarta EE Application Startup Rudy De Busscher
- 3. Rudy De Busscher • Payara • Technical Sales • Developer Advocate • Involved in • Committer of MicroProfile • Committer in Eclipse EE4J groups • Java EE Security API Expert group member @rdebusscher https://blog.payara.fish/ https://www.atbash.be
- 4. Agenda • JVM internals related to Class • Why fast startup important? • Class Data Sharing • Basics • Demos • Conclusion
- 5. Using a class • new MyClass(); • Load Java Byte code (through the Classloader) • Verification of Byte code • Class file derivation (can result in IncompatibleClassChangeError) • Preparation (~ variable initialisation) • Resolution (resolve symbolic references) • Interpreter uses code
- 6. Improvements? • Can we improve this? 1. —noverify option • Opens door for malicious byte code. 2. Native compilation • Slow • Lower peak performance • Obscurity with reflection • No JVTMI, Java Agents, … 3. Class Data Sharing - CDS
- 7. Why Fast Startup Important? • Large surge in requests • React fast • Avoid refused connection during scale up. • (Not cold startup problem with serverless)
- 8. Class Data Sharing - CDS • Load, verify, derive • Perform upfront • Store info in memory mapped read-only file • Loading class is much faster • Memory requirements are lower (can be shared) • Hello world -> 437 classes • Payara Micro -> 15.000 classes
- 9. appCDS • CDS • JVM classes only (~ rt.jar) • AppCDS • Including your class-path
- 10. Using CDS • OpenJDK • -XX:DumpLoadedClassList= • -XX:SharedArchiveFile= • -XX:share:dump|on • OpenJ9 • -Xshareclasses
- 11. Demo
- 12. Gain • OpenJDK • OpenJ9 Startup Time Memory Plain 8,2 sec 49,8 Mb CDS 5,2 sec 49,5 Mb Startup Time Memory Plain 7,9 sec 45.4 Mb CDS 5,1 sec 42,9 Mb 35%
- 13. Docker images • Several steps required • Multi stage Docker file • Shared Archive file can be in Image • Layer is reused • Shared Archive file can be outside • On a volume • Extra step in build or use podman build
- 14. Demo
- 15. “Pitfalls” • Class path issues • Classpath must be identical when using CDS than when CDS is prepared • Dynamic proxies can’t be used • Pre 1.6 class formatted files can’t be used. • Investigate issues -> -Xlog:load+class • Shared Archive file is version and OS dependent
- 16. We’ll Support You With: Let us help you spread the word about our open source software. Join the Reef! • Event, JUG, conference sponsorship • Freebies, swag, handouts, speakers • Promotion and advertising of events and articles • Community forum Learn More: www.payara.fish/reef Payara Reef: Community Growth Program
- 17. Download the open source software: payara.fish/downloads Get Started: payara.fish/get-started