Eclipse Jifa - Tools for Troubleshooting Java Application in the Cloud

https://github.com/eclipse/jifa

Transcript

1. Sanhong Li, Java Champion, co-leader of the GreenTea JUG Denghui Dong, OpenJDK Committer Eclipse Jifa Tools for Troubleshooting Java Application in the Cloud
2. Background Solution Demo Wrap up
3. Garbage Collector Bytecode Interpreter Just-in-time Compiler JVM(Java Virtual Machine) Java HEAP Stack VM data Code Cache Metaspace jmap jstack jcmd pid CodeCache.dump Thread jcmd pid Metaspace.dump Java Diagnostic Tools Overview Internal Use(Alibaba Dragonwell only) Existing Tools verbose gc log
4. Analysis in Eclipse MAT APP ‘Local’ ‘Cloud’ heap dump transferred over network • Requiring heap dump file transfer from cloud to local • Requiring large memory for large heap analysis Challenge#1 Heap Dump Analysis in the Cloud
5. Challenge#2 GC log is hard to read/understand • GC performance is one of the crucial JVM metrics • Deciphering the GC log files is simply daunting
6. Java Issue Finder Assistant Goal: It is designed as a web platform and aims for troubleshooting Java Application in the Cloud
7. Key Feature Roadmap 2020/3 2022/2 2021/12 Initial implementation (Heap Analysis) K8S Scheduling Open-sourced by Alibaba under Eclipse Foundation Master/Worker Pattern • GC Log Analysis • Thread Dump Analysis 2020/8 and other individuals…  Community Involvement Elastic deployment support  Incubation phase, next is to discuss for graduating to Mature phase
8. Background Solution Demo Wrap up
9. Solution • Web Application, frontend and backend separation • Backend is backed by Vert.x(https://vertx.io/) • Frontend is relying on Vue(https://vuejs.org/) • Analytic Engine • HeapDump service is implemented based on Eclipse MAT • GC Log Analyzer
10. Solution (cont.) • Master & Worker Pattern • Master • Route the user requests from the browser to the worker • Schedule new worker on demand(according to the resource usage) for serving the requests • Worker • Do the real analysis work
11. Worker Scheduling • K8S-based: open-source system for automating deployment, scaling, and management of containerized applications. • Kubernetes Java Client: Java library to operate the Kubernetes cluster • https://github.com/kubernetes-client/java/ https://github.com/eclipse/jifa/pull/98/files Implementation details: Master/Scheduler Worker Pod1 Worker Pod2 Worker Podn K8S Cluster Dynamic Pod(worker) creation as needed
12. Different Ways to Upload • Local: Uploaded from your local file • SCP: Copied from a remote host • URL: Pointed to one specified location in the internet • OSS: Transferred from Alibaba OSS • S3: Transferred from Amazon S3 Analytic Engine SCP Storage
13. Different Ways to Deploy • Worker only model • simplest way to get started with Jifa, only one worker for the analysis • Master and Worker model • Consists of only one master and several workers • K8S Model • Rely on the K8S cluster management capacity, and create worker resource as needed https://github.com/eclipse/jifa/tree/master/deploy Guide:
14. Report Example(Heap Analysis) http://jifa.dragonwell-jdk.io/heapDump?file=1617170189191-demo.hprof Report Details per View Different GC Roots Different Views The chain of objects which keep live and traced from GC Root(Thread) Follow the same guide from Eclipse MAT https://help.eclipse.org/latest/index.jsp?topic=/org.eclipse.mat.ui.help/welcome.html
15. GC Log Analyzer Introduction https://github.com/eclipse/jifa/issues/107 • Open sourced as a part of Eclipse Jifa • GC algorithm support • Serial GC / Parallel GC / CMS / G1 / ZGC • Java version support : 8/11(17 is under development) • Feature list • Diagnostic recommendation • Interactive graphs • Key performance indicators • GC pause statistics Implementation Details:
16. Eclipse Jifa at Alibaba  Alibaba uses Java at massive scale  Internal diagnostics platform is based on Eclipse Jifa, used by   Productized as part of Application Real-Time Monitoring Service(ARMS) for Alibaba Cloud users Usage in Alibaba (approx.) 10,000 developers 100,000 applications 1,000,000 JVM instances https://www.alibabacloud.com/product/arms
17. Background Solution Demo Wrap up
18. #1 How to Build • Prerequisites • JDK 11 • Gradle • Build Artifacts • Backend • Master • Worker • Frontend
19. #2 How to Deploy • Worker only model • Master and Worker model(Default) • K8S Model
20. #3 How to Analyze • Step 1: Prepare a dump file • Step 2: Upload it to Jifa • Step 3: Analyze it via the browser
21. Wrap up • Key Takeaways • Eclipse Jifa is a web application, designed for troubleshooting Java application in the cloud • Eclipse Jifa currently supports Heap Dump analysis, GC Log and Thread Dump analysis coming soon • Eclipse Jifa is still in incubation phase(Welcome contribution) • GitHub: https://github.com/eclipse/jifa • Slack: https://app.slack.com/client/T018ZRS4ZC1/C018TCUNE12 Have a try: http://jifa.dragonwell-jdk.io/
22. Thank you