Project CRaC
0 likes50 views
An overview of Project CRaC and JVM performance improvements. In a world where microservices are becoming more of a standard architecture for Java-based applications running in the cloud, the JVM warmup time can become a limitation. A new OpenJDK project called CRaC (Coordinated Restore at Checkpoint) aims to address the JVM warmup problem with a different approach. Basic idea is: take a snapshot of a running JVM, store it in files and restore the JVM at a later point in time or even on another machine.
Project CRaC
- 1. Project CRaC Coordinated Restore at Checkpoint Goran eko
- 2. What the CRaC?! 2 Coordinated Restore at Checkpoint (CRaC) on the Java Virtual Machine https://openjdk.org/projects/crac
- 3. OpenJDK vs Project CRaC OpenJDK is a place, not a thing (73 projects) Source code for the Java Development Kit: Virtual Machine (HotSpot) Java Class Library Java Compiler (javac) Project CRaC Proposed and led by Azul OpenJDK Zulu builds (binary distributions) by Azul include CRaC features OpenJDK vs builds is like Linux vs its distributions 3
- 4. How does JVM work? 4
- 7. Checkpoint & Restore concepts 7
- 8. CRaC API Coordination - ability to react on before and after events (resource management) The API enables coordination of resources during checkout and restore Checkpoint and restore notifications, awareness of checkpoint/restore events (callbacks) Java frameworks and libraries are adopting CRaC, eg. Apache CRaC implementation creates the checkpoint safely Ensures reliable restart from a given checkpoint org.crac dependency 8
- 9. Resource & Context public class MyResource implements Resource { public MyResource() { org.crac.Core.getGlobalContext().register(this); } @Override public void beforeCheckpoint(Context<? extends Resource> context) throws Exception { // free your resources } @Override public void afterRestore(Context<? extends Restore> context) throws Exception { // load your resources } } 9
- 10. Benefits Provides immediate application start with peak performance Faster deployment process as well Eliminates JIT compilation overhead or JVM warm-up Hotspot identification, method compiles, recompiles and deoptimisations Reduces costs in cloud/K8s environments (pay-as-you-use) When scaling resources to meet demand Less time and computing resources used Ideal for apps that perform a lot of work during startup & warmup Major frameworks are supported: Micronaut, Quarkus and Spring (Boot) 10
- 11. Limitations 1. No auto-dump for open files and sockets (we need to take care of that) 2. Must be restored on the same architecture (Linux x64,Arm64 | JDK 17 & 21 Azul Zulu builds of OpenJDK) 3. Changes in the value of the system clock (time difference) (for syncing or caching) 11
- 13. GraalVM Native Images - Ahead-of-Time Ahead of Time is static compilation Native instructions give full speed from the start No need to start the app ahead of time No CPU overhead to compile code at runtime 3x smaller memory consumption Smaller image than the snapshot 13 All classes must be available at build time No JIT (runtime) compilation optimisations Lower performance than JIT optimised JVM Debugging is harder
- 14. How fast is it? 14 * Proof of concept results
- 16. JVM CRaC parameters # START $ java -XX:CRaCCheckpointTo=/crac-files -jar app.jar # CHECKPOINT $ jcmd <PID>/app.jar JDK.checkpoint $ Core.checkpointRestore(); # RESTORE $ java -XX:CRaCRestoreFrom=/crac-files 16
- 17. Demo Time CRaC in Action 17
- 19. 19 Thank you. Java: the elegant simplicity of C++ and the blazing speed of Smalltalk. James Gosling https:/ /openjdk.java.net/projects/crac/ https:/ /github.com/crac