This repo contains my java & Serverless demo for Google cloud Next 2024. This apps provide a chat endpoint that answers users question. Instead of only relying on its trained model, it also uses a RAG (Retrieval Augmented Generation) model to provide more accurate answers. The additional knowledge is extracted from java.dev website and aims to help java enthusiast to get started with Java programming. The app is built using Spring boot, Vertex AI, Langchain4j, and deployed into Google Cloud run.
The main focus of this repo is to compare defirent java app optimizations and check how they affect the (cold) startup time. Below the results I got from running the app with @vCPUs, 2GB memory, 2nd Generation execution env with CPU boost enabled.
| Optimization | Cold start time |
|---|---|
| java 21 with no optimization | 10.301s |
| java 21 with tiered compilation | 8.633s |
| java 21 with Class data sharing (CDS) | 4.818s |
| java 21 with CRaC (Coordinated Restore at Checkpoint) | 1.570s |
| java 21 with Graal VM | 0.911s |
NB: The numbers are calculated based on the reported startup time from spring-boot logs.