Class Data Sharing and JVM Startup Tuning
Accelerate JVM-mode startup with CDS, lazy initialization, and bean instantiation tuning.
Class Data Sharing and JVM Startup Tuning is a free Spring Boot 4 Complete Guide lesson on CoddyKit — lesson 3 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the Spring Boot 4 Complete Guide learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.
Why JVM Startup Still Matters
GraalVM native images give near-instant startup, but most teams still ship a regular JVM build for the vast majority of deployments. The JVM is easier to debug, supports full reflection, and works with every agent and library out of the box.
The good news: a JVM-mode Spring Boot 4 app does not have to be slow to start. Three levers move the needle the most:
- Class Data Sharing (CDS) — skip re-parsing class metadata on every boot.
- Lazy initialization — create beans only when first used.
- Bean instantiation tuning — avoid expensive work in constructors and at refresh time.
This lesson walks through all three, in JVM mode, no native image required.
What Class Data Sharing Actually Does
When the JVM starts, it loads, parses, and verifies hundreds or thousands of classes. Class Data Sharing (CDS) writes the parsed in-memory class representation into a read-only archive file once, then memory-maps that archive on every subsequent start.
The payoff:
- Less class loading and verification work per boot.
- The archive can be shared across multiple JVM processes (same physical memory pages).
Two flavors matter for Spring Boot 4:
- Dynamic CDS (AppCDS) — archive your application classes, not just the JDK ones.
- Project Leyden / Ahead-of-Time cache — an evolution layered on CDS in newer JDKs.
Spring Boot 4's Built-in CDS Support
Spring Boot has first-class support for generating an AppCDS archive. You run the app once in a special training mode, it touches the application context, exits, and dumps an archive. Subsequent runs point the JVM at that archive.
The training run uses the special property below. It performs context refresh and then shuts down without serving traffic:
-Dspring.context.exit=onRefreshstops the app right after the context is ready.- The JVM flags
-XX:ArchiveClassesAtExitcapture the loaded classes.
# Step 1: training run - refresh context, then exit, dumping the archive
java -XX:ArchiveClassesAtExit=app.jsa \
-Dspring.context.exit=onRefresh \
-jar target/myapp.jar
# Step 2: every production start reuses the archive
java -XX:SharedArchiveFile=app.jsa \
-jar target/myapp.jarLayout Optimization with the CDS-Friendly Jar
A standard Spring Boot fat jar nests dependency jars inside it. CDS works best when classes live as plain files on a path the JVM can map directly. Spring Boot's tools layout extracts the app into a directory so CDS can index it cleanly.
Extract the runnable structure with the built-in jar mode, then run from the exploded layout:
-Djarmode=tools extractwrites anapplication/directory with a flat classpath.- Running from the exploded form makes both training and production starts faster and more reproducible.
# Explode the jar into a CDS-friendly directory structure
java -Djarmode=tools -jar target/myapp.jar extract --destination app
# Train against the exploded app
java -XX:ArchiveClassesAtExit=app/app.jsa \
-Dspring.context.exit=onRefresh \
-jar app/myapp.jar
# Production start
java -XX:SharedArchiveFile=app/app.jsa -jar app/myapp.jarVerifying CDS Is Active
Always confirm the archive is actually being used; a typo in the path silently falls back to no CDS. Add class-load logging to see which classes come from the shared archive versus the regular classpath.
Use the diagnostic flag during a one-off check:
-Xlog:class+load:file=cds.logtags each class withsharedwhen it loads from the archive.- Grep the log: a healthy archive shows the bulk of framework classes marked
shared.
# Run with class-load logging and inspect the source of each class
java -XX:SharedArchiveFile=app/app.jsa \
-Xlog:class+load:file=cds.log \
-jar app/myapp.jar
# Count how many classes were served from the shared archive
grep -c 'source: shared objects file' cds.logLazy Bean Initialization
By default Spring eagerly instantiates every singleton bean during context refresh. With many beans, that work dominates startup. Lazy initialization defers each bean's creation until it is first injected or requested.
Turn it on globally with one property:
spring.main.lazy-initialization=truemakes all beans lazy.- Trade-off: errors in a bean's wiring surface at first use instead of at boot, and the first request that touches a cold bean pays its construction cost.
Prefer it for short-lived CLI tasks and dev startup; be cautious in always-on services where a slow first request is worse than a slightly slower boot.
spring.main.lazy-initialization=trueSelective Laziness with @Lazy
Global laziness is blunt. Often you want most beans eager (so failures are caught at boot) but a few heavy beans lazy — a client that opens a slow connection, or a cache that pre-loads a large dataset.
Annotate the bean or the injection point with @Lazy. Spring then injects a proxy and instantiates the real bean on first call.
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.context.annotation.Lazy;
@Configuration
public class ReportingConfig {
// Heavy bean: only built when something actually needs it
@Bean
@Lazy
public ReportGenerator reportGenerator(DataWarehouse warehouse) {
return new ReportGenerator(warehouse);
}
}Keep Bean Constructors Cheap
The single biggest self-inflicted startup wound is doing real work in constructors or @PostConstruct methods: opening connections, warming caches, calling remote services. All of it runs during refresh, serially, blocking startup.
Rules of thumb:
- Constructors should only store collaborators, never perform I/O.
- Defer expensive warm-up to an event that fires after the context is ready, or run it asynchronously.
- Listen for
ApplicationReadyEventfor warm-up that should not block the context from coming up.
import org.springframework.boot.context.event.ApplicationReadyEvent;
import org.springframework.context.event.EventListener;
import org.springframework.scheduling.annotation.Async;
import org.springframework.stereotype.Component;
@Component
public class CacheWarmer {
private final ProductCache cache;
public CacheWarmer(ProductCache cache) {
// cheap: just hold the dependency
this.cache = cache;
}
// Runs after startup, off the critical path
@Async
@EventListener(ApplicationReadyEvent.class)
public void warmUp() {
cache.preload();
}
}Background Bean Instantiation
Spring Boot can instantiate eligible beans on a background thread pool while the main thread continues refreshing. This parallelizes independent, slow-to-construct beans and shrinks wall-clock startup time.
Define a bootstrapExecutor bean; Spring uses it to construct beans that opt in, in the background. Beans with no dependency on those still proceed on the main thread.
import java.util.concurrent.Executor;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.scheduling.concurrent.ThreadPoolTaskExecutor;
@Configuration
public class BootstrapConfig {
// Spring detects a bean named 'bootstrapExecutor' for background init
@Bean
public Executor bootstrapExecutor() {
ThreadPoolTaskExecutor exec = new ThreadPoolTaskExecutor();
exec.setCorePoolSize(4);
exec.setThreadNamePrefix("bg-init-");
exec.initialize();
return exec;
}
}Measuring Startup with the Startup Actuator
Don't guess where the time goes — measure. Spring Boot records a startup timeline of every step (bean instantiation, post-processing, auto-configuration) when you supply a BufferingApplicationStartup.
Wire it in main(), then read the buffered events from the /actuator/startup endpoint to find the slowest steps. Sort by duration and attack the top offenders first.
import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;
import org.springframework.boot.context.metrics.buffering.BufferingApplicationStartup;
@SpringBootApplication
public class ShopApplication {
public static void main(String[] args) {
SpringApplication app = new SpringApplication(ShopApplication.class);
// capture up to 2048 startup steps for /actuator/startup
app.setApplicationStartup(new BufferingApplicationStartup(2048));
app.run(args);
}
}Putting It Together: A Tuned Startup Recipe
A realistic JVM-mode tuning pass for a Spring Boot 4 service combines the techniques in order of impact:
- 1. Extract the jar (
jarmode=tools extract) and generate an AppCDS archive with a training run. - 2. Apply
@Lazyto a handful of genuinely heavy beans; keep the rest eager so config errors fail fast. - 3. Move all warm-up / I/O out of constructors into
ApplicationReadyEventlisteners. - 4. Add a
bootstrapExecutorif you have several independent slow beans. - 5. Use
BufferingApplicationStartupto confirm each change helped.
The launch command then simply points at the archive:
java -XX:SharedArchiveFile=app/app.jsa \
-Dspring.threads.virtual.enabled=true \
-jar app/myapp.jarQuick Check: Choosing the Right Lever
You have an always-on Spring Boot 4 web service. Profiling shows boot time is dominated by re-parsing and verifying framework classes on every restart, and a few requests hit beans that were never warmed. You want faster repeatable startup without risking that the first user request becomes noticeably slow.
Recap
JVM-mode Spring Boot 4 can start fast without going native. Key takeaways:
- CDS memory-maps a pre-parsed class archive; create it with a training run (
-XX:ArchiveClassesAtExit+spring.context.exit=onRefresh) and consume it with-XX:SharedArchiveFile. Extract the jar first for a CDS-friendly layout. - Lazy init defers bean creation: global via
spring.main.lazy-initialization, or surgical via@Lazy. It trades fail-fast and first-request latency for a quicker boot. - Bean tuning: keep constructors cheap, move warm-up to
ApplicationReadyEvent, and parallelize independent slow beans with abootstrapExecutor. - Measure everything with
BufferingApplicationStartupand/actuator/startup— tune the slowest steps, not your assumptions.
Frequently asked questions
Is the “Class Data Sharing and JVM Startup Tuning” lesson free?
Yes — the full text of “Class Data Sharing and JVM Startup Tuning” is free to read here on the web, and the Spring Boot 4 Complete Guide course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the Spring Boot 4 Complete Guide course, upgrade to CoddyKit PRO.
What will I learn in “Class Data Sharing and JVM Startup Tuning”?
Accelerate JVM-mode startup with CDS, lazy initialization, and bean instantiation tuning. You practise Spring Boot 4 Complete Guide with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start Spring Boot 4 Complete Guide?
No prior experience is required. Spring Boot 4 Complete Guide on CoddyKit is structured for beginners through advanced learners; this is — lesson 3 of 4, so you can start here or from the beginning and move at your own pace.
How long does the “Class Data Sharing and JVM Startup Tuning” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this Spring Boot 4 Complete Guide lesson?
Yes. Every Spring Boot 4 Complete Guide lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.
All lessons in this course
- AOT Processing and the Native Build Pipeline
- Runtime Hints for Reflection and Resources
- Class Data Sharing and JVM Startup Tuning
- Diagnosing and Fixing Native Compatibility Issues