Java 25 Performance Improvements

Description

Now that dotCMS runs on Java 25 (runtime 25.0.2-ms, default bytecode 25 — see #35914), a set of language and JVM capabilities become available that let us improve runtime performance, throughput, and resource efficiency across the platform. This Epic is the umbrella / tracking issue for that performance work.

The goal is not a single feature but a coordinated sweep of targeted improvements that take advantage of Java 25, each delivered as its own child PR/issue so they can be reviewed, benchmarked, and rolled back independently.

Why now

• JEP 491 (Java 24+): synchronized no longer pins virtual threads. The classic "virtual threads + JDBC/legacy synchronized ⇒ carrier pinning" objection no longer applies, so we can safely migrate blocking I/O thread pools to virtual threads.
• Modern GC, scoped values, and other Java 21→25 additions open further throughput/latency wins.
• Much of dotCMS's hottest code is I/O-bound (cache persistence, DB access, HTTP), which is exactly where virtual threads and reduced thread-pool contention pay off.

Desired Outcome

We will consider this Epic successful when:

• Hot I/O-bound thread pools are migrated to virtual threads where it measurably helps, with no regression in correctness, memory footprint, or shutdown behavior.
• Each improvement is backed by a before/after benchmark or a clear architectural rationale.
• No increase in error rates (e.g. connection-pool timeouts) or cache rebuild storms under load.
• Changes are individually flag-/config-gated or independently revertable.

Candidate child work

• H22 cache async commits → virtual threads — replace the fixed 5-thread ThreadPoolExecutor with a virtual-thread-per-task executor; preserve backpressure and bound concurrent DB writes against the Hikari pool. (first child PR)
• Audit other blocking thread pools / ExecutorService usages for virtual-thread suitability.
• Evaluate GC / JVM flag tuning for the Java 25 runtime.
• Identify and benchmark additional I/O-bound hotspots (REST, ES, DB access layers).
• Review opportunities from newer concurrency APIs (e.g. scoped values, structured concurrency) where stable.

This list is expected to grow; new child issues/PRs should reference this Epic.

Links

• Java 25 migration: #35914
• CI variant follow-up: #35915

Start by scoping the H22 cache async-commit work, including the fixed five-thread ThreadPoolExecutor, as an independent child issue; no source file or test is named here. Define a before/after benchmark and verify preserved backpressure, bounded Hikari-pool writes, correctness, memory footprint, and shutdown behavior before expanding to other Java 25 performance candidates.