I took one realistic blocking Spring Boot 4 service — a PostgreSQL query through HikariCP plus a downstream HTTP call — flipped spring.threads.virtual.enabled=true, and benchmarked exactly what changed: throughput, tail latency, and where the bottleneck went.
TL;DR: ~6× throughput at 2,500 concurrent users from one flag — but the connection pool quietly becomes your new ceiling, so you have to resize it. I also measured the JEP 491 pinning change (the "avoid synchronized" advice is now out of date).
Full write-up, raw data, charts and a reproducible repo on the Tucanoo blog
Top comments (0)