How can you optimize JVM heap tuning using ergonomics vs manual flags?

JVM ergonomics is the set of default heuristics that pick a garbage collector and heap sizing automatically based on the detected number of CPUs and available memory - for example, by default HotSpot sets initial heap size to roughly 1/64th of physical memory and max heap size to roughly 1/4th, and selects G1 as the collector on typical modern machines.

These defaults work reasonably well for a generic workload, but they don't know your application's actual allocation rate, latency requirements, or deployment constraints, so tuning should start with measuring, not guessing.

1. Enable GC logging (-Xlog:gc*:file=gc.log) and observe real pause times and heap occupancy under realistic load, rather than assuming defaults are wrong.
2. Only override specific ergonomic decisions where the logs show a measured problem - for example, raise -XX:MaxGCPauseMillis if G1 is struggling to hit a target, or switch collectors entirely to ZGC if pause time, not throughput, is the constraint.
3. In containers, explicitly set -Xmx close to, but under, the container's memory limit, since letting ergonomics guess from the wrong visible memory figure was a historical source of OOM-killed containers before the JVM became container-aware in JDK 10+.
4. Avoid copying tuning flags wholesale from unrelated projects or blog posts - a flag set tuned for a different allocation pattern can easily make your specific workload worse, not better.

The general principle is to trust ergonomics as the starting point, then make narrow, measured overrides rather than broad manual configuration from scratch.