G1 and ZGC

Key points

• G1 is the default on anything with 2+ cpus and roughly 1.8 GB. It divides the heap into many equal-sized regions.
• G1 collects the regions holding the most garbage first - the garbage-first in the name - and works towards a pause target.
• -XX:MaxGCPauseMillis=200 is a goal, not a promise. Ask for less and G1 collects smaller batches more often.
• ZGC, enabled with -XX:+UseZGC, does nearly everything concurrently. Pauses stay under a millisecond at any heap size.
• ZGC gives up a little throughput and wants some extra memory. Take that deal when tail latency matters more than raw speed.
• JDK 21 adds generational ZGC behind -XX:+ZGenerational. On a tiny container the Serial collector is still the right answer.

Example

import java.lang.management.GarbageCollectorMXBean;
import java.lang.management.ManagementFactory;

public class Main {
    public static void main(String[] args) {
        Runtime rt = Runtime.getRuntime();
        System.out.println("cpus = " + rt.availableProcessors()
                + ", max heap = " + (rt.maxMemory() >> 20) + " MB");
        System.out.println();
        System.out.println("collectors running right now:");
        for (GarbageCollectorMXBean gc : ManagementFactory.getGarbageCollectorMXBeans()) {
            System.out.println("  " + gc.getName());
        }
        System.out.println();
        System.out.println("G1 Young/Old Generation  -> G1, the default on server-class machines");
        System.out.println("ZGC Cycles / ZGC Pauses  -> ZGC, switched on with -XX:+UseZGC");
        System.out.println("Copy + MarkSweepCompact  -> Serial, chosen on small containers");
    }
}

G1 for most services. ZGC when a 200 ms pause would count as a bug.