ADDITIONAL SYSTEM INFORMATION :
macOS Tahoe 26.2 (ARM64)
java version:
openjdk version "26-ea" 2026-03-17
OpenJDK Runtime Environment (build 26-ea+31-2866)
OpenJDK 64-Bit Server VM (build 26-ea+31-2866, mixed mode, sharing)

A DESCRIPTION OF THE PROBLEM :
When running a simple allocation loop with ParallelGC on JDK 26-ea+31 (macOS Tahoe 26.2, aarch64), the JVM intermittently enters a pathological state where it performs thousands of consecutive Full GCs due to Allocation Failure.

The committed heap remains very small (~22MB) despite -Xmx4g, suggesting heap expansion ergonomics regression.

The same test behaves normally on JDK 25.

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
I use the following command to run the test cases：
java "-Xlog:gc*:file=./ParallelCount26/parallel-26.log" -Xms8m -Xmx4g -XX:+UseParallelGC ParallelCount26.Test


---------- BEGIN SOURCE ----------
package ParallelCount26;

public class Test {
    protected volatile boolean[] field_7978;

    public Test() {
        this.field_7978 = new boolean[16];
        super();
    }

    public static void main(String[] var0) {

        (new Test()).test();
    }

    public void test() {
        this.field_7978 = new boolean[16];

        for(int var1 = 0; var1 < 2000; ++var1) {
            byte[] b = new byte[16*1024*1024];
        }

    }
}



---------- END SOURCE ----------

FREQUENCY :
OFTEN