Unexpected GC behaviour: some data always going into the tenure generation

Question

Ran a very simple code to try to see how garbage collectors function.

String a = null;
while ( true ) {
  a = new String(" no... ");
}

I am using ParallelGC. I printed GC results and here is the first (minor) GCs.

[GC [PSYoungGen: 16448K->1616K(19136K)] 16448K->1624K(62848K), 0.0022134 secs] [Times: user=0.00 sys=0.00, real=0.00 secs]

youngGen came down by 14880K however fullHeap came down by just 14872K

Does this mean 8k has moved to the tenure generation? My understanding is the GC might have been called some instanced of class 'a' must have been marked alive and moved to the tenure generation. Is this understanding correct? Also, is this "Floating Garbage"? Overtime the tenure generation does get filled up and a fullGC is required, however it does take awhile.

Also, in this particular case, shoudln't the entire minor collection get collected and ideally nothing goes into the tenure generation? All these are short lived objects.

Answer 1

You've got 1 instance of String alive when the GC is in progress (the strong reference inside the while loop) so that one is the one that survives hence the 8k.

In this case I wouldn't call the String ref floating garbage. Floating garbage is when an object wasn't ready to be GCed when the GC checked it, but was ready by the time the GC finished. An example of that would be.

Thread1:    Person p = new Person("sammy")

    Thread2:    gc runs and sees that the Person instance is reachable through p.

Thread1:    p = null; // This Person instance is now unreachable.

    Thread2:    GC finishes. The person instance could have been collected but was reachable at the time the collector checked it.