为什么GC无法在同一方法中运行

Question

I'm playing with gc and found interesting situation when gc(I use parallel gc) isn't involved This is my code

public static void main(String[] args) {
       System.out.println(Runtime.getRuntime().freeMemory() / (1024 * 1024) + " free");

    String[] strings = new String[(40 * 1024 * 1024) / Integer.BYTES];
    System.out.println(strings.length);

    System.out.println(Runtime.getRuntime().freeMemory() / (1024 * 1024) + " free");

    strings = new String[(40 * 1024 * 1024) / Integer.BYTES];
    System.out.println(strings.length);

    System.out.println(Runtime.getRuntime().freeMemory() / (1024 * 1024) + " free");

    }

My Java version is :

java version "1.8.0_191"
Java(TM) SE Runtime Environment (build 1.8.0_191-b12)
Java HotSpot(TM) 64-Bit Server VM (build 25.191-b12, mixed mode)

This program was involved with the following argc:

-Xmx64m -XX:+PrintGCDetails -XX:+PrintGCDateStamps -XX:+PrintGCTimeStamps

This is the output:

59 free
10485760
19 free
2019-06-09T14:53:18.868+0600: 0.105: [GC (Allocation Failure) [PSYoungGen: 1640K->480K(18944K)] 42600K->41448K(62976K), 0.0018199 secs] [Times: user=0.00 sys=0.00, real=0.00 secs] 
2019-06-09T14:53:18.870+0600: 0.107: [Full GC (Ergonomics) [PSYoungGen: 480K->0K(18944K)] [ParOldGen: 40968K->41339K(44032K)] 41448K->41339K(62976K), [Metaspace: 3033K->3033K(1056768K)], 0.0174681 secs] [Times: user=0.04 sys=0.00, real=0.02 secs] 
2019-06-09T14:53:18.888+0600: 0.125: [GC (Allocation Failure) [PSYoungGen: 0K->0K(18944K)] 41339K->41339K(62976K), 0.0017270 secs] [Times: user=0.00 sys=0.00, real=0.00 secs] 
2019-06-09T14:53:18.889+0600: 0.126: [Full GC (Allocation Failure) [PSYoungGen: 0K->0K(18944K)] [ParOldGen: 41339K->41321K(44032K)] 41339K->41321K(62976K), [Metaspace: 3033K->3033K(1056768K)], 0.0140842 secs] [Times: user=0.03 sys=0.01, real=0.01 secs] 
Heap
 PSYoungGen      total 18944K, used 491K [0x00000000feb00000, 0x0000000100000000, 0x0000000100000000)
  eden space 16384K, 3% used [0x00000000feb00000,0x00000000feb7afa0,0x00000000ffb00000)
  from space 2560K, 0% used [0x00000000ffd80000,0x00000000ffd80000,0x0000000100000000)
  to   space 2560K, 0% used [0x00000000ffb00000,0x00000000ffb00000,0x00000000ffd80000)
 ParOldGen       total 44032K, used 41321K [0x00000000fc000000, 0x00000000feb00000, 0x00000000feb00000)
  object space 44032K, 93% used [0x00000000fc000000,0x00000000fe85a6d0,0x00000000feb00000)
 Metaspace       used 3064K, capacity 4496K, committed 4864K, reserved 1056768K
  class space    used 336K, capacity 388K, committed 512K, reserved 1048576K
Exception in thread "main" java.lang.OutOfMemoryError: Java heap space
    at lightbox.GcWIthoutComp.main(GcWIthoutComp.java:11)

According to logs , JVM was not able to allocate memory for string second time ,but if I add JVM arg -Xcomp , it will work and provide this output

59 free
10485760
19 free
2019-06-09T15:01:06.593+0600: 0.830: [GC (Allocation Failure) [PSYoungGen: 1982K->512K(18944K)] 42942K->41480K(62976K), 0.0008554 secs] [Times: user=0.01 sys=0.00, real=0.00 secs] 
2019-06-09T15:01:06.594+0600: 0.831: [Full GC (Ergonomics) [PSYoungGen: 512K->0K(18944K)] [ParOldGen: 40968K->411K(37376K)] 41480K->411K(56320K), [Metaspace: 3377K->3377K(1056768K)], 0.0100865 secs] [Times: user=0.01 sys=0.00, real=0.02 secs] 
10485760
20 free
Heap
 PSYoungGen      total 18944K, used 1147K [0x00000000feb00000, 0x0000000100000000, 0x0000000100000000)
  eden space 16384K, 7% used [0x00000000feb00000,0x00000000fec1ed48,0x00000000ffb00000)
  from space 2560K, 0% used [0x00000000ffb00000,0x00000000ffb00000,0x00000000ffd80000)
  to   space 2560K, 0% used [0x00000000ffd80000,0x00000000ffd80000,0x0000000100000000)
 ParOldGen       total 44032K, used 41371K [0x00000000fc000000, 0x00000000feb00000, 0x00000000feb00000)
  object space 44032K, 93% used [0x00000000fc000000,0x00000000fe866c90,0x00000000feb00000)
 Metaspace       used 3399K, capacity 4500K, committed 4864K, reserved 1056768K
  class space    used 334K, capacity 388K, committed 512K, reserved 1048576K

As I understand compiled code is much comfortable for GC in terms of detecting unused memory. Moreover if I remove -Xcomp and move array creation to separate method it will not throw OOM exception:

public static void main(String[] args) {
        System.out.println(Runtime.getRuntime().freeMemory() / (1024 * 1024) + " free");

        allocate();

        System.out.println(Runtime.getRuntime().freeMemory() / (1024 * 1024) + " free");
        allocate();
        System.out.println(Runtime.getRuntime().freeMemory() / (1024 * 1024) + " free");

    }

    private static void allocate() {
        String[] strings = new String[(40 * 1024 * 1024) / Integer.BYTES];
        System.out.println(strings.length);
    }

My questions are:

1. Why does -Xcomp solve problem with allocation as was shown above?

2. Why does moving allocation to separate method solved my problem?

If you have helpful links, please provide them in comments.

Answer 1

The comment is just placed in one line and it is hard to explain something in there. The important difference seems to be, that with -Xcomp, the local variable strings is already available to be collected. So you get:

2019-06-09T15:01:06.593+0600: 0.830: [GC (Allocation Failure) [PSYoungGen: 1982K->512K(18944K)] 42942K->41480K(62976K), 0.0008554 secs] [Times: user=0.01 sys=0.00, real=0.00 secs] 
2019-06-09T15:01:06.594+0600: 0.831: [Full GC (Ergonomics) [PSYoungGen: 512K->0K(18944K)] [ParOldGen: 40968K->411K(37376K)] 41480K->411K(56320K), [Metaspace: 3377K->3377K(1056768K)], 0.0100865 secs] [Times: user=0.01 sys=0.00, real=0.02 secs] 

So you get an Allocation Failure but the GC can free enough memory.

Without -Xcomp, the loal variable strings cannot be collected so the message you get is something like:

2019-06-09T14:53:18.868+0600: 0.105: [GC (Allocation Failure) [PSYoungGen: 1640K->480K(18944K)] 42600K->41448K(62976K), 0.0018199 secs] [Times: user=0.00 sys=0.00, real=0.00 secs] 
2019-06-09T14:53:18.870+0600: 0.107: [Full GC (Ergonomics) [PSYoungGen: 480K->0K(18944K)] [ParOldGen: 40968K->41339K(44032K)] 41448K->41339K(62976K), [Metaspace: 3033K->3033K(1056768K)], 0.0174681 secs] [Times: user=0.04 sys=0.00, real=0.02 secs] 
2019-06-09T14:53:18.888+0600: 0.125: [GC (Allocation Failure) [PSYoungGen: 0K->0K(18944K)] 41339K->41339K(62976K), 0.0017270 secs] [Times: user=0.00 sys=0.00, real=0.00 secs] 
2019-06-09T14:53:18.889+0600: 0.126: [Full GC (Allocation Failure) [PSYoungGen: 0K->0K(18944K)] [ParOldGen: 41339K->41321K(44032K)] 41339K->41321K(62976K), [Metaspace: 3033K->3033K(1056768K)], 0.0140842 secs] [Times: user=0.03 sys=0.01, real=0.01 secs] 

which shows you that the GC tries to free memory but cannot do it.

(Look at the PSYoungGen: It shows how the memory comes down. When it failed without -Xcomp, not much was possible. With -Xcomp it gets down from 1982K to 512K. Also see https://dzone.com/articles/understanding-garbage-collection-log )

I was unable to find the exact stuff, that -Xcomp is doing. So I am sorry that I can only give you the interpretation of the log right now. So it is like setting strings to null before assigning a newly generated array. So if you add a strings = null; before assigning the array a second time: Even without -Xcomp you get more or less the same message of the gc.

And regarding using it inside a method: You have a local variable inside your method. So you assign the array and then the method ends and the local variable can be collected by the garbage collector.