CODEWITHSUNDEEP
javagarbage-collectionjvm
Andrew
Andrew

Reputation: 1002

Force JVM to thrash due to constant GC

I am working on testing how a Java application handles under stress. One of the scenarios I am looking to test is when the JVM is thrashing because of constant garbage collection. One of the conditions I have seen is that a slow memory leak will cause the JVM to enter a state were it will continually spend over 95% of its processor time performing GC. In this state, the JVM is not executing application code. I am interested in writing a Java method that, when executed, will induce this sort of GC behavior. How might I do this?

Upvotes: 2

Views: 1980

Answers (1)

Andreas
Andreas

Reputation: 159086

Create a lot of objects until OutOfMemoryError occurs, then release a few of them. Now enter an allocate-and-release loop.

Example

final int BLOCK_SIZE = 1000000;
List<byte[]> list = new LinkedList<>();
try {
    for (;;) {
        list.add(new byte[BLOCK_SIZE]);
        if (list.size() % 1000 == 0)
            System.out.println(list.size());
    }
} catch (@SuppressWarnings("unused") OutOfMemoryError e) {
    // Ignore
}
list.remove(0); // free some memory
System.out.println((list.size() + 1) + " Memory full");
for (int i = 0; i < 15; i++) {
    System.out.println(i);
    list.add(new byte[BLOCK_SIZE]);
    list.remove(0);
}
System.out.println("Done");

Output (run with -XX:+PrintGC -XX:+PrintGCTimeStamps)

0.089: [GC (Allocation Failure)  131241K->128483K(502784K), 0.0231605 secs]
0.123: [GC (Allocation Failure)  259971K->259265K(634368K), 0.0280219 secs]
0.151: [Full GC (Ergonomics)  259265K->259088K(858112K), 0.0481911 secs]
0.233: [GC (Allocation Failure)  522083K->522044K(858112K), 0.0533938 secs]
0.286: [Full GC (Ergonomics)  522044K->521794K(1285632K), 0.0087057 secs]
0.315: [GC (Allocation Failure)  784782K->784749K(1454592K), 0.0514033 secs]
0.366: [Full GC (Ergonomics)  784749K->784500K(1792512K), 0.0095001 secs]
1000
0.425: [GC (Allocation Failure)  1215843K->1215427K(1792512K), 0.0880220 secs]
0.513: [Full GC (Ergonomics)  1215427K->1215181K(2387968K), 0.0134368 secs]
0.558: [GC (Allocation Failure)  1646521K->1646108K(2724352K), 0.0873853 secs]
0.646: [Full GC (Ergonomics)  1646108K->1645863K(3247616K), 0.0134679 secs]
2000
0.754: [GC (Allocation Failure)  2413882K->2413709K(3247616K), 0.1432810 secs]
0.897: [Full GC (Ergonomics)  2413709K->2413471K(4247552K), 0.0180889 secs]
3000
0.972: [GC (Allocation Failure)  3181488K->3181318K(4877312K), 0.1505734 secs]
1.122: [Full GC (Ergonomics)  3181318K->3181080K(5798400K), 0.0207634 secs]
4000
1.321: [GC (Allocation Failure) -- 4579352K->5802028K(5824000K), 0.2306780 secs]
1.552: [Full GC (Ergonomics)  5802028K->4574691K(7000064K), 0.0412160 secs]
5000
1.683: [GC (Allocation Failure) -- 5802028K->6977828K(7000064K), 0.2326833 secs]
1.915: [Full GC (Ergonomics)  6977828K->5793490K(7000064K), 0.0411068 secs]
6000
7000
2.043: [Full GC (Ergonomics)  6977828K->6974201K(7000064K), 0.1676370 secs]
2.211: [Full GC (Ergonomics)  6977828K->6977131K(7000064K), 0.0248826 secs]
2.236: [Full GC (Allocation Failure)  6977131K->6977119K(7000064K), 1.0004488 secs]
7144 Memory full
0
3.237: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 1.1094352 secs]
1
4.346: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 1.0821330 secs]
2
5.429: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 1.0057587 secs]
3
6.435: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 0.9934119 secs]
4
7.428: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 1.0060575 secs]
5
8.435: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 0.9925522 secs]
6
9.427: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 1.0048095 secs]
7
10.432: [Full GC (Ergonomics)  6977816K->6976142K(7000064K), 0.9968477 secs]
8
11.430: [Full GC (Ergonomics)  6977665K->6976142K(7000064K), 1.0109500 secs]
9
12.441: [Full GC (Ergonomics)  6977480K->6976142K(7000064K), 1.0050952 secs]
10
13.446: [Full GC (Ergonomics)  6977361K->6976142K(7000064K), 0.9977489 secs]
11
14.444: [Full GC (Ergonomics)  6977283K->6976142K(7000064K), 1.0184425 secs]
12
15.463: [Full GC (Ergonomics)  6977232K->6976142K(7000064K), 1.0601230 secs]
13
16.523: [Full GC (Ergonomics)  6977199K->6976142K(7000064K), 1.0214882 secs]
14
17.545: [Full GC (Ergonomics)  6977178K->6976142K(7000064K), 1.0136330 secs]
Done

As can be seen, the first loop is fast, about 7000 iterations in 2 seconds.

Once memory is full, the second loop needs a GC run for each iteration, and each GC run was about 1 second.

Upvotes: 4

Related Questions