java gc full gc_记一次Java服务频繁Full GC的排查过程

现象

从监控来看，堆内存是够用的，但是频繁触发Full GC，每秒钟三次，每次耗时三四秒。

image.png

结合Young GC的信息和堆内存的使用情况，可以发现新生代的内存够用，老生代的内存不够用，频繁Full GC，老生代的内存使用率依旧达到了98%。

2018-11-20T15:02:46.002+0800: 2779214.232: [GC2018-11-20T15:02:46.002+0800: 2779214.232: [ParNew: 1643604K->4513K(1843200K), 0.0129920 secs] 4994360K->3355569K(5257216K), 0.0131040 secs] [Times: user=0.05 sys=0.00, real=0.01 secs]

$ sudo jmap -heap 11927

Attaching to process ID 11927, please wait...

Debugger attached successfully.

Server compiler detected.

JVM version is 24.65-b04-internal

using parallel threads in the new generation.

using thread-local object allocation.

Concurrent Mark-Sweep GC

Heap Configuration:

MinHeapFreeRatio = 40

MaxHeapFreeRatio = 70

MaxHeapSize = 5593104384 (5334.0MB)

NewSize = 2097152000 (2000.0MB)

MaxNewSize = 2097152000 (2000.0MB)

OldSize = 5439488 (5.1875MB)

NewRatio = 2

SurvivorRatio = 8

PermSize = 536870912 (512.0MB)

MaxPermSize = 536870912 (512.0MB)

G1HeapRegionSize = 0 (0.0MB)

Heap Usage:

New Generation (Eden + 1 Survivor Space):

capacity = 1887436800 (1800.0MB)

used = 899619040 (857.9435729980469MB)

free = 987817760 (942.0564270019531MB)

47.66353183322482% used

Eden Space:

capacity = 1677721600 (1600.0MB)

used = 895864384 (854.3628540039062MB)

free = 781857216 (745.6371459960938MB)

53.39767837524414% used

From Space:

capacity = 209715200 (200.0MB)

used = 3754656 (3.580718994140625MB)

free = 205960544 (196.41928100585938MB)

1.7903594970703125% used

To Space:

capacity = 209715200 (200.0MB)

used = 0 (0.0MB)

free = 209715200 (200.0MB)

0.0% used

concurrent mark-sweep generation:

capacity = 3495952384 (3334.0MB)

used = 3432530640 (3273.516311645508MB)

free = 63421744 (60.48368835449219MB)

98.18585217892945% used

Perm Generation:

capacity = 536870912 (512.0MB)

used = 52807616 (50.36126708984375MB)

free = 484063296 (461.63873291015625MB)

9.836184978485107% used

28784 interned Strings occupying 3175392 bytes.

排查过程

这种现象表明，服务存在内存泄露，所以需要分析堆的使用。使用gcore产生core文件，这比直接使用jmap dump快得多，可以避免对线上服务造成影响。core文件的大小跟进程使用的堆大小一样大。

$ sudo gcore 11927

[Thread debugging using libthread_db enabled]

[New Thread 0x487ac940 (LWP 1653)]

[New Thread 0x486ab940 (LWP 1652)]

[New Thread 0x485aa940 (LWP 1651)]

0x00000030f0e07ba5 in pthread_join () from /lib64/libpthread.so.0

Saved corefile core.11927

使用jmap将core文件转换成bin文件。hprof文件跟进程实际使用的堆内存一样大。

$sudo jmap -dump:format=b,file=heap.hprof.bin /opt/taobao/java/bin/java core.11927

Attaching to core core.11927 from executable /opt/taobao/java/bin/java, please wait...

Debugger attached successfully.

Server compiler detected.

JVM version is 24.65-b04-internal

Dumping heap to heap.hprof.bin ...

这个转换过程非常慢。将5.3GB的core转换成4GB的hprof文件，花了将近半个小时。通过top来看，这个dump是单线程做的，一直跑满了一个核。

top - 15:43:38 up 214 days, 1:02, 2 users, load average: 1.69, 1.83, 1.56

Tasks: 142 total, 1 running, 141 sleeping, 0 stopped, 0 zombie

Cpu(s): 51.5%us, 0.8%sy, 0.0%ni, 47.7%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st

Mem: 8194164k total, 8135628k used, 58536k free, 75324k buffers

Swap: 0k total, 0k used, 0k free, 797688k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND

11927 www 20 0 6513m 5.3g 82m S 105.5 67.4 10760:36 java

3764 root 20 0 2458m 791m 12m S 100.5 9.9 18:53.36 jmap

16376 root 20 0 411m 360m 1468 S 0.0 4.5 19:52.15 syslog-ng

11190 agent 20 0 540m 143m 5832 S 0.0 1.8 65:14.77 java

接着就是如何分析这个hprof文件的问题。目前主流的分析工具，有MAT和jhat，下面分别使用二者分析一下。

jhat

jhat看起来比较简单，从分析结果里面可以看出用户自定义类中，SlsRecordWriter类的对象比较多，再做进一步的分析就不知如何下手了。

image.png

image.png

MAT

MAT体验比较好，直接分析出持有内存最大的类，然后从dominator_tree里面可以看到具体的应用情况，可以发现SlsRecordWriter对象很多。需要注意的是，使用MAT分析大的dump文件很吃内存，所以需要调大堆内存。Mac上Eclipse的配置文件路径是/Applications/Eclipse.app/Contents/Eclipse/eclipse.ini。

image.png

image.png

结论

构造一个demo验证一下。下面这个例子，运行一段时间便会触发Young GC(IDEA VM options设置上"-verbose:gc -XX:+PrintGCDetails")，但是从来不会调用Foo的finalize方法，只有停止运行的时候，才会进入hook执行，接着释放对象。如果注释掉addShutdownHook这段代码，就符合期望了。所以addShutdownHook会导致Runtime持有对象，需要注意使用场景。

class Foo {

private String bar = new String("fdsaflkjdsalfjldsajflsajfoiwefnaslfjlaskejfowieajfalfjelajfieslkjflksdajlf");

private int num = 0;

public Foo() {

}

public Foo(int num) {

System.out.println(String.format("Foo initialize: %d", num));

this.num = num;

Runtime.getRuntime().addShutdownHook(new Thread(new Runnable() {

public void run() {

clean();

}

}));

}

void clean() {

System.out.println(String.format("Foo clean: %d", num));

}

@Override

protected void finalize() throws Throwable {

super.finalize();

System.out.println(String.format("foo finalize: %d", this.num));

}

}

public class Demo {

public static void main(String[] args) {

int num = 0;

while (true) {

try {

Foo foo = new Foo(num);

num += 1;

Thread.sleep(100);

}catch (Exception e) {

}

}

}

}

参考资料