[转]java.lang.OutOfMemoryError：GC overhead limit exceeded

我遇到这样的问题，本地部署时抛出异常java.lang.OutOfMemoryError：GC overhead limit exceeded导致服务起不来，查看日志发现加载了太多资源到内存，本地的性能也不好，gc时间消耗的较多。解决这种问题两种方法是，增加参数，-XX:-UseGCOverheadLimit，关闭这个特性，同时增加heap大小，-Xmx1024m。坑填了，but why？

OOM大家都知道，就是JVM内存溢出了，那GC overhead limit exceed呢？

GC overhead limt exceed检查是Hotspot VM 1.6定义的一个策略，通过统计GC时间来预测是否要OOM了，提前抛出异常，防止OOM发生。Sun 官方对此的定义是：“并行/并发回收器在GC回收时间过长时会抛出OutOfMemroyError。过长的定义是，超过98%的时间用来做GC并且回收了不到2%的堆内存。用来避免内存过小造成应用不能正常工作。“

听起来没啥用...预测OOM有啥用？起初开来这玩意只能用来Catch住释放内存资源，避免应用挂掉。后来发现一般情况下这个策略不能拯救你的应用，但是可以在应用挂掉之前做最后的挣扎，比如数据保存或者保存现场（Heap Dump）。

而且有些时候这个策略还会带来问题，比如加载某个大的内存数据时频繁OOM。

假如你也生产环境中遇到了这个问题，在不知道原因时不要简单的猜测和规避。可以通过-verbose:gc -XX:+PrintGCDetails看下到底什么原因造成了异常。通常原因都是因为old区占用过多导致频繁Full GC，最终导致GC overhead limit exceed。如果gc log不够可以借助于JProfile等工具查看内存的占用，old区是否有内存泄露。分析内存泄露还有一个方法-XX:+HeapDumpOnOutOfMemoryError，这样OOM时会自动做Heap Dump，可以拿MAT来排查了。还要留意young区，如果有过多短暂对象分配，可能也会抛这个异常。

日志的信息不难理解，就是每次gc时打条日志，记录GC的类型，前后大小和时间。举个例子。

33.125: [GC [DefNew: 16000K->16000K(16192K), 0.0000574 secs][Tenured: 2973K->2704K(16384K), 0.1012650 secs] 18973K->2704K(32576K), 0.1015066 secs]

100.667:[Full GC [Tenured: 0K->210K(10240K), 0.0149142 secs] 4603K->210K(19456K), [Perm : 2999K->2999K(21248K)], 0.0150007 secs]

GC和Full GC代表gc的停顿类型，Full GC代表stop-the-world。箭头两边是gc前后的区空间大小，分别是young区、tenured区和perm区，括号里是该区的总大小。冒号前面是gc发生的时间，单位是秒，从jvm启动开始计算。DefNew代表Serial收集器，为Default New Generation的缩写，类似的还有PSYoungGen，代表Parallel Scavenge收集器。这样可以通过分析日志找到导致GC overhead limit exceeded的原因，通过调节相应的参数解决问题。

文中涉及到的名词解释，

Eden Space：堆内存池，大多数对象在这里分配内存空间。

Survivor Space：堆内存池，存储在Eden Space的gc中存活下来的对象。

Tenured Generation：堆内存池，存储Survivor Space中存活过几次gc的对象。

Permanent Generation：非堆空间，存储的是class和method对象。

Code Cache：非堆空间，JVM用来存储编译和存储native code。

最后附上GC overhead limit exceed HotSpot的实现：

  1 bool print_gc_overhead_limit_would_be_exceeded = false;
  2 
  3 if (is_full_gc) {
  4 
  5   if (gc_cost() > gc_cost_limit &&
  6 
  7     free_in_old_gen < (size_t) mem_free_old_limit &&
  8 
  9     free_in_eden < (size_t) mem_free_eden_limit) {
 10 
 11     // Collections, on average, are taking too much time, and
 12 
 13     //      gc_cost() > gc_cost_limit
 14 
 15     // we have too little space available after a full gc.
 16 
 17     //      total_free_limit < mem_free_limit
 18 
 19     // where
 20 
 21     //   total_free_limit is the free space available in
 22 
 23     //     both generations
 24 
 25     //   total_mem is the total space available for allocation
 26 
 27     //     in both generations (survivor spaces are not included
 28 
 29     //     just as they are not included in eden_limit).
 30 
 31     //   mem_free_limit is a fraction of total_mem judged to be an
 32 
 33     //     acceptable amount that is still unused.
 34 
 35     // The heap can ask for the value of this variable when deciding
 36 
 37     // whether to thrown an OutOfMemory error.
 38 
 39     // Note that the gc time limit test only works for the collections
 40 
 41     // of the young gen + tenured gen and not for collections of the
 42 
 43     // permanent gen.  That is because the calculation of the space
 44 
 45     // freed by the collection is the free space in the young gen +
 46 
 47     // tenured gen.
 48 
 49     // At this point the GC overhead limit is being exceeded.
 50 
 51     inc_gc_overhead_limit_count();
 52 
 53     if (UseGCOverheadLimit) {
 54 
 55       if (gc_overhead_limit_count() >=
 56 
 57           AdaptiveSizePolicyGCTimeLimitThreshold){
 58 
 59         // All conditions have been met for throwing an out-of-memory
 60 
 61         set_gc_overhead_limit_exceeded(true);
 62 
 63         // Avoid consecutive OOM due to the gc time limit by resetting
 64 
 65         // the counter.
 66 
 67         reset_gc_overhead_limit_count();
 68 
 69       } else {
 70 
 71         // The required consecutive collections which exceed the
 72 
 73         // GC time limit may or may not have been reached. We
 74 
 75         // are approaching that condition and so as not to
 76 
 77         // throw an out-of-memory before all SoftRef's have been
 78 
 79         // cleared, set _should_clear_all_soft_refs in CollectorPolicy.
 80 
 81         // The clearing will be done on the next GC.
 82 
 83         bool near_limit = gc_overhead_limit_near();
 84 
 85         if (near_limit) {
 86 
 87           collector_policy->set_should_clear_all_soft_refs(true);
 88 
 89           if (PrintGCDetails && Verbose) {
 90 
 91             gclog_or_tty->print_cr("  Nearing GC overhead limit, "
 92 
 93               "will be clearing all SoftReference");
 94 
 95           }
 96 
 97         }
 98 
 99       }
100 
101     }
102 
103     // Set this even when the overhead limit will not
104 
105     // cause an out-of-memory.  Diagnostic message indicating
106 
107     // that the overhead limit is being exceeded is sometimes
108 
109     // printed.
110 
111     print_gc_overhead_limit_would_be_exceeded = true;
112 
113  
114 
115   } else {
116 
117     // Did not exceed overhead limits
118 
119     reset_gc_overhead_limit_count();
120 
121   }
122 
123 }

参照&延伸阅读：

http://javaeesupportpatterns.blogspot.com/2012/01/gc-overhead-limit-exceeded-understand.html

http://www.oracle.com/technetwork/java/javase/gc-tuning-6-140523.html

http://reins.altervista.org/java/gc1.4.2_example.html

http://stackoverflow.com/questions/2129044/java-heap-terminology-young-old-and-permanent-generations

http://book.51cto.com/art/201306/399236.htm

https://blogs.oracle.com/jonthecollector/entry/presenting_the_permanent_generation

转自http://www.cnblogs.com/hucn/p/3572384.html