java - G1GC - Tenured 如何在没有混合 GC 的情况下得到清理

Question

我创建了一个测试 Java 程序来帮助了解 G1GC 的工作原理。

我的程序有两个线程 - 一个线程加载一个包含大约 200MB 对象(仅 1000 个字符串)的大型 ArrayList，然后删除一个元素并在循环中添加一个元素，其中包含 sleep。这里的想法是让一些东西慢慢老化并最终获得终身教职。

另一个线程只是在 while 循环中创建一个大字符串，以创建大量垃圾以导致 GC 运行。

这一切似乎都工作正常，但令我困惑的是，tenured 中的浪费似乎在没有运行混合 GC 的情况下得到了清理。

   [Eden: 581.0M(581.0M)->0.0B(573.0M) Survivors: 9216.0K->9216.0K Heap: 904.3M(1024.0M)->330.9M(1024.0M)]
 [Times: user=0.03 sys=0.00, real=0.00 secs]
2017-09-28T16:22:14.889+0000: 2.415: 2017-09-28T16:22:14.889+0000[GC concurrent-root-region-scan-start]
: 2.415: Total time for which application threads were stopped: 0.0081917 seconds, Stopping threads took: 0.0000852 seconds
2017-09-28T16:22:14.889+0000: 2.415: [GC concurrent-root-region-scan-end, 0.0002779 secs]
2017-09-28T16:22:14.889+0000: 2.415: [GC concurrent-mark-start]
2017-09-28T16:22:14.904+0000: 2.430: [GC concurrent-mark-end, 0.0143310 secs]
2017-09-28T16:22:14.905+0000: 2.431: [GC remark 2017-09-28T16:22:14.905+0000: 2.431: [Finalize Marking, 0.0003653 secs] 2017-09-28T16:22:14.905+0000: 2.431: [GC ref-proc2017-09-28T16:22:14.905+0000: 2.431: [SoftReference, 0 refs, 0.0008964 secs]2017-09-28T16:22:14.906+0000: 2.432: [WeakReference, 0 refs, 0.0005380 secs]2017-09-28T16:22:14.907+0000: 2.433: [FinalReference, 2 refs, 0.0005460 secs]2017-09-28T16:22:14.907+0000: 2.434: [PhantomReference, 0 refs, 0 refs, 0.0010243 secs]2017-09-28T16:22:14.908+0000: 2.435: [JNI Weak Reference, 0.0000090 secs], 0.0034419 secs] 2017-09-28T16:22:14.909+0000: 2.435: [Unloading, 0.0008396 secs], 0.0057362 secs]
 [Times: user=0.03 sys=0.00, real=0.01 secs]
2017-09-28T16:22:14.911+0000: 2.437: Total time for which application threads were stopped: 0.0069810 seconds, Stopping threads took: 0.0011536 seconds
2017-09-28T16:22:14.911+0000: 2.437: [GC cleanup 418M->302M(1024M), 0.0014100 secs]
 [Times: user=0.00 sys=0.00, real=0.00 secs]
2017-09-28T16:22:14.912+0000: 2.438: Total time for which application threads were stopped: 0.0015190 seconds, Stopping threads took: 0.0000697 seconds
2017-09-28T16:22:14.912+0000: 2.438: [GC concurrent-cleanup-start]
2017-09-28T16:22:14.912+0000: 2.439: [GC concurrent-cleanup-end, 0.0000987 secs]
2017-09-28T16:22:14.987+0000: 2.513: [GC pause (G1 Evacuation Pause) (young) 2.513: [G1Ergonomics (CSet Construction) start choosing CSet, _pending_cards: 4816, predicted base time: 5.38 ms, remaining time: 44.62 ms, target pause time: 50.00 ms]
 2.513: [G1Ergonomics (CSet Construction) add young regions to CSet, eden: 573 regions, survivors: 9 regions, predicted young region time: 3.54 ms]
 2.513: [G1Ergonomics (CSet Construction) finish choosing CSet, eden: 573 regions, survivors: 9 regions, old: 0 regions, predicted pause time: 8.92 ms, target pause time: 50.00 ms]
2017-09-28T16:22:14.991+0000: 2.517: [SoftReference, 0 refs, 0.0010742 secs]2017-09-28T16:22:14.992+0000: 2.518: [WeakReference, 0 refs, 0.0006150 secs]2017-09-28T16:22:14.992+0000: 2.519: [FinalReference, 0 refs, 0.0004760 secs]2017-09-28T16:22:14.993+0000: 2.519: [PhantomReference, 0 refs, 0 refs, 0.0008707 secs]2017-09-28T16:22:14.994+0000: 2.520: [JNI Weak Reference, 0.0000094 secs] 2.521: [G1Ergonomics (Mixed GCs) do not start mixed GCs, reason: reclaimable percentage not over threshold, candidate old regions: 14 regions, reclaimable: 13071400 bytes (1.22 %), threshold: 5.00 %]
, 0.0077749 secs]
   [Parallel Time: 3.6 ms, GC Workers: 16]
      [GC Worker Start (ms): Min: 2513.1, Avg: 2513.6, Max: 2515.2, Diff: 2.1]
      [Ext Root Scanning (ms): Min: 0.0, Avg: 0.1, Max: 0.2, Diff: 0.1, Sum: 0.9]
      [Update RS (ms): Min: 0.0, Avg: 0.4, Max: 1.1, Diff: 1.1, Sum: 6.3]
         [Processed Buffers: Min: 0, Avg: 1.2, Max: 4, Diff: 4, Sum: 20]
      [Scan RS (ms): Min: 0.0, Avg: 0.0, Max: 0.1, Diff: 0.1, Sum: 0.8]
      [Code Root Scanning (ms): Min: 0.0, Avg: 0.0, Max: 0.0, Diff: 0.0, Sum: 0.1]
      [Object Copy (ms): Min: 1.2, Avg: 2.3, Max: 2.9, Diff: 1.7, Sum: 36.9]
      [Termination (ms): Min: 0.0, Avg: 0.1, Max: 0.2, Diff: 0.2, Sum: 2.1]
         [Termination Attempts: Min: 1, Avg: 1.1, Max: 2, Diff: 1, Sum: 18]
      [GC Worker Other (ms): Min: 0.0, Avg: 0.0, Max: 0.1, Diff: 0.0, Sum: 0.3]
      [GC Worker Total (ms): Min: 1.4, Avg: 3.0, Max: 3.5, Diff: 2.1, Sum: 47.3]
      [GC Worker End (ms): Min: 2516.5, Avg: 2516.6, Max: 2516.6, Diff: 0.1]
   [Code Root Fixup: 0.0 ms]
   [Code Root Purge: 0.0 ms]
   [Clear CT: 0.2 ms]
   [Other: 4.0 ms]
      [Choose CSet: 0.0 ms]
      [Ref Proc: 3.4 ms]
      [Ref Enq: 0.2 ms]
      [Redirty Cards: 0.1 ms]
      [Humongous Register: 0.0 ms]
      [Humongous Reclaim: 0.0 ms]
      [Free CSet: 0.2 ms]
   [Eden: 573.0M(573.0M)->0.0B(607.0M) Survivors: 9216.0K->7168.0K Heap: 787.9M(1024.0M)->221.4M(1024.0M)]

这是 GC 日志的片段。请注意，在这次运行之间:

[Eden: 581.0M(581.0M)->0.0B(573.0M) Survivors: 9216.0K->9216.0K Heap: 904.3M(1024.0M)->330.9M(1024.0M)]
...
[Eden: 573.0M(573.0M)->0.0B(607.0M) Survivors: 9216.0K->7168.0K Heap: 787.9M(1024.0M)->221.4M(1024.0M)]

因此，在这次运行中，Tenured 中的使用率似乎有所下降。然而，没有提到混合GC，事实上它给出的只是:

[G1Ergonomics (Mixed GCs) do not start mixed GCs, reason: reclaimable percentage not over threshold, candidate old regions: 14 regions, reclaimable: 13071400 bytes (1.22 %), threshold: 5.00 %]

所以我的问题是 - 如何在不发生混合(或完整)GC 的情况下清理永久空间？

Answer 1

不需要混合集合来撤出已发现并发循环完全未引用的旧区域。

2017-09-28T16:22:14.911+0000: 2.437: [GC cleanup 418M->302M(1024M), 0.0014100 secs]

还有一些额外的日志记录标志，例如-XX:+G1PrintHeapRegions，可用于跟踪此类清理。请参阅oracle blog关于 G1GC 日志的深入解释。