CMS主要gc频繁发生

Question

我使用的是CMS GC，主要gc每2分钟或更少发生一次

我尝试将CMSMaxAbortablePrecleanTime的值从5增加到10，然后再增加到20，但没有帮助。

有人能帮上忙吗？

下面是我使用的配置

-heap-size=30g 
-XX:PermSize=512m 
-XX:MaxPermSize=512m 
-XX:+UseCompressedOops -Xss256k 
-XX:+UnlockDiagnosticVMOptions 
-XX:ParGCCardsPerStrideChunk=32768 
-XX:NewSize=2g 
-XX:MaxNewSize=2g 
-XX:CMSInitiatingOccupancyFraction=45 
-XX:+UseCMSInitiatingOccupancyOnly 
-XX:CMSMaxAbortablePrecleanTime=20000

这是GC的日志

2016-10-11T08:49:01.983-0500: 407.440: [GC [1 CMS-initial-mark: 13557029K(29360128K)] 13766753K(31247616K), 0.4483970 secs] [Times: user=0.10 sys=0.31, real=0.45 secs]
2016-10-11T08:49:02.431-0500: 407.889: [CMS-concurrent-mark-start]
2016-10-11T08:49:07.764-0500: 413.222: [GC2016-10-11T08:49:07.765-0500: 413.222: [ParNew: 1887488K->209664K(1887488K), 4.2307640 secs] 15444517K->14525267K(31247616K), 4.2309270 secs] [Times: user=4.32 sys=3.49, real=4.23 secs]
2016-10-11T08:49:22.023-0500: 427.480: [GC2016-10-11T08:49:22.023-0500: 427.480: [ParNew: 1887488K->209664K(1887488K), 1.9675120 secs] 16203091K->14978753K(31247616K), 1.9676860 secs] [Times: user=3.20 sys=1.50, real=1.96 secs]
2016-10-11T08:49:28.899-0500: 434.356: [CMS-concurrent-mark: 20.218/26.467 secs] [Times: user=58.98 sys=12.86, real=26.46 secs]
2016-10-11T08:49:28.899-0500: 434.356: [CMS-concurrent-preclean-start]
2016-10-11T08:49:28.997-0500: 434.454: [CMS-concurrent-preclean: 0.096/0.098 secs] [Times: user=0.28 sys=0.01, real=0.10 secs]
2016-10-11T08:49:28.997-0500: 434.454: [CMS-concurrent-abortable-preclean-start]
2016-10-11T08:49:34.215-0500: 439.672: [GC2016-10-11T08:49:34.215-0500: 439.672: [ParNew: 1887488K->209664K(1887488K), 1.9914300 secs] 16656577K->15383793K(31247616K), 1.9915960 secs] [Times: user=3.22 sys=1.24, real=1.99 secs]
2016-10-11T08:49:40.097-0500: 445.554: [CMS-concurrent-abortable-preclean: 9.002/11.100 secs] [Times: user=29.07 sys=3.17, real=11.10 secs]
2016-10-11T08:49:40.098-0500: 445.555: [GC[YG occupancy: 1113559 K (1887488 K)]2016-10-11T08:49:40.098-0500: 445.555: [Rescan (parallel) , 0.6648450 secs]2016-10-11T08:49:40.763-0500: 446.220: [weak refs processing, 0.0005380 secs]2016-10-11T08:49:40.764-0500: 446.221: [class unloading, 0.0115730 secs]2016-10-11T08:49:40.775-0500: 446.233: [scrub symbol table, 0.0120750 secs]2016-10-11T08:49:40.788-0500: 446.245: [scrub string table, 0.0011940 secs] [1 CMS-remark: 15174129K(29360128K)] 16287688K(31247616K), 0.6943980 secs] [Times: user=2.65 sys=0.00, real=0.69 secs]
2016-10-11T08:49:40.793-0500: 446.250: [CMS-concurrent-sweep-start]
2016-10-11T08:49:43.096-0500: 448.553: [GC2016-10-11T08:49:43.096-0500: 448.553: [ParNew: 1887318K->209664K(1887488K), 0.4514140 secs] 16282520K->14815052K(31247616K), 0.4515870 secs] [Times: user=1.67 sys=0.00, real=0.45 secs]
2016-10-11T08:49:46.798-0500: 452.255: [CMS-concurrent-sweep: 5.552/6.005 secs] [Times: user=19.43 sys=1.14, real=6.00 secs]
2016-10-11T08:49:46.798-0500: 452.255: [CMS-concurrent-reset-start]
2016-10-11T08:49:47.236-0500: 452.693: [CMS-concurrent-reset: 0.438/0.438 secs] [Times: user=0.79 sys=0.50, real=0.44 secs]
2016-10-11T08:49:49.244-0500: 454.701: [GC [1 CMS-initial-mark: 13969830K(29360128K)] 15630493K(31247616K), 2.1488470 secs] [Times: user=2.15 sys=0.00, real=2.15 secs]
2016-10-11T08:49:51.394-0500: 456.851: [CMS-concurrent-mark-start]

Answer 1

根据选项CMSInitiatingOccupancyFraction，您的标记阶段将从堆的45%开始，在本例中为14495514K。

根据您的GC日志：

GC2016-10-11T08:49:07.764-0500: 413.222：

2016-10-11T08:49:07.765-0500: 413.222：[ParNew: 1887488K->209664K(1887488K)，4.2307640秒] GC2016-10-11T08:49:07.765-0500: 4.2309270秒

完整GC后的实时数据集为14525267K，高于上述阈值，因此您面临的GC几乎是恒定的。

您应该增加CMSInitiatingOccupancyFraction，以便它可以容纳您的实时数据集。它应该比你的实时数据集+年轻一代的堆大小高一点，在你的情况下大约是16,5G -我想说设置CMSInitiatingOccupancyFraction=55应该会改善你的情况。