高峰请求到来时频繁且长时间Java GC,导致应用停止响应10分钟

Frequently and long-time Java GC when peak requests arrives, which makes application stop response for 10 minutes

我有一个 Java 应用程序提供在线视频播放服务。当高峰流量到来时(比如用户在手机phone点击pushwindow),GC开销很大,我可以看到"promotion failed"和"concurrent mode failure"的信息GC日志,而且GC非常耗时,几乎需要10分钟才能完成一个GC周期,导致应用长时间无法响应。

我知道问题应该是内存使用的错误代码设计,谁能帮忙弄清楚如何解决 GC 问题?谢谢。

这是我的服务器环境:

cat /etc/redhat-release
CentOS release 6.8 (Final)

free -h
             total       used       free     shared    buffers     cached
Mem:           31G        31G       302M        16K       260M       6.5G
-/+ buffers/cache:        24G       7.1G
Swap:         2.0G       564M       1.4G

java -version
java version "1.7.0_131"
OpenJDK Runtime Environment (rhel-2.6.9.0.el6_8-x86_64 u131-b00)
OpenJDK 64-Bit Server VM (build 24.131-b00, mixed mode)

这是我原来的 JVM 设置:

55634 XmlConfiguration -Xmx22528m -Xms22528m -verbose:gc -XX:NewRatio=3
-XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:+CMSParallelRemarkEnabled
-Xloggc:/data/logs/jetty/gc.log -XX:GCLogFileSize=10M -XX:NumberOfGCLogFiles=10
-XX:+UseGCLogFileRotation -XX:+PrintGCDateStamps -XX:+PrintGCTimeStamps
-XX:+PrintGCDetails -XX:+DisableExplicitGC -XX:+ScavengeBeforeFullGC
-XX:+CMSScavengeBeforeRemark -XX:+UseCMSInitiatingOccupancyOnly
-XX:CMSInitiatingOccupancyFraction=70
-Djetty.home=/usr/local/jetty-distribution-8.1.16.v20140903

我尝试了一些解决方案,但问题仍然存在。

1)调整JVM参数: -XX:CMSInitiatingOccupancyFraction=50,使CMS GC更早,为应用程序线程提供更多空闲内存。

2) 调整JVM参数:-XX:NewRatio=2,让新生代变大

3) 调整JVM参数:-XX:PermSize=512m,-XX:MaxPermSize=512m,使持久化生成更大

4) 单机限速,可以保护下线应用

这是 GC 日志:

2019-08-17T11:03:01.119+0800: 48511.609: [GC2019-08-17T11:03:01.119+0800: 48511.610: [ParNew: 5132248K->576704K(5190464K), 0.5585610 secs] 13334008K->9118076K(22491968K), 0.5594890 secs] [Times: user=3.56
 sys=0.01, real=0.56 secs]
2019-08-17T11:03:03.773+0800: 48514.264: [GC2019-08-17T11:03:03.774+0800: 48514.264: [ParNew: 5190464K->576704K(5190464K), 0.6915700 secs] 13731836K->9578991K(22491968K), 0.6925000 secs] [Times: user=4.10
 sys=0.01, real=0.69 secs]
2019-08-17T11:03:07.237+0800: 48517.728: [GC2019-08-17T11:03:07.238+0800: 48517.728: [ParNew: 5190464K->396073K(5190464K), 0.5246610 secs] 14192751K->9710333K(22491968K), 0.5255170 secs] [Times: user=2.92
 sys=0.02, real=0.53 secs]
2019-08-17T11:03:10.301+0800: 48520.791: [GC2019-08-17T11:03:10.301+0800: 48520.791: [ParNew: 5009786K->576704K(5190464K), 0.2799780 secs] 14324046K->9906749K(22491968K), 0.2808270 secs] [Times: user=2.09
 sys=0.01, real=0.28 secs]
2019-08-17T11:03:12.171+0800: 48522.662: [GC2019-08-17T11:03:12.172+0800: 48522.662: [ParNew: 5190464K->517429K(5190464K), 0.7336400 secs] 14520509K->10324876K(22491968K), 0.7346510 secs] [Times: user=4.0
5 sys=0.03, real=0.74 secs]
2019-08-17T11:03:14.893+0800: 48525.383: [GC2019-08-17T11:03:14.893+0800: 48525.384: [ParNew: 5131189K->504298K(5190464K), 0.2094840 secs] 14938636K->10311745K(22491968K), 0.2103470 secs] [Times: user=1.6
4 sys=0.00, real=0.21 secs]
2019-08-17T11:03:18.874+0800: 48529.365: [GC2019-08-17T11:03:18.875+0800: 48529.365: [ParNew: 5117791K->576704K(5190464K), 0.6094130 secs] 14925238K->10624093K(22491968K), 0.6103270 secs] [Times: user=4.0
6 sys=0.01, real=0.61 secs]
2019-08-17T11:03:21.653+0800: 48532.143: [GC2019-08-17T11:03:21.653+0800: 48532.144: [ParNew: 5190439K->576704K(5190464K), 0.7770410 secs] 15237828K->11148654K(22491968K), 0.7779820 secs] [Times: user=4.1
4 sys=0.03, real=0.78 secs]
2019-08-17T11:03:24.691+0800: 48535.182: [GC2019-08-17T11:03:24.692+0800: 48535.182: [ParNew: 5190464K->576704K(5190464K), 0.6323050 secs] 15762414K->11476700K(22491968K), 0.6331560 secs] [Times: user=4.2
2 sys=0.06, real=0.64 secs]
2019-08-17T11:03:27.652+0800: 48538.142: [GC2019-08-17T11:03:27.652+0800: 48538.142: [ParNew: 5183665K->571966K(5190464K), 0.7238850 secs] 16083661K->11972084K(22491968K), 0.7247810 secs] [Times: user=4.1
7 sys=0.12, real=0.73 secs]
2019-08-17T11:03:30.631+0800: 48541.121: [GC2019-08-17T11:03:30.631+0800: 48541.121: [ParNew: 5185726K->576704K(5190464K), 0.6396980 secs] 16585844K->12346323K(22491968K), 0.6405600 secs] [Times: user=3.6
8 sys=0.09, real=0.64 secs]
2019-08-17T11:03:33.075+0800: 48543.566: [GC2019-08-17T11:03:33.076+0800: 48543.566: [ParNew: 5190464K->576704K(5190464K), 1.5363980 secs] 16960083K->12790823K(22491968K), 1.5372590 secs] [Times: user=4.0
7 sys=0.75, real=1.53 secs]
2019-08-17T11:03:34.619+0800: 48545.109: [GC [1 CMS-initial-mark: 12214119K(17301504K)] 12795454K(22491968K), 0.6847160 secs] [Times: user=0.68 sys=0.00, real=0.68 secs]
2019-08-17T11:03:35.305+0800: 48545.796: [CMS-concurrent-mark-start]
2019-08-17T11:03:37.124+0800: 48547.614: [GC2019-08-17T11:03:37.124+0800: 48547.614: [ParNew: 5190464K->576704K(5190464K), 2.1277390 secs] 17404583K->13245164K(22491968K), 2.1285870 secs] [Times: user=5.0
0 sys=1.60, real=2.13 secs]
2019-08-17T11:03:40.637+0800: 48551.128: [GC2019-08-17T11:03:40.638+0800: 48551.128: [ParNew: 5190464K->576704K(5190464K), 2.2917500 secs] 17858924K->13825008K(22491968K), 2.2926540 secs] [Times: user=5.9
6 sys=1.64, real=2.29 secs]
2019-08-17T11:03:44.240+0800: 48554.730: [GC2019-08-17T11:03:44.240+0800: 48554.731: [ParNew: 5190464K->576704K(5190464K), 1.5847800 secs] 18438768K->14442978K(22491968K), 1.5857620 secs] [Times: user=5.8
3 sys=0.96, real=1.58 secs]
2019-08-17T11:03:47.129+0800: 48557.619: [GC2019-08-17T11:03:47.129+0800: 48557.620: [ParNew: 5190464K->576704K(5190464K), 2.6088650 secs] 19056738K->14989342K(22491968K), 2.6097680 secs] [Times: user=5.5
9 sys=2.00, real=2.61 secs]
2019-08-17T11:03:51.153+0800: 48561.643: [GC2019-08-17T11:03:51.153+0800: 48561.644: [ParNew: 5190464K->576704K(5190464K), 3.4004900 secs] 19603102K->15553978K(22491968K), 3.4015160 secs] [Times: user=7.1
7 sys=2.46, real=3.40 secs]
2019-08-17T11:03:55.936+0800: 48566.426: [GC2019-08-17T11:03:55.936+0800: 48566.426: [ParNew: 5190464K->576704K(5190464K), 1.9874080 secs] 20167738K->16393777K(22491968K), 1.9883410 secs] [Times: user=7.0
7 sys=1.27, real=1.99 secs]
2019-08-17T11:03:59.169+0800: 48569.659: [GC2019-08-17T11:03:59.169+0800: 48569.660: [ParNew (promotion failed): 5190464K->5190464K(5190464K), 151.9715540 secs]2019-08-17T11:06:31.141+0800: 4
8721.631: [CMS2019-08-17T11:06:41.022+0800: 48731.513: [CMS-concurrent-mark: 19.582/185.717 secs] [Times: user=322.79 sys=60.12, real=185.68 secs]
 (concurrent mode failure): 16268450K->5994133K(17301504K), 39.5874300 secs] 21007537K->5994133K(22491968K), [CMS Perm : 106999K->106688K(169984K)], 191.5598280 secs] [Times: user=242.58 sys=45.41, real=1
91.53 secs]
2019-08-17T11:07:12.062+0800: 48762.552: [GC2019-08-17T11:07:12.062+0800: 48762.552: [ParNew: 4613760K->576704K(5190464K), 1.3077710 secs] 10607893K->7468992K(22491968K), 1.3085280 secs] [Times: user=9.17
 sys=0.02, real=1.31 secs]
2019-08-17T11:07:14.615+0800: 48765.105: [GC2019-08-17T11:07:14.615+0800: 48765.105: [ParNew: 5190464K->576704K(5190464K), 0.8007980 secs] 12082752K->8128668K(22491968K), 0.8016420 secs] [Times: user=5.38
 sys=0.00, real=0.80 secs]
2019-08-17T11:07:16.460+0800: 48766.950: [GC2019-08-17T11:07:16.460+0800: 48766.950: [ParNew: 5190464K->576704K(5190464K), 1.4912750 secs] 12742428K->9224109K(22491968K), 1.4921230 secs] [Times: user=10.0
3 sys=0.00, real=1.49 secs]
2019-08-17T11:07:19.175+0800: 48769.665: [GC2019-08-17T11:07:19.175+0800: 48769.666: [ParNew: 5190464K->576704K(5190464K), 0.8212860 secs] 13837869K->9797270K(22491968K), 0.8221130 secs] [Times: user=5.63
 sys=0.00, real=0.82 secs]
2019-08-17T11:07:21.087+0800: 48771.578: [GC2019-08-17T11:07:21.088+0800: 48771.578: [ParNew: 5190464K->576704K(5190464K), 0.8776180 secs] 14411030K->10567781K(22491968K), 0.8785350 secs] [Times: user=5.7
3 sys=0.01, real=0.88 secs]
2019-08-17T11:07:23.122+0800: 48773.613: [GC2019-08-17T11:07:23.123+0800: 48773.613: [ParNew: 5190464K->576704K(5190464K), 0.8694820 secs] 15181541K->11179123K(22491968K), 0.8703220 secs] [Times: user=5.7
7 sys=0.01, real=0.88 secs]
2019-08-17T11:07:25.040+0800: 48775.531: [GC2019-08-17T11:07:25.041+0800: 48775.531: [ParNew: 5190464K->576704K(5190464K), 0.5756330 secs] 15792883K->11513169K(22491968K), 0.5764910 secs] [Times: user=3.8
1 sys=0.02, real=0.58 secs]
2019-08-17T11:07:26.778+0800: 48777.269: [GC2019-08-17T11:07:26.779+0800: 48777.269: [ParNew: 5190464K->576704K(5190464K), 0.5531170 secs] 16126929K->11900488K(22491968K), 0.5542050 secs] [Times: user=3.6
4 sys=0.01, real=0.56 secs]
2019-08-17T11:07:29.753+0800: 48780.243: [GC2019-08-17T11:07:29.753+0800: 48780.243: [ParNew: 5190464K->484832K(5190464K), 0.4668920 secs] 16514248K->12147612K(22491968K), 0.4677250 secs] [Times: user=3.0
1 sys=0.00, real=0.46 secs]
2019-08-17T11:07:38.895+0800: 48789.385: [GC2019-08-17T11:07:38.895+0800: 48789.385: [ParNew: 5098592K->576704K(5190464K), 0.2818390 secs] 16761372K->12241212K(22491968K), 0.2826330 secs] [Times: user=2.2
4 sys=0.00, real=0.28 secs]
2019-08-17T11:07:46.308+0800: 48796.799: [GC2019-08-17T11:07:46.309+0800: 48796.799: [ParNew: 5190464K->576704K(5190464K), 0.4371320 secs] 16854972K->12504773K(22491968K), 0.4380620 secs] [Times: user=2.9
7 sys=0.01, real=0.44 secs]
2019-08-17T11:07:50.904+0800: 48801.394: [GC2019-08-17T11:07:50.904+0800: 48801.395: [ParNew: 5190089K->436781K(5190464K), 0.3699670 secs] 17118159K->12646070K(22491968K), 0.3708110 secs] [Times: user=2.4
5 sys=0.01, real=0.38 secs]
2019-08-17T11:07:51.277+0800: 48801.767: [GC [1 CMS-initial-mark: 12209288K(17301504K)] 12646582K(22491968K), 0.2209210 secs] [Times: user=0.22 sys=0.00, real=0.22 secs]
2019-08-17T11:07:51.498+0800: 48801.989: [CMS-concurrent-mark-start]
2019-08-17T11:07:55.619+0800: 48806.109: [CMS-concurrent-mark: 4.119/4.121 secs] [Times: user=14.62 sys=0.38, real=4.12 secs]
2019-08-17T11:07:55.619+0800: 48806.110: [CMS-concurrent-preclean-start]
2019-08-17T11:07:55.739+0800: 48806.229: [CMS-concurrent-preclean: 0.108/0.120 secs] [Times: user=0.34 sys=0.02, real=0.12 secs]
2019-08-17T11:07:55.739+0800: 48806.230: [CMS-concurrent-abortable-preclean-start]
2019-08-17T11:07:57.305+0800: 48807.795: [GC2019-08-17T11:07:57.305+0800: 48807.795: [ParNew: 5050541K->230975K(5190464K), 0.1115260 secs] 17259830K->12440264K(22491968K), 0.1124190 secs] [Times: user=0.8
4 sys=0.01, real=0.12 secs]
 CMS: abort preclean due to time 2019-08-17T11:08:01.862+0800: 48812.353: [CMS-concurrent-abortable-preclean: 5.958/6.123 secs] [Times: user=14.89 sys=0.63, real=6.12 secs]
2019-08-17T11:08:01.864+0800: 48812.355: [GC[YG occupancy: 2379557 K (5190464 K)]2019-08-17T11:08:01.865+0800: 48812.355: [GC2019-08-17T11:08:01.865+0800: 48812.355: [ParNew: 2379557K->306806K(5190464K),
0.1407880 secs] 14588846K->12516095K(22491968K), 0.1417330 secs] [Times: user=1.03 sys=0.01, real=0.15 secs]
2019-08-17T11:08:02.007+0800: 48812.497: [Rescan (parallel) , 0.0565410 secs]2019-08-17T11:08:02.063+0800: 48812.554: [weak refs processing, 0.0011670 secs]2019-08-17T11:08:02.064+0800: 48812.555: [scrub
string table, 0.0062860 secs] [1 CMS-remark: 12209288K(17301504K)] 12516095K(22491968K), 0.2064110 secs] [Times: user=1.43 sys=0.01, real=0.21 secs]
2019-08-17T11:08:02.071+0800: 48812.562: [CMS-concurrent-sweep-start]
2019-08-17T11:08:07.810+0800: 48818.301: [GC2019-08-17T11:08:07.811+0800: 48818.301: [ParNew: 4920566K->429676K(5190464K), 0.1617750 secs] 13061507K->8570616K(22491968K), 0.1628060 secs] [Times: user=1.23
 sys=0.01, real=0.16 secs]
2019-08-17T11:08:12.512+0800: 48823.002: [GC2019-08-17T11:08:12.513+0800: 48823.003: [ParNew: 5043436K->454510K(5190464K), 0.2457460 secs] 9206556K->4675336K(22491968K), 0.2466240 secs] [Times: user=1.75
sys=0.01, real=0.25 secs]
2019-08-17T11:08:13.447+0800: 48823.937: [CMS-concurrent-sweep: 10.944/11.376 secs] [Times: user=44.77 sys=1.97, real=11.37 secs]
2019-08-17T11:08:13.447+0800: 48823.938: [CMS-concurrent-reset-start]
2019-08-17T11:08:13.513+0800: 48824.003: [CMS-concurrent-reset: 0.065/0.065 secs] [Times: user=0.31 sys=0.07, real=0.07 secs]
2019-08-17T11:08:17.937+0800: 48828.427: [GC2019-08-17T11:08:17.937+0800: 48828.428: [ParNew: 5068270K->472234K(5190464K), 0.2318510 secs] 8860461K->4341829K(22491968K), 0.2327560 secs] [Times: user=1.63
sys=0.00, real=0.24 secs]
2019-08-17T11:08:25.183+0800: 48835.673: [GC2019-08-17T11:08:25.184+0800: 48835.674: [ParNew: 5085994K->561958K(5190464K), 0.1959590 secs] 8955589K->4485967K(22491968K), 0.1969300 secs] [Times: user=1.46
sys=0.00, real=0.20 secs]
2019-08-17T11:08:32.218+0800: 48842.708: [GC2019-08-17T11:08:32.218+0800: 48842.709: [ParNew: 5175718K->488604K(5190464K), 0.2410410 secs] 9099727K->4502228K(22491968K), 0.2419920 secs] [Times: user=1.71
sys=0.00, real=0.24 secs]
2019-08-17T11:08:39.239+0800: 48849.729: [GC2019-08-17T11:08:39.240+0800: 48849.730: [ParNew: 5102364K->477861K(5190464K), 0.2332310 secs] 9115988K->4571114K(22491968K), 0.2343110 secs] [Times: user=1.63
sys=0.00, real=0.24 secs]
2019-08-17T11:08:46.610+0800: 48857.100: [GC2019-08-17T11:08:46.610+0800: 48857.101: [ParNew: 5091621K->429776K(5190464K), 0.2107400 secs] 9184874K->4611486K(22491968K), 0.2116540 secs] [Times: user=1.51
sys=0.01, real=0.21 secs]
2019-08-17T11:08:54.909+0800: 48865.399: [GC2019-08-17T11:08:54.909+0800: 48865.400: [ParNew: 5043536K->549821K(5190464K), 0.1887970 secs] 9225246K->4731531K(22491968K), 0.1900180 secs] [Times: user=1.43
sys=0.00, real=0.19 secs]
2019-08-17T11:09:03.947+0800: 48874.437: [GC2019-08-17T11:09:03.947+0800: 48874.438: [ParNew: 5163581K->486679K(5190464K), 0.2226900 secs] 9345291K->4724142K(22491968K), 0.2236710 secs] [Times: user=1.61
sys=0.01, real=0.22 secs]
2019-08-17T11:09:12.631+0800: 48883.122: [GC2019-08-17T11:09:12.632+0800: 48883.122: [ParNew: 5100439K->390273K(5190464K), 0.1972480 secs] 9337902K->4690593K(22491968K), 0.1983700 secs] [Times: user=1.41
sys=0.00, real=0.20 secs]
2019-08-17T11:09:20.105+0800: 48890.595: [GC2019-08-17T11:09:20.105+0800: 48890.596: [ParNew: 5004004K->344026K(5190464K), 0.2136360 secs] 9304325K->4735593K(22491968K), 0.2147120 secs] [Times: user=1.47
sys=0.00, real=0.22 secs]

查看您的 GC 日志,CMS 似乎不是您应用的正确选择。

考虑尝试其他 GC 算法:

  • 并行 GC(它会给你 10-30 秒的定期暂停,但不是 10 分钟)
  • 如果你可以升级到 Java 8,G1 可能是一个选项(它可能很好也可能很差 - 看你的运气)

如果您想坚持使用 CMS ...

... I can see info about "promotion failed" and "concurrent mode failure" in GC log, and the GC is very time consuming, almost takes 10 min to finish one GC cycle ...

你的晋升率很高(100+MiB/s)

2019-08-17T11:07:38.895+0800: 48789.385: [GC2019-08-17T11:07:38.895+0800: 48789.385: [ParNew: 5098592K->576704K(5190464K), 0.2818390 secs] 16761372K->12241212K(22491968K), 0.2826330 secs] [Times: user=2.2
4 sys=0.00, real=0.28 secs]
2019-08-17T11:07:46.308+0800: 48796.799: [GC2019-08-17T11:07:46.309+0800: 48796.799: [ParNew: 5190464K->576704K(5190464K), 0.4371320 secs] 16854972K->12504773K(22491968K), 0.4380620 secs] [Times: user=2.9
7 sys=0.01, real=0.44 secs]

CMS 需要更大的旧 space 大小来处理这样的吞吐量并保持低碎片化。不过,您的服务器似乎已达到内存限制。

我终于找到了根本原因。代码在高峰流量到来时会产生大量对象(超过10GB),导致young GC频繁发生,CMS GC fallback到SerialOld GC,造成长时间STW。