为何在使用CMS gc算法时会出现连续两次full gc
现象:
jstat -gcutil pid 1000观察到的情况,段时间内连续两次full gc
S0 S1 E O P YGC YGCT FGC FGCT GCT 59.33 0.00 63.98 69.66 59.31 24338 274.969 307 17.349 292.318 59.33 0.00 86.19 69.66 59.31 24338 274.969 307 17.349 292.318 0.00 60.52 10.22 70.10 59.31 24339 275.006 308 17.373 292.379 0.00 60.52 30.03 70.10 59.31 24339 275.006 308 17.373 292.379 0.00 60.52 53.57 70.10 59.31 24339 275.006 308 17.373 292.379 0.00 60.52 76.45 70.10 59.31 24339 275.006 308 17.373 292.379 0.00 60.52 93.77 70.10 59.31 24339 275.006 308 17.373 292.379 61.16 0.00 15.66 70.53 59.31 24340 275.040 308 17.373 292.413 61.16 0.00 40.71 67.96 59.31 24340 275.040 309 17.399 292.439 61.16 0.00 66.44 59.90 59.31 24340 275.040 309 17.399 292.439
?同一时间的gc log
第一次
[GC [1 CMS-initial-mark: 1220647K(1740800K)] 1227199K(2385920K), 0.0057570 secs] [Times: user=0.00 sys=0.00, real=0.00 secs]
在旧生代空间为1220647K触发marking操作,后面1227199K(2385920K)=当前总体jvm内存使用(maxMem)
initial? 标记从根集合中可直接访问的对象,要停顿整个应用
原文:This is initial Marking phase of CMS where all the objects directly reachable from roots are marked and this is done with all the mutator threads stopped.
[CMS-concurrent-mark: 1.084/1.084 secs] [Times: user=2.43 sys=0.11, real=1.08 secs]
完成mark耗时1.084 secs 并发执行
[CMS-concurrent-preclean: 0.008/0.009 secs] [Times: user=0.01 sys=0.00, real=0.01 secs]
开始执行清理,目的是减少remark的时间
?
第二次
[GC [ParNew: 579913K->4387K(645120K), 0.0080390 secs] 1800561K->1226923K(2385920K), 0.0082860 secs] [Times: user=0.05 sys=0.00, real=0.01 secs]
minor gc
? CMS: abort preclean due to time [CMS-concurrent-abortable-preclean: 2.715/5.047 secs] [Times: user=4.42 sys=0.40, real=5.05 secs]
停止执行preclean 默认是eden到达50%或者real time=5secs停止
[GC[YG occupancy: 153139 K (645120 K)][Rescan (parallel) , 0.0348570 secs][weak refs processing, 0.0005070 secs] [1 CMS-remark: 1222535K(1740800K)] 1375675K(2385920K), 0.0355150 secs] [Times: user=0.19 sys=0.00, real=0.04 secs]
remark会停顿整个应用
[CMS-concurrent-sweep: 1.358/1.360 secs] [Times: user=1.85 sys=0.10, real=1.36 secs]
[CMS-concurrent-reset: 0.003/0.003 secs] [Times: user=0.00 sys=0.00, real=0.00 secs]
gcutil 看到连续两次 full gc
原因分析:
目前来看,只有当initial-mark和remark时才会停顿整个应用,这个两个时间点jstat -gcutil表现都为full gc次数加一
?
real != user+sys
原因:user和sys代表处于用户和系统态的时间,不包含block的时间.但是由于多cpus这个东东是累加的.所以一般来说user和sys相加都大于real
real是从启动到终止的真实时间(现实中消耗),包含block.
