GC暂停时间过长——排查分析

告警

本次GC日志分析

2022-11-17T17:58:50.518+0800: 1217960.132: [GC (Allocation Failure) 2022-11-17T17:58:50.518+0800: 1217960.132: [ParNew: 1382400K->153600K(1382400K), 0.
5626158 secs] 3419277K->2410488K(4040704K), 0.5628652 secs] [Times: user=1.07 sys=0.00, real=0.56 secs]
2022-11-17T17:58:51.084+0800: 1217960.698: [GC (CMS Initial Mark) [1 CMS-initial-mark: 2256888K(2658304K)] 2419662K(4040704K), 0.0349632 secs] [Times:
user=0.05 sys=0.00, real=0.03 secs]
2022-11-17T17:58:51.119+0800: 1217960.734: [CMS-concurrent-mark-start]
2022-11-17T17:58:51.626+0800: 1217961.240: [CMS-concurrent-mark: 0.506/0.506 secs] [Times: user=1.03 sys=0.00, real=0.51 secs]
2022-11-17T17:58:51.626+0800: 1217961.240: [CMS-concurrent-preclean-start]
2022-11-17T17:58:51.634+0800: 1217961.248: [CMS-concurrent-preclean: 0.008/0.008 secs] [Times: user=0.02 sys=0.00, real=0.01 secs]
2022-11-17T17:58:51.634+0800: 1217961.248: [CMS-concurrent-abortable-preclean-start]
 CMS: abort preclean due to time 2022-11-17T17:58:56.994+0800: 1217966.608: [CMS-concurrent-abortable-preclean: 4.854/5.360 secs] [Times: user=5.29 sys
=0.03, real=5.36 secs]
2022-11-17T17:58:56.997+0800: 1217966.611: [GC (CMS Final Remark) [YG occupancy: 629022 K (1382400 K)]2022-11-17T17:58:56.997+0800: 1217966.611: [Resca
n (parallel) , 0.0937759 secs]2022-11-17T17:58:57.091+0800: 1217966.705: [weak refs processing, 0.0002312 secs]2022-11-17T17:58:57.091+0800: 1217966.70
5: [class unloading, 0.0450819 secs]2022-11-17T17:58:57.136+0800: 1217966.750: [scrub symbol table, 0.0198218 secs]2022-11-17T17:58:57.156+0800: 121796
6.770: [scrub string table, 0.0021679 secs][1 CMS-remark: 2256888K(2658304K)] 2885910K(4040704K), 0.1613101 secs] [Times: user=0.26 sys=0.00, real=0.16
 secs]
2022-11-17T17:58:57.159+0800: 1217966.773: [CMS-concurrent-sweep-start]
2022-11-17T17:58:58.567+0800: 1217968.181: [CMS-concurrent-sweep: 1.398/1.409 secs] [Times: user=1.44 sys=0.00, real=1.41 secs]
2022-11-17T17:58:58.567+0800: 1217968.182: [CMS-concurrent-reset-start]
2022-11-17T17:58:58.576+0800: 1217968.190: [CMS-concurrent-reset: 0.008/0.008 secs] [Times: user=0.01 sys=0.00, real=0.01 secs]

STW的两个阶段，共耗时 0.19s

• 初始标记 CMS Initial Mark: 耗时0.03s
• 重新标记 CMS Final Remark: 耗时0.16s

并发阶段，耗时

• 并发标记 CMS-concurrent-mark: 耗时0.51s
• 并发预清理 CMS-concurrent-preclean：耗时0.01s
• 并发可中断预清理 CMS-concurrent-abortable-preclean: 耗时5.36s
• 并发清理 CMS-concurrent-sweep：1.41s

告警和监控的问题

经过对gc日志的分析，发现，实际暂停时间和告警内容不符，查看监控，也发现了这个问题

和运维确认后，监控和告警的指标，并不是gc暂停时间，而是gc总时长。
因此，运维侧，需要优化，告警、监控相关说明
监控最好能够添加暂停时长的指标 @运维同鞋

有内存不足的表现

查看堆监控发现，

• eden去和s区经常被打满，对象进入老年代
• 使得fullgc较为频繁
• 没有内存泄露的表现（没有持续增长，无法回收），不需要再dump分析

老年代够用，但新生代小了

结论与方案

• gc暂停时间并不长，只是gc整个过程较长
• 告警中的时间并不是gc暂停时间，运维侧需要调整告警内容与监控的指标，细化暂停时长的监控@运维同鞋
• 堆内存，老年代够用，但新生代小了，建议老年代保持不变，新生代加内存，
  • 当前是堆内存共4GB，新生代为1500m
  • 新生代加到2500m，然后持续观察