序
本文主要介绍的是一款jvm排查工具箱：jvm-tools。除了对基本jvm封装外，还提供了jmx访问以及火焰图的生成。
下载
- sjk.jar - all commands without mxdump
- sjk-plus.jar - all commands
实例
cpu and memory usage
java -jar sjk-0.9.jar ttop -o CPU -n 10 -p pid
java -jar sjk-0.9.jar ttop -o USER -n 10 -p pid
java -jar sjk-0.9.jar ttop -o SYS -n 10 -p pid
java -jar sjk-0.9.jar ttop -o ALLOC -n 10 -p pid
一个命令搞定排查造成cpu高的线程
root@d3daf1eca8ca:/# java -jar sjk-0.9.jar ttop -o CPU -n 10 -p 1
Monitoring threads ...
2018-01-01T07:18:26.773+0000 Process summary
process cpu=6.88%
application cpu=1.37% (user=1.30% sys=0.07%)
other: cpu=5.51%
thread count: 25
GC time=0.00% (young=0.00%, old=0.00%)
heap allocation rate 390kb/s
safe point rate: 1.5 (events/s) avg. safe point pause: 0.16ms
safe point sync time: 0.01% processing time: 0.02% (wallclock time)
[000068] user= 1.30% sys= 0.02% alloc= 386kb/s - RMI TCP Connection(1)-172.17.0.2
[000070] user= 0.00% sys= 0.03% alloc= 4268b/s - JMX server connection timeout 70
[000047] user= 0.00% sys= 0.01% alloc= 0b/s - NioBlockingSelector.BlockPoller-1
[000060] user= 0.00% sys= 0.01% alloc= 52b/s - http-nio-8080-AsyncTimeout
[000058] user= 0.00% sys= 0.00% alloc= 23b/s - http-nio-8080-ClientPoller-0
[000014] user= 0.00% sys= 0.00% alloc= 0b/s - container-0
[000002] user= 0.00% sys= 0.00% alloc= 0b/s - Reference Handler
[000003] user= 0.00% sys= 0.00% alloc= 0b/s - Finalizer
[000004] user= 0.00% sys= 0.00% alloc= 0b/s - Signal Dispatcher
[000013] user= 0.00% sys= 0.00% alloc= 0b/s - ContainerBackgroundProcessor[StandardEngine[Tomcat]]
histo
java -jar sjk-0.9.jar hh -n 10 --dead -p pid
java -jar sjk-0.9.jar hh -n 10 --dead-young -p pid
java -jar sjk-0.9.jar hh -n 10 --live -p pid
查看对象统计
root@d3daf1eca8ca:/# java -jar sjk-0.9.jar hh -n 10 --live -p 1
# Instances Bytes Type
1: 67025 9944432 [C
2: 9382 2209656 [I
3: 18413 1620344 java.lang.reflect.Method
4: 65810 1579440 java.lang.String
5: 4140 1283520 [B
6: 35735 1143520 java.util.concurrent.ConcurrentHashMap$Node
7: 8514 946936 java.lang.Class
8: 15267 610680 java.util.LinkedHashMap$Entry
9: 9027 505512 java.util.LinkedHashMap
10: 10414 494656 [Ljava.lang.Object;
Total 475830 28175752
gc
java -jar sjk-0.9.jar gc -p pid
查看gc日志
root@d3daf1eca8ca:/# java -jar sjk-0.9.jar gc -p 1
MBean server connected
Collecting GC stats ...
[GC: Copy#112 time: 4ms mem: Survivor Space: 0k+135k->135k[max:17088k] Compressed Class Space: 5190k+0k->5190k[max:1048576k] Eden Space: 18496k-18496k->0k[max:137152k] Metaspace: 41247k+0k->41247k Tenured Gen: 27515k+0k->27515k[max:342720k]]
[GC: Copy#113 time: 2ms interval: 6649ms mem: Survivor Space: 135k+304k->439k[max:17088k,rate:45.83kb/s] Compressed Class Space: 5243k+0k->5243k[max:1048576k,rate:0.00kb/s] Eden Space: 18496k-18496k->0k[max:137152k,rate:-2781.77kb/s] Metaspace: 41696k+0k->41696k[rate:0.00kb/s] Tenured Gen: 27515k+0k->27515k[max:342720k,rate:0.00kb/s]]
[GC: Copy#114 time: 3ms interval: 675ms mem: Survivor Space: 439k+779k->1219k[max:17088k,rate:1154.48kb/s] Compressed Class Space: 5338k+0k->5338k[max:1048576k,rate:0.00kb/s] Eden Space: 18496k-18496k->0k[max:137152k,rate:-27401.48kb/s] Metaspace: 42470k+0k->42470k[rate:0.00kb/s] Tenured Gen: 27515k+0k->27515k[max:342720k,rate:0.00kb/s]]
stack
## collect 30 seconds
java -jar sjk-0.9.jar stcap -o dump.std -t 30000 -p pid
java -jar sjk-0.9.jar stcap -o dump.std -p pid
dump线程堆栈
root@d3daf1eca8ca:/# java -jar sjk-0.9.jar stcap -o dump.std -p 1
Writing to /dump.std
Collected 506
Collected 1012
Collected 1518
Collected 2001
Collected 2507
//......
Trace dumped: 123211
查看堆栈或生成火焰图
java -jar sjk-0.9.jar ssa -f dump.std --print
java -jar sjk-0.9.jar ssa -f dump.std --histo
java -jar sjk-0.9.jar ssa -f dump.std --flame > flame.svg
屏幕快照 2018-01-01 下午3.26.18.png
jmx
java -jar sjk-0.9.jar --help mx
java -jar sjk-0.9.jar mx -mi -b java.lang:type=Threading -p pid
java -jar sjk-0.9.jar mx -mg -b java.lang:type=Memory -f HeapMemoryUsage -p pid
查看jmx
root@d3daf1eca8ca:/# java -jar sjk-0.9.jar mx -mg -b java.lang:type=Memory -f HeapMemoryUsage -p 1
java.lang:type=Memory
committed: 68194304
init: 33554432
max: 508887040
used: 49297880
小结
使用ttop命名查看cpu高的线程，以及使用stack stcap和ssa生成火焰图相当便利，值得推荐。
