Skip to content

Latest commit

 

History

History
92 lines (75 loc) · 2.81 KB

session-java-virtual-machine-analysis.md

File metadata and controls

92 lines (75 loc) · 2.81 KB
Raw

Description

This session describes basic things to analyze the behavior of java virtual machines

The Garbage Collection Log

  • The following settings/arguments are useful to analyze the behavior of java virtual machines 
    java -XX:+PrintGCDetails -XX:+PrintTenuringDistribution -XX:+PrintGCApplicationConcurrentTime \
   -XX:+PrintGCApplicationStoppedTime -XX:+PrintGCDateStamps -XX:+PrintHeapAtGC \
   -XX:+UseGCLogFileRotation -XX:+PrintTenuringDistribution -XX:NumberOfGCLogFiles=5 \
   -XX:GCLogFileSize=5M -Xloggc:/var/log/foo/bar/java_gc.log -verbose:gc ....
  • How to find the most recent GC Log 
    cd /var/log/foo/bar/
ls -latr
  • Add a timestamp to the logfile name 
    -Xloggc:/var/log/foo/bar/java_gc-$(date --date="today" "+%Y-%m-%d_%H-%M-%S").log

GC Logs can by analyzed by:

Using heapdumps find memory leaks or to size the heap for performance

Its a good idea to have enough space (minimal = -Xmx + 50%) for creating heapdumps.

  • Automatic heapdumps Create a heapdump when a out of memory situation appears and stop server immediately. 
    java -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/var/log/tomcat/ -XX:OnOutOfMemoryError="kill -9 %p" ....
  • Create a manual heapdump 
    su - <java process user>
jmap -heap:file=heap.dump.out,format=b [PID]
time jmap -F -dump:file=$(hostname -f)_$(date --date="today" "+%Y-%m-%d_%H-%M-%S").hprof,format=b `pgrep java`

Heapdumps can be analyzed with: (change heap of this tools to load huge heapdumps)

Interacting with a running jvm

NOTE: If you call tools from the jdk/jre invoke them with the same user and with the same jdk/jre like the running jvm

  • Create threadump to stdout 
    kill -QUIT $(pgrep java)
  • Which java are we running currently and with which user? 
    ps auxwwww|grep java
readlink -f <java process>
  • Print current state of the Heap 
    jmap -heap <pid>
  • Get a histogramm of all loaded classes 
    jmap -histo 29655
  • Invoke a manual GC, useful to get a idea how much memory is really needed 
    jcmd 12572 GC.run
  • Get a Threadump 
    jstack <pid>

Analyzing deployments

  • Get a list of all classes 
    find . -name "*.jar" |while read A; do unzip -l $A; done|grep class|awk '{print $4}'|grep -e "\.class$"|sort -u|sed '~s,\.class$,,;~s,/,.,g'
  • Analyze a deployment if there are duplicate classes 
    find . -name "*.jar" |while read A; do unzip -l $A; done|grep class|awk '{print $4}'|sort|uniq -c |sort -n|awk '{if ($1 > 1){print $0}}' |tee duplicate-classes