JVM tuning - nimrody/knowledgebase GitHub Wiki

• Displaying traces and overview of tracing methods: https://thume.ca/2023/12/02/tracing-methods/ (tracing / profiling)

• JVM Honest Profiler

• Production options:

  java -Xms3500M -Xmx3500M -XX:+UseConcMarkSweepGC -XX:+CMSParallelRemarkEnabled -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/home/ec2-user/policy.hprof -cp 'policy/*' MainWebServer config/qa.new.config dev.nimrod.tensera config logs

• JVMtop (https://github.com/patric-r/jvmtop/blob/master/doc/ConsoleProfiler.md)

• Using strace to get system call statistics

  sudo strace -c -f -p

• And to monitor specific system call such as fsync:

  sudo strace -e fsync -T -f -p

• JVM perf

• JVM perf flamegraph

• JVM Production flags

• Tuning garbage collection

• Connecting to remote JVM

• Histogram of heap

  jmap -histo:live <pid>
  

  Or remove the :live suffix to get all objects and avoid running a full GC

• Heap dump:

  jmap -dump:live,format=b,file=/tmp/dump.hprof <pid>
  

  (again, remove :live suffix to avoid full GC and get all objects)

  OR

  jcmd 1  GC.heap_dump filename=dump.hprof
  

  (add -all to get all objects and avoid a full GC)

• Exploring heap dump with jhat

  jmap -dump:file=eclipse.heap 16608 Dumping heap to C:\Program Files (x86)\Java\jdk1.6.0_11\bin\eclipse.heap ... Heap dump file created

  jhat -J-Xmx1G eclipse.heap Reading from eclipse.heap... Dump file created Wed Oct 14 08:41:07 MSD 2009 Snapshot read, resolving... Resolving 2300585 objects... Chasing references, expect 460 dots............................................. ................................................................................ ................................................................................ ....................................... Eliminating duplicate references................................................ ................................................................................ ................................................................................ .................................... Snapshot resolved. Started HTTP server on port 7000 Server is ready.

  And open a browser on localhost:7000

• Monitoring garbage collection using jstat

  jstat –gc <vmid> 1000
  

• Asking the JVM to print GC stats:

  -verbose:gc
  

• Dump heap on OOM error

  -XX:+HeapDumpOnOutOfMemoryError
  -XX:HeapDumpPath=/path/
  

• Print all JVM flags and their values java -XX:+PrintFlagsFinal. Note the the following options

  • UseParallelGC
  • UseParallelOldGc
  • UseParallelNewGc
  • UseConcurrentGC
  • UseG1GC

(See here and here for more information on the above options)

• Use jcmd <pid-or-main-class-name> help to get the list of available commands

  jcmd ReportsWriterMain Thread.print      # print all active threads
  
  jcmd ReportsWriterMain GC.heap_dump xx   # dump heap to file xx in the directory where the jvm was started
  

• Use jhat <filename> to analyze heap dumps produced by jcmd GC.heap_dump (or jvisualvm for a gui)

• Printing generated assembly (needs the hsdis-amd64 DLL available using LD_LIBRARY_PATH)

  java -XX:+UnlockDiagnosticVMOptions -XX:PrintAssemblyOptions=hsdis-print-bytes -XX:CompileCommand=print,MyClass.mymethod TestCode

• Getting the number of context switches for a running process:

  sudo perf stat -e cs -p

http://www.javamem.com/post.cfm/15-useful-performance-tools-for-java-environments

• profiler4j [https://blog.idrsolutions.com/2014/06/java-performance-tuning-tools/profiler4j/] [https://www.openhub.net/p/profiler4j], [http://freecode.com/projects/profiler4j]

• jprof

• Running JVM under docker

• To use jmap or jcmd run it with the user that started the JVM. I.e., sudo -u <jvm-user> jmap ...