Search code examples
javalinuxmultithreadingwebsphere

Multithreading java application is failing sometimes to get all threads on Websphere

I have an exhaustive java application (JRE 1.8.0_201) that requires to be executed in threads, It is deployed on IBM Websphere 9.

At WAS startup time the application executions seems to be fine, however sometimes later the multithreading capability is lost and the executions occurs using only one thread.

If we restart again the websphere the situation turns into normality, the executions takes again the proper parallel threads.

Is there some WAS configurations that I'm missing? or at java or OS side?

Below the HW architecture on Linux:

  • Architecture: x86_64
  • CPU op-mode (s): 32-bit, 64-bit
  • Byte Order: Little Endian
  • CPU (s): 4
  • On-line CPU (s) list: 0-3
  • Thread (s) per core: 1
  • Core (s) per socket: 1
  • Socket (s): 4
Solution

  • To get maximum information about the problem:

    Thread Dump

    Creates a listing of the current state of all threads within JVM. It’s advisable to do this listing repeatedly (in differently named outputs, don’t overwrite your dumps) within a few tens of seconds/minutes and in your case maybe hours. This allows you to capture changes and possible dead-lock. If nothing has changed and thread is on the same row, then there might be a problem.

    How to make thread dump:

    Open the console (CMD), switch to “/Java/jdkx.x.x_xx /bin” (JDK) using the “cd” command. Then type “jstack JVM_PID > /thread_dump_1“. Replace the “JVM_PID” with the process ID of running JVM (Task Manager – Details – PID column).

    You can also use tool like VisualVM to take a thread dump:

    enter image description here

    How read thread dumps:

    Great article: https://dzone.com/articles/how-to-read-a-thread-dump

    Look for blocked and waiting threads. Probably there will be multiple threads stucked on same line and only one thread making some progress.

    How read thread dumps in your situation:

    I would make a few thread dumps right after the restart and then when problem occurs. If you can distinguish when one thread ends and another one get time for progress I would make another thread dump and compare them (before first end and after second start working).