Version

    Chapter 31. Troubleshooting Worker

    Worker Logs

    When investigating issues with Worker itself or jobs running in Worker, there are several logs with useful information:

    Logs found in the Monitoring → Server Logs section of the Server Console.

    • COMMON log - the main Server log contains also information related to Worker. This log contains the full command line used to start Worker, this allows you to check the command line arguments. Additionally, standard output of the Worker process is redirected to this log - this is useful especially if the Worker process crashes during startup.

      The COMMON log file is located in ${java.io.tmpdir}/cloverlogs/all.log

    • WORKER log - the main Worker log provides information about Worker startup, initialization, executed jobs, runtime activities, etc. The initialization details contain information about Worker's JNDI resources, etc.

      The WORKER log file is located in ${java.io.tmpdir}/cloverlogs/worker_[nodeID].log

      You can also open this log via the Go to logs action in the Worker section of the Monitoring page.

    Worker Command Line

    The full command line that was used to start the Worker process can be found in:

    • Monitoring section, use the Show command line action on Worker. For more details, see Showing Worker's Command Line Arguments.

    • The COMMON log of the Server (found in the Monitoring → Server Logs page). See section above for more details.

    Investigate the command line options in case Worker does not correctly start or if the configuration of the running Worker is not correct.

    Performance Logging

    CloverDX regularly collects performance metrics and stores them in the performance log. The metrics are such as CPU load, garbage collector activity, used heap, thread counts etc. The performance log is an additional tool to analyze an incident, see Performance Log for more details.

    Worker Does Not Start

    If Worker does not start, check the following:

    • Server's COMMON log and the WORKER log, see above. Look for errors during Worker startup and initialization.

    • Worker's command line arguments, see above. Look for invalid command line arguments. Additionally, check the custom JVM arguments set on Worker, in the Worker tab of Setup or via the worker.jvmOptions configuration property.

    Restarting Worker

    If Worker gets into an unrecoverable state (e.g. out of heap memory, etc.) and you fix the source issue, you can restart it from the Monitoring section (For more details, see Restarting the Worker.):

    • restart immediately, which will abort jobs currently running in Worker;

    • restart after running jobs finish, in case the currently running jobs are crucial.

    Worker does not start

    In case Worker does not start (i.e. remains in the STARTING status), check the COMMON log first.

    Worker Crashes

    If common log (all.log file) contains row similar to the following one, the worker crashed due to exhausted heap space. 

    2018-03-22 16:08:29,008[s StdOut reader] WorkerProcess    INFO      [worker0@N1:10500]: java.lang.OutOfMemoryError: Java heap space

    You can configure it to generate heap dump for further investigation. To do so, add -XX:+HeapDumpOnOutOfMemoryError to JVM arguments in Worker configuration (Configuration > Setup > Worker). The generated file can be investigated with tools like jvisualvm or jhat.

    Another cause of the crash can be swapped Worker's main memory. If there is insufficient free space in the main memory and the pages of Worker process are swapped on the hard drive, Worker is slowed down, it does not receive heart beat from the Server in time and it kills itself. Lower the maximum heap size of Worker to avoid swapping. Note that Java process uses also non-heap memory, e.g. metaspace or direct memory.

    To investigate usage of direct memory, add -XX:NativeMemoryTracking=summary to Worker's JVM arguments (Configuration > Setup > Worker). The details on native memory usage can be displayed with jcmd <pid> VM.native_memory summary. See https://docs.oracle.com/javase/8/docs/technotes/guides/troubleshoot/tooldescr007.html for details on native memory tracking.

    Worker Hangs

    Usually, it is caused by garbage collector. Try tweaking the garbage collection.

    Another cause can be swapping of worker process pages on hard drive.

    Issues with Classloading

    To debug issues with classloading, add -verbose:class to JVM arguments of Worker. Loaded and unloaded classes will be printed to the output. The output can be seen in the common log.