Version

    Chapter 19. Logging

    Main Logs
    Another Useful Logging Settings
    Access Log in Apache Tomcat
    Application Server Logs
    Graph Run Logs
    Server Audit Logs
    Designer-Server Integration Logs
    Performance Log
    Job Queue Log

    Main Logs

    Since 5.2, CloverDX Server uses the Log4j 2 library for logging. The WAR file contains the default Log4j 2 configuration.

    Location of the Log4j 2 configuration file is:

    • Server Core: [clover.war]/WEB-INF/log4j2.xml

    • Worker: [clover.war]/WEB-INF/worker/cloveretl.server.worker.jar/log4j2.xml

    By default, log files are produced in the directory specified by the java.io.tmpdir system property in the cloverlogs subdirectory.

    The java.io.tmpdir system property usually points to a common system temp directory, i.e. /tmp. On Apache Tomcat, it is usually the $TOMCAT_HOME/temp directory.

    You can customize the logging configuration for both Core and Worker by supplying own log4j2.xml:

    • Server Core:

      1. Define a new system property log4j.configurationFile with the full path to the file:

        -Dlog4j.configurationFile=file:///C:/path/to/log4j2-server.xml

        or

        -Dlog4j.configurationFile=file:///opt/path/to/log4j2-server.xml

      2. Restart CloverDX Server.

    • Worker:

      1. Using worker.jvmOptions, define a new system property log4j.configurationFile with the full path to the file:

        worker.jvmOptions= -Dlog4j.configurationFile=file:///C:/path/to/log4j2-worker.xml

        or

        worker.jvmOptions= -Dlog4j.configurationFile=file:///opt/path/to/log4j2-worker.xml

      2. Restart Worker.

    [Important]Important

    You must provide complete configuration as the supplied file will completely override log4j configuration shipped with CloverDX Server. It is therefore best to copy out and modify the configuration shipped with Server (see above for location).

    If you used a custom Log4j configuration in CloverDX Server 5.1 and older, you will have to migrate to Log4j 2 configuration in CloverDX Server 5.2.

    For configuration details, see Configuration of Log4j 2.

    Note: since such a configuration overrides the default configuration, it may influence Graph run logs. So your own log configuration has to contain following fragment to preserve Graph run logs:

    <logger name="Tracking" additivity="false">
      <level value="debug"/>
    </logger>

    Another Useful Logging Settings

    These system properties allow for logging of HTTP requests/responses to stdout:

    Client side:

    com.sun.xml.ws.transport.http.client.HttpTransportPipe.dump=true (For more information, see the Working with CloverDX Server.)

    Server side:

    com.sun.xml.ws.transport.http.HttpAdapter.dump=true

    Access Log in Apache Tomcat

    If you need to log all requests processed by the server, add the following code to $CATALINA_HOME/conf/server.xml.

    <Valve className="org.apache.catalina.valves.AccessLogValve" directory="logs"
        prefix="localhost_access_log" suffix=".txt"
        pattern="%h %l %u %t %D %r %s %b" />

    The format defined above has following meaning

    [IP address] [date-time] [processing duration in milliseconds] [method] [URL] [protocol+version] [response code] [response size]

    The log will look like the next line

    172.17.30.243 - - [13/Nov/2014:12:53:03 +0000] 2 "POST /clover/spring-rpc/clusterNodeApi HTTP/1.1" 200 1435"

    See also Valve in documentation on Apache Tomcat.

    Application Server Logs

    If you use Apache Tomcat, it logs into $CATALINA_HOME/logs/catalina.out file.

    Graph Run Logs

    Each graph or jobflow run has its own log file – for example, in the Server Console, section Execution History.

    By default, these log files are saved in the subdirectory cloverLogs/graph in the directory specified by java.io.tmpdir system property.

    It’s possible to specify a different location for these logs with the CloverDX graph.logs_path property. This property does not influence main Server logs.

    Server Audit Logs

    Server Audit Log logs operations called on ServerFacade and JDBC proxy interfaces.

    Audit logging can be enabled by setting (adding) the value of CloverDX property logging.logger.server_audit.enabled to true. In server GUI, you can change the property value in ConfigurationSetupConfiguration File. Audit logging is disabled by default.

    The name of output file is server-audit.log. The file is in the same directory as main server log files. Default log level is DEBUG, so all operations which may do any change or another important operations (e.g. login or openJdbcConnection) are logged. To enable logging of all operations, change log level to TRACE in the log4j configuration.

    Each logged operation is logged by two messages: entering method and exiting method (if the exception is raised, it's logged instead of output parameters)

    • Entering method (marked as "inputParams"). All method's parameters (except for passwords) are printed.
    • Exiting method (marked as "outputParams"). Method's return value is printed.
    • Exception in method (marked as "EXCEPTION"). Exception's stacktrace is printed.

    Message also contains:

    • username, if the user is known
    • client IP address, if it's known
    • Cluster node ID
    • Interface name and the operation name

    Values of transient and lazy initialized (in entity classes) fields and fields with binary content are not printed.

    Designer-Server Integration Logs

    The logging of Designer-Server integration can be enabled with logging.logger.server_integration.enabled configuration property. The name of the log file is server-integration.log.

    The log format is date and time, IP address of Designer, user name, operation, result of the operation (success/failure) and duration in milliseconds.

    2018-03-07 16:42:00,525 10.0.3.2 user=clover, operation=executeGraph SUCCESS duration=576 ms

    Performance Log

    The performance log measures performance metrics that are useful for diagnostics of incidents such as slowdown of processing, restarts of Worker or Server Core etc. The metrics are such as CPU load, Java heap size, number of open files etc. CloverDX logs values of the metrics in regular intervals into the performance log.

    Example of the performance log:

    Performance log example

    Figure 19.1. Performance log example


    In the above example, the jobs in Worker require large heap space, which will soon cause the Worker's heap to fill and garbage collector to consume all CPU time of the Worker. The Worker will stop responding to Server and will be restarted.

    Each line of the performance log is one entry, which is a JSON. The line contains a timestamp and values of metrics from Server Core and Worker. There can be short gaps in the log, e.g. under very high load the Worker might not be able to send the metric values for every entry. Long gaps indicate a more severe issue.

    The performance log can be found in:

    • Server Console Monitoring > Server Logs > PERFORMANCE page
    • {java.io.tmpdir}/cloverlogs/performance.log

    The performance log can be disabled by setting configuration property logging.logger.performance.enabled=false

    Table 19.1. Performance Log Metrics

    Metric Label in LogMetricDescriptionUnitCollected in
    time Timestamp

    Date and time when metrics in this entry were collected.

    The timestamps should be even, i.e. there should not be significant time gaps between the log entries. By default, Server logs the metrics every 3 seconds. If the log entry contains only metrics from Worker (i.e. Server Core wasn't able to collect the metrics), then the timestamp will be for the time when Worker collected the metrics.

    timestamp core/worker
    hbAge Heartbeat age

    Age of latest Worker heartbeat (Worker sends regular heartbeat to Server Core to report its health). Measured as a difference between the current time and the time when the latest Worker heartbeat was received by Server Core.

    Heartbeat age should be typically a few seconds. Tens of seconds indicate a problem, typically overloaded Worker or Server Core. For example when running out of free heap the Worker will spend its time in GC and will have less capacity to send heartbeat. When the heartbeat age gets over a limit value (about a minute), the Server Core will restart the Worker.

    seconds core
    cGC Server Core garbage collector activity

    Percentage of CPU time spent in garbage collection for the Server Core, since the time of previous performance log entry.

    Typical values should be <20%, with occasional higher spikes. Long intervals of high GC activity indicate that the Server Core is running out of free heap.

    0-100% core
    wGC Worker garbage collector activity

    Percentage of CPU time spent in garbage collection for the Worker, since the time of previous performance log entry.

    Typical values should be <20%, with occasional higher spikes. Long intervals of high GC activity indicate that the Worker is running out of free heap. High Worker GC activity will often be accompanied by short gaps in metrics reported by the Worker.

    0-100% worker
    cHeap Server Core heap usage

    Amount of used java heap memory in Server Core JVM.

    The value of used heap can rise up to maximum heap size of Server Core. The maximum heap size of Server Core is set as the heap size for the Application Container running CloverDX Server, e.g. via the -Xmx argument. Look in the all.log to find the maximum heap size - Server Core logs system information when starting. As the used heap size nears the maximum heap size, Server Core GC activity will increase.

    GB core
    wHeap Worker heap usage

    Amount of used java heap memory in Worker JVM.

    The value of used heap can rise up to maximum heap size of Worker. Maximum size of Worker heap is set in the Worker configuration (Configuration > Setup > Worker, or worker.maxHeapSize configuration property). Additionally, Worker logs the maximum Worker heap size in the worker.log as part of system information log. As the used heap size nears the maximum heap size, Worker GC activity will increase.

    GB worker
    sysCPU System CPU load

    Total CPU load of the machine on which Server runs. The system CPU load is caused by all processes running on the machine. The measured CPU load is average over last 10 seconds.

    If the system CPU load is significantly higher than the sum of Worker and Server Core CPU load for a long time, this indicates another CPU demanding process is running on the machine. Alternatively, high swap usage can cause high system CPU load.

    0-100% core
    cCPU Server Core process CPU load

    CPU load that Server Core process introduces on the system. The Server Core CPU load is average over last 10 seconds.

    Typically, the Server Core CPU load should be low (<10%), as most processing is done by jobs running in the Worker.

    0-100% core
    wCPU Worker process CPU load

    CPU load that Worker process introduces on the system. The Worker CPU load is average over last 10 seconds.

    0-100% worker
    jobs Running jobs

    Number of currently running CloverDX jobs. If this Server is part of CloverDX Cluster, only jobs running on this node are counted.

    number core
    jobQueue Size of job queue

    Number of currently enqueued CloverDX jobs, see Job Queue for more information about the job queue. If this Server is part of CloverDX Cluster, only jobs enqueued on this node are counted.

    number core
    cThread Server Core threads

    Number of currently live threads in Server Core JVM.

    Typical values are up to hundreds. If the values continuously increase that can indicate a thread leak.

    number core
    wThread Worker threads

    Number of currently live threads in Worker JVM.

    Typical values are up to hundreds or low thousands. If the values continuously increase that can indicate a thread leak.

    number worker
    swap System swap

    Amount of memory swapped by OS to disk.

    Swap should be used as little as possible, as swapped processes are significantly slower. On Linux, swap should be in hundreds of MB or less.

    GB core
    cFD Server Core file descriptors

    Number of currently used file descriptors by Server Core.

    This metric is not available on Windows. If the values continuously increase that can indicate a file descriptor leak. Maximum number of open file descriptors has a limit configured in the OS.

    number core
    wFD Worker file descriptors

    Number of currently used file descriptors by Worker.

    This metric is not available on Windows. If the values continuously increase that can indicate a file descriptor leak. Maximum number of open file descriptors has a limit configured in the OS.

    number worker
    cUptime Server Core uptime

    Uptime of Server Core, i.e. of the Application Container running Server Core.

    seconds core
    wUptime Worker uptime

    Uptime of Worker.

    Useful metric to see when the Worker was restarted - the uptime is re-set to 0.

    seconds worker

    Example Scenarios:

    • Out of heap memory - if the Server Core or Worker run out of heap, it will be visible in the performance log by high GC activity (cGC or wGC) and by their heap usage (cHeap or wHeap) rising to the max heap available.

      Worker running out of heap example

      Figure 19.2. Worker running out of heap example


    • High swap usage - if the system swap is used too much (e.g. several GB) as indicated by the swap metric, then CloverDX will be typically slowed down noticeably, e.g. the jobs will run much slower. This can be caused by insufficient system memory for the heap size (remember that Java needs memory outside of heap). Or some other process is taking the memory. This is often accompanied by high system CPU load.

      High swap usage example

      Figure 19.3. High swap usage example


    • Low limit of file descriptors - CloverDX uses a high number of open files for high performance processing. Default OS limit on maximum open files can be low (e.g. 4096). If the file descriptor metrics (cFD or wFD) rise to a high number and then an issue occurs (typically accompanied by "Too many open files" exceptions in all.log or job logs), then the cause is low limit on maximum open file descriptors (see the section called “Maximum Number of Open Files” in our documentation).

      Low file descriptor limit example

      Figure 19.4. Low file descriptor limit example


    • Other process creating CPU load - if the system CPU load (sysCPU) is high, and significantly higher than Server Core + Worker CPU loads (cCPU + wCPU), then there's probably some other process producing high CPU load.

    Job Queue Log

    The job queue log contains metrics useful for analyzing job queue behavior (see Job Queue). CloverDX logs values of the metrics in regular 3 second intervals into the job queue log.

    Each line of the queue log is one entry, which is a JSON. The line contains a timestamp and values of metrics.

    The job queue log can be found in:

    • Server Console Monitoring > Server Logs > Job Queue
    • {java.io.tmpdir}/cloverlogs/job-queue.log

    Table 19.2. Job Queue Log Metrics

    Metric Label in LogMetricDescriptionUnit
    time time

    Date and time when metrics in this entry were collected.

    The timestamps should be even, i.e. there should not be significant time gaps between the log entries. By default, Server logs the metrics every 3 seconds.

    timestamp
    queued Queue Size

    Number of enqueued jobs waiting for execution. The jobs are in the Enqueued state.

    Maximum queue length is configurable jobqueue.maxQueueSize.

    number
    running Running jobs

    Number of currently running CloverDX jobs. If this Server is part of CloverDX Cluster, only jobs running on this node are counted.

    number
    started Started jobs in interval

    Number of jobs started in the time interval from the previous log entry (3 seconds). Jobs that have queueing disabled (e.g. Data Services, jobs with queueable execution property set to false etc.) are not counted. The number includes both jobs that were enqueued and those that were not (but could have been).

    number
    cpu System CPU load

    Total CPU load of the machine on which Server runs. The system CPU load is caused by all processes running on the machine.

    0-100%
    loadStep Load Step

    This value represents number of enqueued jobs started in one batch, see Job Queue Algorithm for more details.

    If there are available resources, this queue executes loadStep number of jobs at once.

    Advanced metric.

    number
    queueExecutions Queue Executions Counter

    This value counts events when the job queue starts a new batch of enqueued jobs.

    This counter is reset if the job queue becomes empty or loadStep has been increased.

    Advanced metric.

    number
    avgCpu Average CPU load

    Average CPU load measurement, which is used for heuristic that can increase loadStep.

    Advanced metric.

    0-100%
    coreLowMem Core Low Memory Mode

    Core heap memory consumption exceeded safety threshold, queue enters emergency mode (see Job Queue Emergency Mode).

    boolean
    workerLowMem Worker Low Memory Mode

    Worker heap memory consumption exceeded safety threshold, queue enters emergency mode (see Job Queue Emergency Mode).

    boolean
    nonBlockedJobs Running jobs not blocked

    Counts currently running jobs, used only in emergency mode.

    Advanced metric.

    number

    Monitor Log

    The Monitor log tracks changes of Monitors which are part of the Operations Dashboard. The Monitor log contains entries such as deteriorating or improving health of a Monitor, which items started to fail and which recovered, manual reset of health state of monitored items etc. This allows you to analyze what was happening to your data processing in the past.

    The Monitor log can be found in:

    • Server Console Monitoring > Server Logs > Monitor page
    • {java.io.tmpdir}/cloverlogs/monitor.log

    Monitor log

    Figure 19.5. Monitor log