Pages

Tuesday, October 1, 2013

Deep dive into IBM JDK -Xtrace

-Xtrace Walkthrough

Thanks to Tom Bitonti for this blog post
The goal is to generate trace for an IBM WebSphere Application Server process using the extended java command line option "-Xtrace". The trace is to include entry and exit trace for one constructor and for two methods, including a listing of parameter values, and is to include stack trace the constructor and for one of the methods.
Documentation for the "-Xtrace" java command line option is available here:
http://pic.dhe.ibm.com/infocenter/java7sdk/v7r0/topic/com.ibm.java.lnx.70.doc/diag/tools/tracing.html
Some samples are provided here:
https://www.ibm.com/developerworks/community/blogs/738b7897-cd38-4f24-9f05-48dd69116837/entry/dump_the_stack_trace_information_of_a_method_using_xtrace15?lang=en
While specified for Linux JVMs, this documentation applies to other platforms which use an IBM JDK.
These notes are intended as an addition to the IBM WebSphere Application Server DynaCache Blog:
http://wasdynacache.blogspot.com/2011/07/jvm-xtrace-holy-grail-of-websphere.html
Notes:
  1. "-Xtrace", a command line option to JVM's, is not to be confused with "X-Trace", which is a network diagnostic tool (http://x-trace.net/wiki/doku.php).
  2. The "jstacktrace" trigger action is available in the IBM JVM since IBM Java 6 SR5.
  3. This documentation in for "-Xtrace" in side IBM JVM's. Options for non-IBM JVMs may differ substantially from those described by this documentation.

Goal

Generate trace for an IBM WebSphere Application Server process using the extended java command line option "-Xtrace". Include entry and exit trace for one constructor and for two methods, including a listing of parameter values. Include stack trace the constructor and for one of the methods.
The target constructor is of class:
org.eclipse.emf.ecore.impl.EClassifierImpl
The target methods are:
org.eclipse.emf.ecore.impl.ENamedElementImpl.setName
org.eclipse.emf.ecore.sdo.util.SDOUtil.adaptType

Tracing Method Entry and Exit

Using the "-Xtrace" syntax, entry and exit trace is generated using the "methods" option, while stack trace is generated using "trigger" options. Following the "-Xtrace" documentation:
  • To generate method trace to standard output, "print=mt" must be specified together with method specifications of the target constructor and methods.
  • The method specification places "/" between package names and between a package name and a class name, with "." between a class name and a method name.
  • To include parameter values in the trace, the method specifications are specified with a "()" suffix.
  • A wild card syntax is available, but since specific methods are targetted, this example does not use wild cards.
Following these guidelines, the target constructor and methods are specified as:
org/eclipse/emf/ecore/impl/EClassifierImpl.<init>()
org/eclipse/emf/ecore/impl/ENamedElementImpl.setName()
org/eclipse/emf/ecore/sdo/util/SDOUtil.adaptType()
Combining these method specifications with the "print=mt" constant value and with "-Xtrace" and "methods" option generates:
-Xtrace:print=mt,methods={org/eclipse/emf/ecore/impl/EClassifierImpl.<init>(),org/eclipse/emf/ecore/impl/ENamedElementImpl.setName(),org/eclipse/emf/ecore/sdo/util/SDOUtil.adaptType()}
Notes:
  1. A constructor is specified the same as a method, but using "<init>" as the method name.
  2. No syntax is available to select among multiple constructors for the same class, nor to select among several methods having different parameters.
  3. No syntax is available to restrict a method to invocations on a subclass. In the example, the interest is in "setName" on "EClassifierImpl" instances, but "setName" is implemented on "ENamedElementImpl".
  4. No syntax for selecting static methods or static initialization is currently known.

Displaying Stacks

Stack trace is generate using the "trigger" option and specifying "jstacktrace" as the trigger action. Combining the method specifications with these options generates:
-Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/impl/EClassifierImpl.<init>,jstacktrace}
-Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/sdo/util/SDOUtil.adaptType,jstacktrace}
For simplicity, the method specifications are presented using two distinct -Xtrace command line stanzas.

Completed Command Line Options

The completed command line options are:
-Xtrace:print=mt,methods={org/eclipse/emf/ecore/impl/EClassifierImpl.<init>(),org/eclipse/emf/ecore/impl/ENamedElementImpl.setName(),org/eclipse/emf/ecore/sdo/util/SDOUtil/adaptType()}
-Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/impl/EClassifierImpl.init,jstacktrace}
-Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/sdo/util/SDOUtil/adaptType,jstacktrace}
Notes:
  1. These options may be combined into a single -Xtrace stanza.
  2. The rules for the use of "print=mt" are unclear. Available documentation presents this as required text when outputting to Standard Output.
    1. The option "iprint=mt" may also be used when outputting to Standard Output. "iprint=mt" adds indentation to the output.
  3. To redirect output to a file, the option "output" may be used. This writes to a binary format which must be decoded, for example, using "com.ibm.jvm.format.TraceFormat" which is provided with the JDK. See the linked documentation for additional details on using "output" and on displaying the binary encoded output.
  4. The -Xtrace syntax must be followed exactly. Errors in the -Xtrace options can cause the JVM to fail to start. An error will be displayed. The error might describe the -Xtrace syntax. The error might indicate that the JVM initialization failed, with no clear tie to the fautly -Xtrace stanza.

Setting the Server Configuration

These are specified to the IBM WebSphere Application Server process through the generic JVM arguments attribute, using either the Administrative Console (admin console), or using the Administrative Scripting Client (wsAdmin). This attribute is specified through a "JavaVirtualMachine" entry within a "ProcessDefinition" entry. With no other generic JVM arguments specified, the XML text shows as:
genericJvmArguments="-Xtrace:print=mt,methods={org/eclipse/emf/ecore/impl/EClassifierImpl.<init>(),org/eclipse/emf/ecore/impl/ENamedElementImpl.setName(),org/eclipse/emf/ecore/sdo/util/SDOUtil.adaptType()} -Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/impl/EClassifierImpl.<init>,jstacktrace} -Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/sdo/util/SDOUtil.adaptType,jstacktrace}"
For example, the XML text within the JVM entry shows as:
<jvmEntries xmi:id="JavaVirtualMachine_1183122130078"
    verboseModeClass="false"
    verboseModeGarbageCollection="false"
    verboseModeJNI="false"
    runHProf="false"
    hprofArguments=""
    debugMode="false"
    debugArgs="-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=7782"
genericJvmArguments="-Xtrace:print=mt,methods={org/eclipse/emf/ecore/impl/EClassifierImpl.<init>,org/eclipse/emf/ecore/impl/ENamedElementImpl.setName(),org/eclipse/emf/ecore/sdo/util/SDOUtil.adaptType()} -Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/impl/EClassifierImpl.init,jstacktrace} -Xtrace:print=mt,trigger=method{org/eclipse/emf/ecore/sdo/util/SDOUtil.adaptType,jstackTrace}">
<systemProperties xmi:id="Property_1" name="propertyName_1" value="propertyValue_1" required="false"/> <systemProperties xmi:id="Property_2" name="propertyName_2" value="propertyValue_2" required="true"/>
</jvmEntries>
 


No comments:

Post a Comment

Note: Only a member of this blog may post a comment.