Oracle® Application Server Containers for J2EE User's Guide
10g Release 2 (10.1.2) Part No. B14011-01 |
|
Previous |
Next |
This appendix describes common problems that you may encounter when using OC4J and explains how to resolve them. It includes the following topics:
This section describes common problems and solutions. It contains the following topics:
Unable to Restart OC4J After Abnormal Termination When OracleAS JMS is Active
Connection Timeouts Through a Stateful Firewall Affect System Performance
OPMN-Managed OC4J Unable to Access EJB Resources Via the Default RMI Port
Application Performance Impacted by Garbage Collection Pauses
Problem
When persistence is enabled in OracleAS JMS, the JMS server creates persistent queues/topics. It also creates lock files (*.lock) associated with these queues/topics in the /persistence directory. If the JVM is terminated abnormally, such as with kill -9
, the lock files are not deleted. This creates a condition in which OC4J cannot be restarted.
Solution
Manually delete all .lock files from the /persistence
directory.
Problem
The common scenario is that failover is seen from OC4J instance A to instance B, but not back again from B to A.
Solution
Standalone OC4J does not require stateful replication to be set up globally for all applications, but instead allows each Web module to configure replication through the <cluster-config>
element in its orion-web.xml
descriptor file. Ensure that this element is populated correctly in each Web module's descriptor.
Note that OPMN-managed OC4J only supports clustering at the global level, and not at the application/module level.
Problem
In this scenario, you wish to use a later version of the JDK with OC4J than the version certified for use with all Application Server Components. However, using a later version of the JDK globally for all Oracle Application Server components increases the risk of breaking certification.
Solution
To use the later JDK version with OC4J only, specify its location in the <java-bin>
element in the opmn.xml
configuration file. For example:
<module-data> <category id="start-parameters"> <data id="java-bin" value="/myjavalocation/jdk/bin/java"/> </category> </module-data>
Problem
This error indicates that the heap size of the Java instance is lower than the memory required by applications running within OC4J.
Solution
Increase the heap size by setting -Xmx
to the desired amount of memory in the <java-option>
element in opmn.xml
:
<module-data> <category id="start-parameters"> <data id="java-options" value="-Xmx256M" /> </category> </module-data>
Alternatively, you can set a system property at OC4J startup:
java -Xmx256M -jar oc4j.jar
If running under Unix/Linux, verify that ulimit
settings allow the JVM process to allocate this much memory.
Problem
To improve performance the MOD_OC4J component in each httpd process maintains open TCP connections to the AJP port within each OC4J instance it sends requests to. In situations where a firewall exists between OHS and OC4J, packages sent via AJP are rejected if the connections can be idle for periods in excess of the inactivity timeout of stateful firewalls.
However, the AJP socket is not closed; as long as the socket remains open, the worker thread is tied to it and is never returned to the thread pool. OC4J will continue to create more threads, and will eventually exhaust system resources.
Solution
The OHS TCP connection must be kept "alive" to avoid firewall timeout issues. This can be accomplished using a combination of OC4J configruation parameters and Apache runtime properties.
Set the following parameters in the httpd.conf
or mod_oc4j.conf
configuration files. Note that the value of Oc4jConnTimeout
sets the length of inactivity, in seconds, before the session is considered inactive.
Oc4jUserKeepalive on
Oc4jConnTimeout 12000
(or a similar value)
Also set the following AJP property at OC4J startup to to enable OC4J to close AJP sockets in the event that a connection between OHS and OC4J is dropped due to a firewall timeout:
ajp.keepalive=true
For example:
java -Dajp.keepalive=true -jar oc4j.jar
Problem
OC4J cannot access EJB resources via the default RMI port when running as a component of Oracle Application Server.
Solution
The most common cause is that a user more familiar with Standalone OC4J is reading the RMI port from rmi.xml
, unaware that the value specified in this file is not used in an OPMN-managed environment.
OPMN-managed OC4J instances use dynamic RMI port assignment. The port value ranges are specified in the <port>
element in opmn.xml
or are determined using dynamic opmn:ormi
lookup from the application client.
See the Oracle Process Manager and Notification Server Administrator's Guide for more information.
Problem
An application running on OC4J appears unresponsive, with simple requests experiencing noticable delays. The cause is that the the JVM has crossed the low memory threshold and is running a full garbage collection to free up memory.
Solution
Consider using the incremental low pause collector, which avoids long major garbage collection pauses by doing portions of the major collection work at each minor collection. This collector (also known as the train collector) collects portions of the tenured generation - a memory pool holding objects that are typically collected in a major collection - at each minor collection. The result is shorter pauses spread over many minor collections.
Note that the incremental collector is even slower than the default tenured generation collector when considering overall throughput.
To use the incremental collector, the -Xincgc
option must be passed in on the Java command line at application startup. Set the inital and maxmium size of the young generation (object pool) to the same value using the XX:NewSize
and -XX:MaxNewSize
options. Set the initial and the maximum Java heap sizes to the same value using the -Xms
and -Xmx
options.
For example, to use this collector with a server with 1GB of physical memory:
java -server -Xincgc -XX:NewSize=64m -XX:MaxNewSize=64m -Xms512m -Xmx512m
For more information on garbage collection tuning, read "Tuning Garbage Collection with the 1.4.2 JavaTM Virtual Machine" which is available at http://java.sun.com/docs/hotspot/gc1.4.2/
Problem
If the OC4J process memory is growing consistently during program execution, then you may have references to invalid symbolic links in your global application.xml
file. This problem is usually characterized by a growth in the C heap and not a growth in Java object memory, as one would see with a more traditional Java object memory leak. OC4J loads all resources using the links in the application.xml file. If these links are invalid, then the C heap continues to grow, causing OC4J to run out of memory.
Solution
Ensure that all symbolic links are valid, and restart OC4J.
In addition, keep the number of JAR files OC4J is configured to load to a minimum . Eliminate all unused JAR files from the configuration and from the directories OC4J is configured to search. OC4J searches all JAR files for classes and resources, thereby causing the file cache to use extra memory and processor time.
You can control the loading more precisely if your <library>
elements in the application.xml
file point to the individual JAR and ZIP files that are needed, instead of to the directories where they reside.
You can search for additional solutions on the following Oracle support-oriented Web sites:
Oracle Application Server Release Notes, available on the Oracle Technology Network at http://www.oracle.com/technology/documentation/index.html
Oracle MetaLink, available at http://metalink.oracle.com
If you still cannot find a solution for the problem you are facing, please log a service request.