Description of problem:
If there are multiple resources on the same agent parsing the log4j log files and producing events from them, some of them would randomly stop collecting the events for no apparent reason.

The reason why they fail is that Log4JLogEntryProcessor class in the plugin-api module uses shared (static) DateFormat instances to parse different supported date strings in the log files. The DateFormat is not thread-safe and would randomly fail if multiple simultaneous parse() calls were made on them.

Version-Release number of selected component (if applicable):
4.5.0-SNAPSHOT

How reproducible:
always

Steps to Reproduce:
1. Setup multiple resources on the same agent for parsing log4j logs
2. Make the managed resources write something into the log files
3. Wait a bit for the agent to generate events from the log file entries.
  
Actual results:
Randomly, some of the resources will no longer report new events even though the underlying log files change.

Expected results:
The event generation works as expected.

Additional info:
This bug has been fixed in a separate branch with the work on the hadoop plugin by the following commits:

feature/hadoop-plugin http://git.fedorahosted.org/cgit/rhq/rhq.git/diff/?id=f719aa14b69abda856fe51e1d45db3f9b7588c85
Author: Lukas Krejci <lkrejci>
Date:   Mon Aug 6 20:48:13 2012 +0200

    Make the date formats used by the log4j log processor instance fields so
    that they are not shared by multiple instances that would use those formats
    from different threads.
    The DateFormat is not a thread safe class.

feature/hadoop-plugin http://git.fedorahosted.org/cgit/rhq/rhq.git/diff/?id=c82eb40efe105f8d719cbc16592add9e259dbd10
Author: Lukas Krejci <lkrejci>
Date:   Mon Aug 6 19:56:30 2012 +0200

    More rigorous error handling while processing the log files to prevent
    a runtime exception cancel the scheduled job for log processing for given
    resource.