[COCOON-1977] Unsynchronized access to HashMap in ResourceReader

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.1.9, 2.1.10
• Fix Version/s: 2.1.11, 2.2
• Component/s: * Cocoon Core
• Urgency: Urgent
• Assignee: Jörg Heinicke
• Reporter: Ellis Pritchard
• Created: 04/Jan/07 15:24
• Updated: 04/Jan/07 23:53
• Resolved: 04/Jan/07 23:19

Description

I've just had a production server lock up, and when I forced a stack trace I saw lots of these:

"TP-Processor30" daemon prio=5 tid=0x00ce2a50 nid=0x52 runnable [bccfd000..bccffc30]
        at java.util.HashMap.get(HashMap.java:325)
        at org.apache.cocoon.reading.ResourceReader.getLastModified(ResourceReader.java:238)
        at org.apache.cocoon.components.pipeline.AbstractProcessingPipeline.process(AbstractProcessingPipeline.java:468)
        at org.apache.cocoon.components.treeprocessor.sitemap.ReadNode.invoke(ReadNode.java:84)
 etc.

What I've noticed is that the 'documents' HashMap in ResourceReader (a static member) is accessed in an unsynchronized fashion; neither is documents synchronized, or are the puts and gets to it. This is a potential (and actual!!) hazard, and can lead to crashes or hangs.

Suggest that line 94 of ResourceReader.java is changed from:

    private static final Map documents = new HashMap();

to

    private static final Map documents = Collections.synchronizedMap(new HashMap());


Work-around:
enable quick-modified-test in configuration which by-passes use of the document URI cache:

<map:reader logger="sitemap.reader.resource" name="resource" pool-max="32" src="org.apache.cocoon.reading.ResourceReader">
      <quick-modified-test>true</quick-modified-test>
</map:reader>