Regex validation fails in multi-threaded, multi-processor environment

• Key: XMLBEANS-46
• Type: Bug
• Status: Resolved
• Resolution: Fixed
• Priority: Major
• Assignee: Kevin Krouse
• Reporter: Dejan Predovic
• Component/s: Validator
• Affects Version/s: Version 1.0.3, Version 2
• Fix Version/s: unspecified

Description

When using validate() method in multi-threaded, multi-cpu environment, there is a pretty high probability of getting false XMLErrors for elements defined with regex patterns in schema.

import EDU.oswego.cs.dl.util.concurrent.CountDown;
import junit.framework.TestCase;
import org.apache.xmlbeans.XmlException;
import org.apache.xmlbeans.XmlOptions;
import org.apache.xmlbeans.XmlError;

import java.util.Collection;
import java.util.ArrayList;
import java.util.Iterator;

public class XmlBeansTest extends TestCase {

    public void testMultipleThreads() throws XmlException, InterruptedException {
        final int totalThreadCount = 100;
        CountDown latch = new CountDown(totalThreadCount);
        CountDown start = new CountDown(1);

        long now = System.currentTimeMillis();

        System.out.println("Starting");
        for (int i = 0; i < totalThreadCount; i++) {
            new ParseAndValidate(latch, start).start();
        }

        start.release();
        latch.acquire();

        System.out.println("Finished: " + (System.currentTimeMillis() - now));
    }

    public static void main(String[] args) throws Exception, InterruptedException {
        XmlBeansTest xmlBeansTest = new XmlBeansTest();
        xmlBeansTest.setUp();
        xmlBeansTest.testMultipleThreads();
    }

    private class ParseAndValidate extends Thread {

        private CountDown countDownLatch;
        private CountDown start;
        private final String xmldata = ...;

        public ParseAndValidate(CountDown countDownLatch, CountDown start) {
            this.countDownLatch = countDownLatch;
            this.start = start;
        }

        public void run() {
            try {
                start.acquire();
                boolean valid = parseAndValidate();
                if (!valid) {
                    System.out.println("Not Valid!!!");
                }
                countDownLatch.release();
            } catch (XmlException e) {
                e.printStackTrace();
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
        }

        private boolean parseAndValidate() throws XmlException {
            MyXml xml = MyXml.Factory.parse(xmldata);
            return validate(xml);
        }

        private boolean validate(MyXml rdd) {
Collection errors = new ArrayList();
XmlOptions validateOptions = new XmlOptions();
validateOptions.setErrorListener(errors);
boolean valid = rdd.validate(validateOptions);
if (!valid) {
for (Iterator iterator = errors.iterator(); iterator.hasNext();) {
XmlError xmlError = (XmlError) iterator.next();
System.out.println("XML Error - " + xmlError.getMessage() + " at\n" + xmlError.getCursorLocation().xmlText());
}
}
return valid;
        }
    }
}

The code above, when executed on a MP machine, has a pretty high probability (5%+ I'd say) of generating at least one failed validation. On machines with hyperthreading, the probability is even higher (10%+).
I only saw it fail on regex patterns. On SP machines it never fails. The only workaround I've found is encasing a validate method in a globally synchronised block (not really a workaround, I know).
The code also runs very slowly on MP machines with sun jdk, but that's unrelated to the problem I guess.