1) the scanner's time doesn't really need to be part of the unit tests, as that can be tested separately and we know it'll work. So we can just call scan() instead. See code in 2)

2) I've also moved it to use a CountDownLatch on a afterChangeSetApplied event.
    private void scan(KnowledgeAgent kagent) {
        // Calls the Resource Scanner and sets up a listener and a latch so we can wait until it's finished processing, instead of using timers
        final CountDownLatch latch = new CountDownLatch( 1 );

        KnowledgeAgentEventListener l = new KnowledgeAgentEventListener() {
            ...

            public void afterChangeSetApplied(AfterChangeSetAppliedEvent event) {
                latch.countDown();
            }
        };

        kagent.addEventListener( l );

        this.scanner.scan();

        try {
            latch.await( 10, TimeUnit.SECONDS );
        } catch ( InterruptedException e ) {
            throw new RuntimeException( "Unable to wait for latch countdown", e);
        }

        if ( latch.getCount() > 0 ) {
            throw new RuntimeException( "Event for KnowlegeBase update, due to scan, was never received" );
        }

        kagent.removeEventListener( l );
    }

3) All file deletes go through the FileManager. I always do multiple attempts at deleting a file and throw an exception if it was not successfuly
    public boolean deleteFile(File file) {
        // This will attempt to delete a file 5 times, calling GC and Sleep between each iteration
        // Sometimes windows takes a while to release a lock on a file
        if ( !file.delete() ) {
            int count = 0;
            while ( !file.delete() && count++ < 5 ) {
                System.gc();
                try {
                    Thread.sleep( 250 );
                } catch ( InterruptedException e ) {
                    throw new RuntimeException( "This should never happen" );
                }
            }
        }

        return !file.exists();

    }

3) File writting always waits 1000ms when the file already exists (http and linux round to nearest second) so it ensures lastmodified is always atleast 1000ms after lastread. See code in 4)

4) All file writting goes via the FileManager where I know attempt to write any file 5 times, calling GC and wait between each iteration.
    public void write(File f,
                      String text) throws IOException {
        if ( f.exists() ) {
            // we want to make sure there is a time difference for lastModified and lastRead checks as Linux and http often round to seconds
            // http://saloon.javaranch.com/cgi-bin/ubb/ultimatebb.cgi?ubb=get_topic&f=1&t=019789
            try {
                Thread.sleep( 1000 );
            } catch ( Exception e ) {
                throw new RuntimeException( "Unable to sleep" );
            }
        }

        // Attempt to write the file
        BufferedWriter output = new BufferedWriter( new FileWriter( f ) );
        output.write( text );
        output.close();

        // Now check the file was written and re-attempt if it was not
        // Need to do this for testing, to ensure the texts are read the same way, otherwise sometimes you get tail \n sometimes you don't
        String t1 = StringUtils.toString( new StringReader( text ) );

        int count = 0;
        while ( !t1.equals( StringUtils.toString( new BufferedReader( new FileReader( f ) ) ) ) && count < 5 ) {
            // The file failed to write, try 5 times, calling GC and sleep between each iteration
            // Sometimes windows takes a while to release a lock on a file
            System.gc();
            try {
                Thread.sleep( 250 );
            } catch ( InterruptedException e ) {
                throw new RuntimeException( "This should never happen" );
            }
            output = new BufferedWriter( new FileWriter( f ) );
            output.write( text );
            output.close();
            count++;
        }

        if ( count == 5 ) {
            throw new IOException( "Unable to write to file:" + f.getCanonicalPath() );
        }
    }

So far between those things the code is now more robust, it's just taking time to move it all across :)

We should probably make separate unit tests specifically for ResourceScannerImpl, on it's own separate from the rest of hte stack.

Mark
On 04/12/2010 13:26, Esteban Aliverti wrote:

Mark, let me know if you need some help since this is one of the task in my (almost eternal) TODO list. You can take a look at KnowledgeAgentEventListenerTest.java. Using KnowledgeAgentEventListener, you can be notified when the rule base is recreated and there is no need to wait for x seconds anymore.
Take a look at the waitUntilChangeSetApplied() method there.

Best,

XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Esteban Aliverti
- Developer @ http://www.plugtree.com
- Blog @ http://ilesteban.wordpress.com

On Sat, Dec 4, 2010 at 6:38 AM, Mark Proctor <mproctor@codehaus.org> wrote:

The KnowlegeAgent tests have been fragile for a while, related to "wait"
code and file issues, particularly on windows. Something has changed
recently and now they aren't working on the linux hudson server either.

So I'm now in the process of refactoring the tests to make them more
robust, and they will hopefully run faster too, as I'll use call backs
when something has finished instead of waiting for X seconds in the hope
that something has finished.

Mark

_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev
_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev