Using Hudson CI server to build multiple projects with FlexUnit4 tests on a windows box in parallel

June 29th, 2010

Recently I setup the Hudson CI server for my splink projects. As Hudson is very well documented and super-easy to use, I had it up and running within less than 5 minutes.

1. Download the latest Hudson.war
2. Run Hudson "java -jar hudson.jar"
3. Install Hudson as a windows service

My next step was to add build jobs for the splink projects. For each project (splinklibrary, splinkresource and deepsplink I did

1. Add a "Free Style" job
2. Configure the source code repository (firstly I had to install the Hudson Mercurial plugin, but for instance SVN is supported out of the box.
3. Define when to run the job (i.e. every hour, after a commit, after another build, ...)
4. Add ANT scripts and define which tasks are to be run. (i.e. compile, run-tests, package, release, ...)
5. Define what to do after the ANT scripts have been successfully run. (i.e. publish documentation, publish test results, archive releases, ...)
6. Save the job

Now I could click the 'build button' and Hudson successfully built my project. Great!

But wait,..

Next I configured the three jobs to run every hour. But almost always one of the jobs failed. After looking at the console output of the failed builds I discovered the problem:

The FlexUnit tasks communicate with the swf which runs the FlexUnit tests over a socket on port 1024. So if two jobs run at a time, they interfere with each other because both jobs try to use port 1024 when they run the FlexUnit tests.

So I needed to assign unique TCP ports to avoid these port collisions and the Hudson Port Allocator Plug-in seems the best tool for the job. Just install it here
With the port allocator in place I could allocate one or more ports for a job. My first solution was to just define port 1024 for each job and be done. Port Allocator then queues all the jobs which use port 1024 instead of running them in parallel. On the upside: No more failed builds. But on the downside the builds take longer, as they can't execute in parallel anymore.
The answer to this was to make the Hudson Port Allocator Plug-in pick a free random port and convey the port to the build script. Because if each job is assigned with it's own port for the FlexUnit socket communication, jobs can't jam each other anymore.

Make Hudson Port Allocator convey a free port to the build script

Now I could grab the PORT enviroment variable in my build file and use it to configure the Hudson ANT task.

 
<property environment="env"/>
<condition property="PORT" value="${env.PORT}" else="1024">
	<isset property="env.PORT" />
</condition>
 
<flexunit
	swf="${tests}/TestRunner.swf"
	toDir="${report}"
	haltonfailure="false"
	headless="false"
	verbose="true"
	port="${PORT}"
	localTrusted="true">
</flexunit>
 

But as the FlexUnit CIListener class also needs to know the port, I employed ANT's echo task to write the port number into a file:

 
<echo file="${tests}/port">${port}</echo>
 
 
core = new FlexUnitCore();
core.addListener(new CIListener(port));
 

And to get the port inside my TestRunner class I had to just load the "port" file:

 
public function TestRunner() {
	var loader : QUrlLoader = new QUrlLoader(new URLRequest("port"));
	loader.register(QEvent.COMPLETE, function(e : QEvent):void {
		start(loader.getContent());
	});
	loader.register(QEvent.ERROR, function(e : QEvent):void {
		start();
	});
	loader.start();
}
 
private function start(port : uint = 1024) : void {
	core = new FlexUnitCore();
	core.addListener(new TraceListener());
	core.addListener(new CIListener(port));
	core.run(IntegrationSuite);
}
 

* QUrlLoader is part of the splinklibrary project.

By checking the console output after running a build job I was able to verify that Hudson Port Allocator indeed picked different ports for each job:

hudson-allocating
hudson-allocating-2


No Comments »




googlecode: converting svn to mercurial

June 20th, 2010

Just converted the splink googlecode repositories from svn to mercurial dvcs. (including the complete commit history) It worked like a charm once I figured that I had to use TortoiseHg's hg.exe to perform the conversion on my windows machine. TortoiseHg includes all neccessary extensions (python+svn bindings) and works out of the box.

1. Convert the library:

 
mkdir mylibrary
hg.exe convert http://mylibrary.googlecode.com/svn mylibrary
cd mylibrary
hg.exe push https://mylibrary.googlecode.com/hg
 

2. Convert the wiki:

 
mkdir mylibrary-wiki
hg.exe convert http://mylibrary.googlecode.com/svn/wiki mylibrary-wiki
cd mylibrary-wiki
hg.exe push https://wiki.mylibrary.googlecode.com/hg
 

No Comments »




Powered by WordPress