Category Archives: Other

Milestone – a full virtual Asterisk phone system

About four years ago, we switched our office phone system from a proprietary ISDN PBX to Asterisk with Cisco 7940 handsets. It was an instant success and has been running without interruption ever since. It routes incoming faxes to email, provides voicemail, soft-phones and conference rooms, allows us to take our extensions with us wherever we are in the world, provides us with unlimited incoming and outgoing lines over SIP gateways and all the other amazing stuff which Asterisk provides. It was Asterisk which first made me realize that open-source systems can be a whole lot better than closed-source – especially in the proprietary world of telephony.

The only thing which bothered me about our Asterisk PBX was that it was running on a Dell server equipped with a rather obscure Junghanns quad-port ISDN board. It worried me that if either the server or the board died, it would take us quite some time to restore service.

What I really wanted was Asterisk running as a virtual machine, so I could back it up like all our other VMs and move it around within our network infrastructure as needed. Up until now, this wasn’t practical because of high resolution timer issues with the conferencing features. Finally, with the newest Asterisk versions (we’re using the latest beta of version 1.6.2) running on the newest Linux kernels (Debian), its feasible to run it as a VM.

So, as of today, we have our Asterisk PBX virtualized – no moving parts. It connects to our ISDN lines via a Voxtream Parlay voXip gateway. We’re running it on VMWare server 1.0.10 on a dual quad-core Xeon 5420 server with 12GB RAM (the server is running plenty of other VMs alongside the Asterisk VM) . It uses a scarcely noticeable amount of CPU during regular telephony where Asterisk is just connecting SIP end-points. Only conference-rooms, where the Asterisk server is transcoding multiple streams of voip traffic, actually require much CPU resources. Even there, based on our tests, we don’t expect any significant burden on the server.

Wicket component strings

Wicket components have a convenient getString method to load a (translatable string from a properties file associated with the component (or any component in its hierarchy). It also allows you to provide parameters to those strings in the form “${count} new messages” where count is looked up as a property of the model you pass to getString.

Sometimes, however, I’d prefer to just have simple numbered parameters and provide them as varargs to getString(). So, in my properties file, I’d have “${0} new messages” and I’d call “getString(id, 15)” to get “15 new messages”. To do this, you need to add a method something like this:

This method takes your varargs and creates an object array which the standard getString function can use to look up your parameters by index.

Using the HTML TBODY tag as a container for table rows

A common issue which Wicket developers face is the need for an AJAX update of the HTML output of a repeater (such as a ListView) which is part of a HTML table (where the repeating element is represented by a single <tr wicket:id=”mylist”> line in the table). The problem is that you can’t just use “target.addComponent(mylist)” – you need some kind of container around the repeater. Using a nested table or div as a container may interfere with your table layout.

The solution is to use a <tbody wicket:id=”mylistcontainer”> tag around the <tr> which you can add to your page with a WebMarkupContainer and then pass to target.addComponent. It provides you with a container around one or more table rows which you can address from Wicket without messing up your table layout.

Form fields within a Wicket TabbedPanel

Here’s something which took me several days to figure out. If you have form fields within a TabbedPanel (or AjaxTabbedPanel), how do you ensure that they get validated and submitted correctly when the user switches tabs . The TabbedPanel unloads panels as the user switches tabs so only the currently selected tab gets submitted. One suggestion (from Julian Sinai) was to use AjaxFormValidatingBehavior as follows:

AjaxFormValidatingBehavior.addToAllFormComponents(form, “onblur”);

This ensures that form fields get submitted every time they lose the focus (which also happens when the user switches tabs). This however still leaves us with the problem that the tab switch has already occurred before you have a chance to react to validation errors.

The solution I eventually found was to prevent the user switching tabs until any form fields within the selected tab validate correctly.

I did this by overloading the newLink method of the TabbedPanel and returning an AjaxSubmitLink instead of the standard AjaxFallbackLink (wouldn’t it make sense to make AjaxSubmitLink the default if the Panel contains form fields?). Since the AjaxSubmitLink needs a form, I needed to additionally extend the AbstractTab used by the TabbedPanel to query its panel for a form.

Here’s the overload of the newLink method:

Here’s the subclass of the AbstractTab to provide access to the form:

Here’s how the tabs are added:


And here’s how it looks in practice – in the example below, the user is attempting to switch tabs before all required fields have been filled out in the current tab.

The user attempts to switch tabs before all required fields are filled

Installing JRebel

We just started using JRebel on a wicket project with hibernate and Spring. JRebel (formerly known as JavaRebel) provides true hot-swap development under Eclipse – i.e. you never have to republish and wait for hibernate and Spring to reload everything – whenever you save a class JRebel notices and simply reloads the class into the JVM. This greatly increases productivity on larger projects and, perhaps more importantly, removes a major annoyance factor relative to other languages which handle this better.

Installation was a bit confusing because JRebel does not watch for changed classes in the directory where Eclipse puts them by default (at least that was the case for our installation of Eclipse on Mac OS X). To workaround this, you need to add a rebel.xml file to the WEB-INF/classes directory of your project as follows:


<?xml version="1.0" encoding="ISO-8859-1"?>
<application
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xmlns="http://www.zeroturnaround.com"
xsi:schemaLocation="http://www.zeroturnaround.com/alderaan/rebel-2_0.xsd">
<classpath fallback="true">
<dir name="/Users/userx/Documents/workspace/projectx/build/classes"/>
</classpath>
</application>

where the “dir name” element is set to the directory where Eclipse writes the compiled .class files for your project.

In eclipse, you need to modify the launch configuration for tomcat and add the following to the arguments:


-noverify
-javaagent:/path/to/jrebel.jar
-Drebel.trace.log=true
-Drebel.wicket_plugin=true

Corporate backup 2.0

Anyone who has a Mac is familiar with Time Machine, the almost magical, continuous backup capability of OSX. What many people may not know is that Time Machine is based on concepts which have been freely available for quite some time and which can easily be applied to corporate-wide backups. Because corporate backup is considered expensive to implement, many companies have outdated legacy backup systems based on tapes, tape-robots and offsite transport and storage of tapes. These systems are hopelessly outdated and can no longer keep up with the every increasing storage capacity of the disks they should be backing up and the decreasing backup time window in which backups should be completed.

We have approximately 3TB of data (consisting of about 40 databases, 200 virtual machines and hundreds of thousands of files) on our servers and workstations which need to be backed up. About a year ago we installed a comapny-wide backup to disk with offsite replication and versioning which has been providing us with continuous backup ever since. It continuously replicates a 4TB local RAID-6 disk-array offsite to a versioning 5TB RAID-6 disk array over a dedicated 4Mbps line, 24 hours a day, using rsync for the replication and snapshots based on Linux filesystem hard-links for versioning. Its implemented entirely on standard Linux components (zero license costs) and has been running without a glitch for over a year. Thanks to this system, we not only have an offsite backup of all business-critical data, but we can step back to any version of a database or virtual machine from yesterday, two days ago, four days ago, a week old, a month old etc. I can’t imagine why any company would still want to install a propietary backup system when such perfect technology is freely available.

Turnkey appliances

This blog is running on a Turnkey WordPress appliance (www.turnkeylinux.org). Virtual appliances are of course fantastic if they work – a fully configured, just-works server which you can download and provision in seconds (this blog took about 5 minutes to download, install on a VMWare Server virtual machine and get running). Until now, most appliances we tried had enough gotchas to make us return to manual installation on a generic distribution, but the Turnkey appliances seem to be perfect. Based on Ubuntu or Debian and with just enough stuff pre-installed to make them useful, while still being compact enough to compete with a manual installation. Thanks Turnkey!