I started off this post with the title "Our Quest for Agility" but it kind of trailed off into being about our latest customer-dashboard tool. I'll do another post about our agile quest later.
The Greenback project
Before the summer, I was issued with the task of setting up a customer-management portfolio tool, which I named Greenback (as in, money, because that's what we want from our customers). The first goal was to provide our customer-managers with an overview of which customers exists, where they are installed, which applications they are running and other configuration details. Greenback is basically a web application with a big table with our customers and some filters to see the customers for a single manager, or with a particular configuration. Our customer-managers can also edit, add and comment all the entries, naturally.
We have a rather large portfolio of customers (200+), and each of these are hosted with their own application of our software. Through development, we strive to keep all these customers in a working state for testing, but it's not always easy. Developing a feature for one customer will some times break the functionality in many others. Maintaining the test-data for every customer is also a challenge.
Automatic deployment and web testing
At the same time, I started pondering about how we could continuously test all these customers' applications. First thing, we got automatic deployment set up. I then made a simple web-test that basically tried loading the front page of a customer's application on the nightly deployment. I then looped this test over every customer, every morning (in our CI-environment/Hudson).
At the same time, we made a big overhaul of our test data to get every customer's test-data into working condition. It didn't take long before the first customer's application broke again on the nightly build. I started wondering: How can we keep every customer in a working state? We need some kind of proper reporting that people will respond to, because scrolling through a console output to see which applications are broken simply does not work over time.
Then it occurred to me. Why not report the results of the web-tests back to the Greenback overview? I put together a little call-back at the end of our customer-test that wrote the result into Greenback, merely as a traffic light: Red meant that the setup scripts for the customer were not working at all, yellow meant that the application was working, but the front page wasn't working properly, and green meant everything is OK. I then made a new column in our customer table with the color of the test-result, and a link to the nightly deployment for those who want to investigate any problems, or test a particular application.
Later on, I also added a little test-history so customer-managers and developers alike could track when a certain application stopped working (we can't always drop what we're doing and fix test data every time we see a red light). Next in line is to produce a feed that people can subscribe to, to see any changes in the state of applications.
The great advantage of this is that the customer-managers themselves can take ownership in making sure that their applications are working. It was all in all pretty easy to set up (a few weeks of work, all in all), and I hope others with similar needs will find some inspiration in this.