In the present organisation that i work for , we work as four teams continuously developing code and checking it into TFS and as a consequence of this we have adopted the process of continuous integration using Cruise Control.Net . Well using Cruise Control.Net along side Nant to have automated build processes is nothing new, but what we have achieved and are planning to achieve is hopefully a unique implementation of a automated build platform. We have about nine different projects having there own automated builds and since Nant is not able to scale across multiple CPUs, the resources on the server are underutilized and wait times have increased considerably. So as a solution to this we have not installed two instances of Cruise Control on the main build server one called the current build which developers use to build binaries and test , while the nightly build runs on a scheduled basis generating the end product installers. Sometimes people may still need a setup and nightly build allows forcing builds manually using the CC Tray application.
So to achieve this there were a few bottle necks we had to work around.
- Cruise Control does not install two instances completely, for starters it does install the windows service for itself when you install a second instance. So you have to manually install the service using the installutil.exe in the .Net framework.
- Since you will have two instances building simultaneously you will need to isolate the identities under which these build run and the physical locations of these builds.
- The cruise control server web dashboard also has to be configured manually as separate virtual directories.
- The cruise control manager for each instance has to be configured to use a different port, the default install of CC.Net using 21234, we could use something like 21244 as a series . In case you have a firewall make sure your network admin allows requests to this port (CC Tray uses these ports to communicate with the server)
Now that said we have a single server which has two instances of Cruise Control, The build outputs are copied on to a network share. We now have another build server where we are going to replicate the set up of the first server and then split the CC projects across the two physical servers to share the load of the builds. It all looks simple, but thanks to the guy who authored a Templated build where I have been able to enhance it to achieve the following, We have been constantly updating the build or the last one year and we are close to achieving end to end automation for the product, so when we check-in code to TFS , we are able to dish out a CD ISO image for the products to the network share in the nightly build. We also have switches as properties in the build scripts which can allow the current builds to create setups incase the nightly build is overloaded. Due to constant development work and support issues we have to squeeze work like this as part of our non functional sprint work. But then scrum treats engineering work as Non functional and is rightly justified in doing so.