Someone just asked me to define “best practices” for a collection of application deployments. The question was impossible to answer because the applications we were talking about were all bespoke, so each one had good and bad ways of deploying them. It would take an age to go through each one and explain which is the best way of doing their unique installation operations.

However, I’ve given it some thought and I think there are still a handful of best practices for application deployments which can pretty much extend across almost all applications, certainly all the ones we were talking about. Here are some of what I would define as best practices for application deployments.

1. Keep the installation structure SIMPLE. Files and directories should be kept to a minimum. Don’t install anything that’s never going to be used.
2. Always get rid of old files. When something goes wrong on a production host, the last thing I want to be doing is trawling through random directories and copies of old files to find what’s gone wrong.
3. Automate it – this almost goes without saying, but deployments should NOT be manual, there’s far too much room for human error. Use a tool for doing deployments, something that supports the native OS operations, like rpm using yum for RedHat. Alternatively, if you’re deploying to multiple different OSs, try using a scripting language to script the deployments.
4. Don’t over do it with the symlinks. Use them only if you have to. It’s too easy to end up with symlinks pointing to the wrong place, or to break them altogether. It’s also a bad idea for your applications themselves to rely on symlinks. I would rather enforce standardisation of the environments and have my applications use real paths than rely on symlinks. Symlinks simply add another level of configuration and a reliance on something else which is all too breakable.
5. Delete everything first. If you’re simply deploying a new directory or package, completely remove the existing one. Take a backup if necessary, but delete that backup at the end if the deployment is successful. This is similar to point 2, but more robust. I think that if at all possible, you shouldn’t rely on sync tools like rsync/xcopy/robocopy to do your deployments. If your time and bandwidth allows, delete everything first and upload the complete new package, not just a delta.
6. Have a roll back strategy. Things can sometimes go wrong and often the best policy is to roll-back to a known-working version. Keeping a backup of the last known working version locally on the target machine can often be the quickest and simplest method, but again I would avoid this option if bandwidth allows. I don’t like having old versions sitting around on servers, it leads to cluttered production boxes. I would much rather do a roll-back using the same mechanism used for doing a normal deployment.
7. Don’t make changes to your deploy mechanism or deploy scripts between deploying to different environments. This is just common sense, but I’ve seen a process where, in order to actually deploy something onto a production server, the deploy script had to be manually changed! Suffice to say I wasn’t a fan of that idea.

As you can probably appreciate, this list is generally not written with installers (such as msi files) in mind, maybe I’ll look at that another time.