Robert Hope

Whenever I start a new project, in particular a web project of some kind, there are several steps that I take in preparation for scaling the project or supporting different business functions. Let’s face it, not everything that the website might expect to do should be done by the website; a classic example of this might be notification that a subscription is about to expire. That sort of occasionally scheduled business process is certainly doable in the context of web development, but in general it’s not recommended.

One that might be less apparent might be tasks such as general email delivery, say a forgot password email, where under high traffic one might want to off-load the email delivery to a secondary process rather than hold up rendering of a web page to the browser while the email is built, formatted, and ultimately delivered.

To support a wide variety of potential offline processes, I generally will setup what I might refer to as “harnesses” for three different types of processing. These harnesses are fairly generic, interacting with an interface implementation and typically based of the standard .NET configuration model for instantiation of the class implementing the interface. In many cases, the interface implementation is common to all three harnesses, such that the processes are interchangeable.

There are three basic timing elements for things you might want to accomplish offline from your website. The first timing element I would refer to as “do something repeatedly every so often”. The second timing element I would refer to as “do something on a scheduled basis”, whether that schedule is every few hours or once a day or even once a month. The last timing element I generally prepare for is a “one off” or “one time” execution.

It should be noted that this sort of architecture presupposes that you have full control of your computing environment either through ownership of the servers or access via some sort of cloud computing or virtual hosting service, such that you can install and run items from the console or command line. Obviously this would not be possible if all you had was a web host for your website.

Under those assumptions, I will create three things. First, I will create a Windows Service. This service’s sole purpose would be to take a configured set of objects that implement my interface and run them repeatedly at a specified interval, say once every three minutes. A good example of this might be a process that monitors an email inbox for new messages and processes them in some fashion. Because this is a Windows Service, it might be wise to give each object its own thread, or if you are the latest .NET platform, its own Task.

Second, I will create a standalone console application that I intend to schedule to run regularly. This console application will also load up a configured set of interfaced objects and run them a single time when the scheduled task executes. A good example of this might be some sort of nightly statistical analysis that needs to be done for reporting. In the same sense as the Windows Service, if you have a lot of objects, it might be wise to allow for sequencing some of them in order while noting which ones are truly independent, and then multi-threading them or assigning them Tasks in the proper order.

Last, I will create a near-replica of the standalone console application above, but likely without the multi-threading in place as this is considered to be a one time execution. The application above and this one might even go so far as to be exact copies deployed separately, with one scheduled and one not, if the different requirements for each don’t stray. Common uses of this would be for one-time data conversions, say for example you had inherited a poor database structure that had a person’s name all in one field and you wanted to split into first and last name.

Once I have these three harnesses built, it becomes very easy to generate a plug-and-play approach to any tasks that would need to be scheduled or executed in a predictable fashion without constantly creating new services or new executables to handle the work. This would then allow me to very easily move tasks that my web project might eventually find overwhelming or detrimental to performance off into background tasks without constantly reinventing the wheel. It also allows me, if I so choose, to install or write some code to monitor the execution of these harnesses without having to rewrite the monitoring code every time as well.

This approach has been helpful on several projects and saved me a lot of work down the line.