VMware offers training and certification to turbo-charge your progress.Learn more
In preparation for the upcoming releases of Spring Cloud Stream (SCSt) 3.0.0 - Horsham and Spring Cloud Function (SCF) 3.0.0, we’ve been publishing a series of posts discussing and showcasing new features and enhancements. We provided motivation and justification for moving away from the annotation-based programming model to the functional model, and then we provided more details on the functional approach, as well as reactive functions. In another post, which is relevant to what we're going to be discussing here, Artem demonstrated the benefits of combining functional approaches with the Spring Integration project. We also talked about event routing in our last post.
In this post, we discuss function composition and Enterprise Integration Patterns (EIP), their commonalities, their differences, and how one can complement the other in the context of SCSt.
"There is no such a thing as a complex problem, since every complex problem is nothing more than an array of simple problems."
Function composition is a feature of SCF that lets you compose several functions together in a declarative way. The following example shows how to do so:
Here, we effectively provided a definition of a single function that is itself a composition of a function named
uppercase and a function named
reverse. You can also argue that we've orchestrated a simple pipeline consisting of running the
uppercase function and then sending its output to the
reverse function. The term orchestration is important here, and we cover it in more detail later in the post.
Enterprise Integration Patterns is a set of patterns that let you describe a business case as a collection of clearly defined and well understood patterns. Some examples are filter, transformer, router, and so on. For more details on EIP, see this link. Spring provides a reference implementation of EIP thru Spring Integration framework. For example, by using the same two function example as before, we can construct a pipeline that uses SI's Java DSL, as follows:
IntegrationFlow.fromChannel(inputChannel) .transform(uppercase) .transform(reverse);
For more on SI's Java DSL, see the Java DSL documentation as well as this quick tutorial.
The core point here is that we've just demonstrated two ways of addressing the same problem thru orchestrating a pipeline. Both can be looked at as orchestrators of complex functionality:
Why is this important? As stated earlier, there is no such a thing as a complex problem, since every complex problem is nothing more than an array of simple problems. So complexity is a composition. However, even complexity can be looked at as something straight forward or something complex. Let’s look at few use cases to set the context:
In the spirit of breaking complexity into simplicity, the first case can be broken down into two services in sequence: compute -> save (similar to our previous uppercase -> reverse example). The second case, while similar, contains a decision point that can then trigger a loop-back that contains some type of additional service call (and so on). In other words, it is not as straightforward as simple compute -> save.
To start, let’s quickly state that EIP and Spring Integration as its implementation can easily handle both use cases. They provide:
On the other hand, SCF, with its function composition feature, can easily handle the first use case and rightfully so. After all,
compute -> save is a collection of functionalities aligned in sequence -
computeFunction.andThen(saveFunction).andThen(..) or using SCF notation
computeFunction | saveFunction.
Also, with SCF composition, it is significantly simpler and more performant, given the differences in the internal implementation between SCF and SI. However, the second use case (which is not fully aligned in a sequence of steps) would be difficult if not impossible to implement with function composition. That is where using a framework such as SI would be the preferred option.
The good thing is that, when broken down, the complexity could still be realized as functions that are recognized as first class citizens by SCF and SI, as described in this post by Artem Bilan. This means that you can defer the decision about your orchestration approach until later to chose SI or SCF or combination of both.
SCF Composition is suited better for orchestrating functionalities that are aligned in sequence, and SI is the better choice for everything else that fits into the category of EIP.