SOA Anti-Pattern : The Knot
SOA Anti-Pattern : The Knot
Join the DZone community and get the full member experience.Join For Free
SnapLogic is the leading self-service enterprise-grade integration platform. Download the 2018 GartnerMagic Quadrant for Enterprise iPaaS or play around on the platform, risk free, for 30 days.
Everything starts oh so well. Embarking on a new SOA initiative the whole team feels as if it is pure green field development. We venture on - The first service is designed. Hey look it got all these bells and whistles; we are even using XML so it must be good. Then we design the second service, it turns out the first service has to talk to the second – and vice versa. Then comes a third, it has to talk to the other two. The forth service only talks to a couple of the previous ones. The twelfth talks to nine of the others and the fourteenth has to contact them all – yep our services are tangling up together into an inflexible, rigid knot.
The above scenario might sound to you like a wacky and improbable scenario - why would anyone in the right mind do something like that? Let’s take another look, with a concrete example this time and see how the road to hell is paved with good intentions. In Figure 10.1 below we see a vanilla ordering scenario. An ordering service sends the order details to a stock service, where the items are identified in the stock, marked for delivery and then sent to a delivery service which talks to external shipping companies such as DHL, FedEx etc.
Figure 10.1 a vanilla ordering scenario. An ordering service sends the order to a stock service, which provisions the goods to a delivery service which is responsible to send the products to the customer
If we think about it more we’ll see that when an item is missing from the stock we probably have to talk to external suppliers, order the missing items and wait for their arrival- so the whole process is not immediate. Furthermore since the process takes time, it seems viable to cancel the process if an order is cancelled. It seems we have two options (see Figure 10.2) either the ordering service will ask the two other services to cancel processing related to the order or the two services call the ordering service before they decide what to do next. Naturally the system wouldn’t stop here, we would want to introduce more services and more connections e.g. an Accounts Payable service that interacts with the external suppliers, the stock service and the delivery service(since we also need to pay shipping companies) etc.
Figure 10.2 a little more realistic version of the Ordering scenario from figure 10.1. Now we also need to handle missing items in the stock, cancelled orders and paying external suppliers. In this scenario the services get to be more coupled. For instance the Ordering service is now aware of the delivery service and not just the stock service.
Well, yes – and no. Yes it is normal for services to connect to each other. After all, creating a system in an SOA is connecting services together. As for the “no” part, the problem lies with the way we develop these integrations if you are not careful it is easy to get all the integration lines in a big, ugly mess – a knot
A knot is an Anti-pattern where the services are tightly coupled by hardcoded point-to-point integration and context specific interfaces
For instance, what happens when we want to reuse the ordering service mentioned above? No problem, we just call it from the new context. Alas, the knot prevents us from reusing it without hauling in the rest of the baggage - all the other services we defined above (the stock, delivery etc.) if the new context is not identical in it ordering processes and matches what we already have we can’t use it. Or we can’t use it without adding one-off interfaces where we add specific messages for the new context and all sort of “if” statements to distinguish between the old and the new behavior. Another option is to make this distinction in the original messages, which either not possible or forces us to make sure the other services are still functioning. In any event it is a big mess.
Let’s recap. We moved to SOA to get flexibility, increase reuse/use within our systems, prevent spaghetti point to point integration – what we see here is not flexible, hard to maintain and basically it seems like we are back in square one and we invested gazillions of dollars to get there.
1.1.2 CausesHow did that happen? How can a wonderful, open standards, distributed, flexible SOA deteriorate to an unmanageable knot?
It is tempting to dismiss the knot as the result of lack of adequate planning. If we only planned everything in advance we wouldn’t be in this mess now. Well, besides the point that trying to plan everything ahead of time is an anti-pattern in itself (an organizational anti-pattern – which isn’t in the scope of this book). There’s still a good chance you’d get to a Knot anyway since the problems are inherent in the way business work.
Figure 10.3 the Knot anti-pattern is similar in both effect and origin to the spaghetti integration in non-SOA environments
The workflodize pattern suggests adding a workflow engine inside the service to handle both Sagas (i.e. long running operations, see chapter 5) and added flexibility. The “added flexibility” is the card we want to play here. When we express the connections as steps in the workflow they are not part of our services’ business logic. They are also easier to change in a configuration-like manner both of these points are big plusses.
Still, a better way to solve the service to service integration problem is to use an external orchestration engine. The idea of using the Orchestrated Choreography pattern is to enable Business Process Management- or a way for the organization to control and verify it processes are carried out as intended (you need an orchestration engine for that but it helps…). In the context of solving or avoiding the Knot anti-pattern, Orchestrated Choreography is better than Workflodize since it centralizes and externalizes all the interactions between services and thus effectively removing all the problematic code from the services themselves. Note that there’s a fine line between externalizing flow and externalizing the logic itself (see discussion in Orchestrated Choreography pattern, in chapter 7).
Figure 10.4 the Ordering service using the Inversion of Communications pattern. Now the service doesn’t know/depend on other services directly. It is only aware of the business events of new order and cancelled order which are relevant to the business function that the service handled
1.1.1 Known Exceptions
In a sense the Knot is a distributed version of an anti-pattern described by Brian Foote and Joseph Yoder as “Big Ball of Mud” – spaghetti code where different types of the system tied to each other in unmanageable ways. The reason for mentioning the connection is that the reason that “Big Ball of Mud” might be considered a pattern rather than an anti-pattern also apply here:
[when] you need to deliver quality software on time on budget… focus first of feature and functionality, then focus on architecture and performance
Starting out on a large project, such as moving an enterprise to SOA, is difficult enough as it is. You can’t figure everything in advance; you need to deliver something – so as Nike says “just do it”. Get something done. You do need to be prepared to let go and redesign further down the road. In the current system I’m working on – a visual recognition/search engine for mobile, we went with a “knot” approach for the first release. The simplicity of the implementation, i.e. less investment in infrastructure, ad hoc integration etc. enabled us to deliver a first working version in less than 6 months. These 6 months also helped us understand the domain we are operating in much better and more importantly get to market with the feature the business needed in the schedule the business wanted. We spent the next 6 month rewriting the system in a proper way, including applying the Inversion of Communications pattern mentioned above.
Opinions expressed by DZone contributors are their own.