IBM Cloud Authors: Yeshim Deniz, Elizabeth White, Pat Romanski, Liz McMillan, Stefan Bernbo

Related Topics: IBM Cloud

IBM Cloud: Article

Discovering and Documenting Business Application Patterns

Discovering and Documenting Business Application Patterns

You're probably saying to yourself, "Oh, no! Not another patterns article!" Technically, that's what this is. However, instead of simply showing you a finished pattern, we're going to look at pattern discovery. And, while what we'll talk about can help you capture fundamental patterns (that is, if there are any fundamental patterns left to be defined) we're going to focus on the capture of more workaday patterns. These are patterns that may be useful to only you and your team and might never be made into formal patterns or even shared outside of your group. In fact a pattern purist might say they aren't even patterns.

We believe that what they're called isn't important, instead what is important is making knowledge gained solving one problem useful when solving another similar problem (or even more simply to document what we've done and why we've done it when we have to go back and maintain the code). To effectively do this we have to have a means of capturing, refining, and sharing the knowledge we gain. We believe that capturing this knowledge as a pattern is a great way to do this (and by the way, in doing so, we also believe that you are going to learn more about your design than you expected, very likely including some ways to improve it).

We're going to look at a business application pattern that we discovered and documented. We'll take you through the steps we went through as we discovered and captured it. As we progress through the steps, we'll show you not only what we learned as we captured this pattern, but also things we learned as we captured other patterns.

We'll look at our initial solution to a problem and why we thought it might be a pattern - even with just one instance of the problem. In our next installment, we'll look at another business problem to which we were able to apply our original pattern and, more important, how applying the pattern helped us to define and refine it. In our final installment, we'll look at applying the pattern (with others) to the construction of applications, components, and Web services.

The Problem
The first time we encountered the problem of managing configurable balance information was in the area of warehouse management. We needed some way to manage cached product balances (how much of any particular product was in the warehouse), but we needed to do so in a way that would allow users of our product balances to keep the balances on only the information they wanted. For example, some businesses need to keep track of a product's lot - the "batch" that it was part of. This is especially important when the product is dyed cloth, for example, since only bolts of cloth from the same lot will have exactly the same color.

Other businesses don't care about lots at all. Some businesses use detailed product balances (down to individual zones and bins) for defining pick lists, while other businesses need to track only gross amounts to ensure that product orders can be fulfilled within specified lead times. Our product balances needed to support all of these situations flexibly and without a lot of overhead.

We also knew that we had situations in which there would need to be more product information taken into account when caching the balance. For example, sometimes a product may have multiple colors, so color needs to be taken into account in the balances. On the other hand, many companies will have a different product for each color, so they don't need to define additional attributes to manage color within a specific product. All these possibilities added up to the need for a lot of flexibility in the information the balances could be kept for.

The Solution
Our initial solution was to create a multivalued map that tied items of interest (i.e., the attributes we needed to track) to the cached balance. In other words, we created a lightweight object called the ProductBalanceMapKey that contained the things we supported caching balances over and used it as a key in a map to a number.

Figure 1 shows an example of our simple cached product balances implementation. In our original solution each product had separate balances, so the product ID didn't need to be part of the ProductBalanceMapKey. However, we were interested in the balances for each warehouse and lot, so they do show up in our example. (We're interested in the warehouse because it allows us to know what the delivery time to a customer is - quick if they are next door and slow if they are in a different country.)

While there could be an entry for every possible ProductBalanceMapKey in our map, that wouldn't be the most efficient design. Instead of returning a null when there isn't an entry for a key, we would prefer to return a balance of zero. This behavior could be coded by every user of the map. However, it makes more sense to encapsulate this behavior and the map into a CachedProductBalances class.

Besides being able to retrieve balances, we also needed to keep them up to date. This design allows us to do that - simply build a ProductBalanceMapKey with the right information and modify or add the associated value. This behavior is also encapsulated by the CachedProductBalances class.

The retrieve and update methods are written in terms of the ProductBalanceMapKey. They only use equals (during the map lookup) and don't need to know the details of what's in the key. This design allows us to fulfill our requirement of adding information we're keeping balances over by changing the ProductBalanceMapKey to add the information. However, this doesn't allow us to easily ignore information, for example, in the case where lots are passed in, but we don't need to keep our product balances based on lots.

We resolved this problem by introducing a usage indicator within CachedProductBalances for each piece of information in the key (a Boolean array). We then implemented the CachedProductBalances methods to replace the value in the key with a placeholder whenever the indicator for that key entry is turned off. Thus, for our example above we have a usage array set to all true. If we were to change the lot indicator to false, then the methods on CachedProductBalances would (as part of their processing) replace the values for lot with a placeholder.

The implementation of updateBalance() accumulates entries with duplicate keys; in this case lot values are ignored during the accumulation process. Note that CachedProductBalances doesn't know it is working with a lot. If we think of the key as being an array of information, it simply knows that the information in the same relative position is false and should be replaced.

What we ended up with was a special kind of map key, a key that contained information (that could be changed programmatically), with an associated usage array (that could be dynamically configured). This simple design is shown in Figure 2.

The Start of a Pattern
At this point, we'd normally be happy that we were done and move on to the next problem. However, in this case we stopped and took a longer look at what we'd done. Why this particular problem? On our project we had a number of team members who were very experienced in the business domains we were targeting. They were able to use this experience to tell us that similar problems were going to come up. For example, they knew that we were going to need to cache account balances both for general ledger accounts (for financial reporting and analysis) and customers (for credit checking purposes).

Had we not had the advantage of our domain experts' insight, we would have waited until at least the second or third time we encountered a similar problem before we even thought of our original solution as a candidate for a pattern. Trying to generalize a pattern from a single use is difficult, time-consuming, and error-prone - if not impossible. In general, when evaluating existing designs, you have to be on the lookout for similarities, but don't go overboard - not everything is a pattern and some patterns aren't worth capturing.

Probably the biggest driving force behind pattern definition and capture is the cost/benefit trade-off related to capturing a pattern. Who will be the consumers of our pattern, and how much will they benefit from our pattern documentation? We're very pragmatic in the capture of our patterns. When we identify a design as a candidate pattern we begin by being very informal about documentation - especially when the developer who worked on the first instance is going to be around to answer questions. An important factor in how much information we choose to document is how broadly we expect the pattern to be used. If it's only going be used within the team, often informal documentation is sufficient - there is somebody a new user can talk to about the pattern and the concepts behind it. On the other hand, when a pattern is something that will go outside your organization, it is a good idea to formally document it, since the customer normally can't simply stop by and ask you a question.

Remember that candidate patterns aren't worth spending a lot of time documenting. The second and third applications of the pattern are very likely to dramatically change the pattern. Spending lots of time documenting your candidate pattern is a waste of time, especially if you discover you were wrong and the candidate pattern turns out not to be a pattern (or simply a pattern that it isn't worth your time to document). Often, just documenting what you've done in a simple UML model with associated notes discussing why you think this might be a pattern and some of the key ideas behind it (in other words, leaving a trail of bread crumbs for you to follow later as you refine your pattern) is plenty of detail for this level of pattern definition.

The most important notes to capture in a candidate pattern are any alternatives you explored and why you rejected them. For example, in our case study we looked at simultaneously having ProductBalanceMapKeys of differing sizes. We rejected this design because although it could be done, there was too much overhead associated with it. We may come back later and readdress these alternatives as we apply the pattern to subsequent uses. However, this documentation serves as a starting point and, when the pattern changes, it helps us understand whether or not the pattern still applies in the original case.

As you refine your pattern, you may find that it no longer applies to the first case that started you thinking of the design as a pattern. In other words, the first case may have been an exception to the pattern. The key is not to get trapped by trying to have the pattern solve every single case. SanFrancisco Design Patterns has families of patterns that solve similar problems (proxy, mediator, facade), and you may have run into a similar situation. If the pattern is getting in the way of solving the problem, or adding complexity without value, it is time to start questioning the value or applicability of the pattern.

Even if the original problem turns out to be covered by the final pattern, you need to keep in mind that you may need to rework the original solution. Again this is a trade-off. The solution you have is working, so you don't have to make it adhere to the final pattern. You need to make sure that something wasn't discovered that would be of value in the original case and you have to weigh the value of consistency (which becomes more important when the pattern is exposed to customers and also when maintaining the solution).

The identification of a potential pattern may make you rearrange your development schedule. In our case we started looking at account balances earlier than we originally planned and we put one of our better developers on it. In addition, as the technical leads we paid more attention to this component than we originally planned, periodically reviewing the pattern as it was being developed and actively looking for other appropriate applications of the pattern.


  • When identifying potential patterns:
    - Use your team's technical and domain experts to guide you to interesting areas within your design.
    - Wait for the second or even third occurrence. Be watchful for these.

  • Document patterns to the right level.
    - Candidate pattern documentation shouldn't consume lots of time.
    - Patterns for team use can be very informal.
    - Patterns for customer use usually end up being very formal.

  • Don't get trapped by patterns (or the thrill of capturing them).
    - Sometimes the original problem can't be solved by the final pattern.
    - Sometimes there isn't a pattern.

  • Consider reapplying the pattern once it is refined.
    - May be something new of benefit for the original implementation.
    - May need consistency across implementations.

  • Potential patterns may cause development plan changes.
    - Look at other uses earlier.
    - Change assignments to more experienced developers.
    - Team leaders may be more involved.

    In part 2 of this series, we'll show how our account balances requirements affected our original candidate pattern and how we used that information to provide both our pattern and our original product balances implementation.

  • More Stories By Brent Carlson

    Brent Carlson is vice president of technology and cofounder of LogicLibrary, a provider of software development asset (SDA) management tools. He is the coauthor of two books: San Francisco Design Patterns: Blueprints for Business Software (with James Carey and Tim Graser) and Framework Process Patterns: Lessons Learned Developing Application Frameworks (with James Carey). He also holds 16 software patents, with eight more currently under evaluation.

    Comments (0)

    Share your thoughts on this story.

    Add your comment
    You must be signed in to add a comment. Sign-in | Register

    In accordance with our Comment Policy, we encourage comments that are on topic, relevant and to-the-point. We will remove comments that include profanity, personal attacks, racial slurs, threats of violence, or other inappropriate material that violates our Terms and Conditions, and will block users who make repeated violations. We ask all readers to expect diversity of opinion and to treat one another with dignity and respect.

    IoT & Smart Cities Stories