Home > Articles > Programming

  • Print
  • + Share This
This chapter is from the book

This chapter is from the book

Reasons to Break the Rules

An experienced DDD practitioner may at times decide to persist changes to multiple Aggregate instances in a single transaction, but only with good reason. What might some reasons be? I discuss four reasons here. You may experience these and others.

Reason One: User Interface Convenience

Sometimes user interfaces, as a convenience, allow users to define the common characteristics of many things at once in order to create batches of them. Perhaps it happens frequently that team members want to create several backlog items as a batch. The user interface allows them to fill out all the common properties in one section, and then one by one the few distinguishing properties of each, eliminating repeated gestures. All of the new backlog items are then planned (created) at once:

public class ProductBacklogItemService ... {
    public void planBatchOfProductBacklogItems(
        String aTenantId, String productId,
        BacklogItemDescription[] aDescriptions) {

        Product product =
                    new TenantId(aTenantId),
                    new ProductId(productId));

        for (BacklogItemDescription desc : aDescriptions) {
            BacklogItem plannedBacklogItem =


Does this cause a problem with managing invariants? In this case, no, since it would not matter whether these were created one at a time or in batch. The objects being instantiated are full Aggregates, which maintain their own invariants. Thus, if creating a batch of Aggregate instances all at once is semantically no different from creating one at a time repeatedly, it represents one reason to break the rule of thumb with impunity.

Reason Two: Lack of Technical Mechanisms

Eventual consistency requires the use of some kind of out-of-band processing capability, such as messaging, timers, or background threads. What if the project you are working on has no provision for any such mechanism? While most of us would consider that strange, I have faced that very limitation. With no messaging mechanism, no background timers, and no other home-grown threading capabilities, what could be done?

If we aren’t careful, this situation could lead us back toward designing large-cluster Aggregates. While that might make us feel as if we are adhering to the single transaction rule, as previously discussed it would also degrade performance and limit scalability. To avoid that, perhaps we could instead change the system’s Aggregates altogether, forcing the model to solve our challenges. We’ve already considered the possibility that project specifications may be jealously guarded, leaving us little room for negotiating previously unimagined domain concepts. That’s not really the DDD way, but sometimes it does happen. The conditions may allow for no reasonable way to alter the modeling circumstances in our favor. In such cases project dynamics may force us to modify two or more Aggregate instances in one transaction. However obvious this might seem, such a decision should not be made too hastily.

Cowboy Logic

AJ: “If you think that rules are made to be broken, you’d better know a good repairman.”

Consider an additional factor that could further support diverging from the rule: user-aggregate affinity. Are the business workflows such that only one user would be focused on one set of Aggregate instances at any given time? Ensuring user-aggregate affinity makes the decision to alter multiple Aggregate instances in a single transaction more sound since it tends to prevent the violation of invariants and transactional collisions. Even with user-aggregate affinity, in rare situations users may face concurrency conflicts. Yet each Aggregate would still be protected from that by using optimistic concurrency. Anyway, concurrency conflicts can happen in any system, and even more frequently when user-aggregate affinity is not our ally. Besides, recovering from concurrency conflicts is straightforward when encountered at rare times. Thus, when our design is forced to, sometimes it works out well to modify multiple Aggregate instances in one transaction.

Reason Three: Global Transactions

Another influence considered is the effects of legacy technologies and enterprise policies. One such might be the need to strictly adhere to the use of global, two-phase commit transactions. This is one of those situations that may be impossible to push back on, at least in the short term.

Even if you must use a global transaction, you don’t necessarily have to modify multiple Aggregate instances at once in your local Bounded Context. If you can avoid doing so, at least you can prevent transactional contention in your Core Domain and actually obey the rules of Aggregates as far as you are able. The downside to global transactions is that your system will probably never scale as it could if you were able to avoid two-phase commits and the immediate consistency that goes along with them.

Reason Four: Query Performance

There may be times when it’s best to hold direct object references to other Aggregates. This could be used to ease Repository query performance issues. These must be weighed carefully in the light of potential size and overall performance trade-off implications. One example of breaking the rule of reference by identity is given later in the chapter.

Adhering to the Rules

You may experience user interface design decisions, technical limitations, stiff policies, or other factors in your enterprise environment that require you to make some compromises. Certainly we don’t go in search of excuses to break the Aggregate Rules of Thumb. In the long run, adhering to the rules will benefit our projects. We’ll have consistency where necessary, and support for optimally performing and highly scalable systems.

  • + Share This
  • 🔖 Save To Your Account