Question 1

What are the three layers of an enterprise application?

Accepted Answer

Enterprise applications are typically split into three layers of responsibility: presentation (interacting with the user or client program), domain or business logic (the calculations, validations, and decisions the application performs), and data source (communicating with databases, messaging, and other systems). The key rule is that the domain and data source must never depend on the presentation, which keeps presentations swappable and the lower layers testable.

Question 2

What is the difference between a Transaction Script and a Domain Model?

Accepted Answer

A Transaction Script organizes domain logic as one procedure per business action, which is simple and well understood but duplicates logic and tangles as complexity grows. A Domain Model spreads behavior across a web of interconnected objects organized around domain nouns, each holding its own validations and calculations. The model costs more to learn and to persist, but it absorbs complex, evolving logic far more gracefully than a script.

Question 3

When should I use a Domain Model instead of a Transaction Script?

Accepted Answer

Use a Transaction Script when domain logic is simple, because a Domain Model's learning curve and data-source complexity won't be repaid. Switch to a Domain Model as logic grows complex, since scripts hit a wall where each new feature gets exponentially harder. There is no precise complexity metric, so the call is a judgment made by experienced people. Refactoring a script toward a model is reasonable; going the other way rarely is.

Question 4

What is the difference between Active Record and Data Mapper?

Accepted Answer

Active Record is a domain object that wraps a database row and knows how to load and save itself; it fits when the model is simple and maps closely to tables. Data Mapper is a separate layer that moves data between objects and the database while keeping them fully independent, so a rich Domain Model can use inheritance and fine-grained objects free of the schema. Use Active Record for simple cases and Data Mapper for complex ones.

Question 5

What is a Unit of Work?

Accepted Answer

A Unit of Work tracks every object created, modified, or deleted during a business transaction and coordinates writing the changes to the database at commit time. It avoids many tiny database calls, removes the need to keep a transaction open across a whole interaction, and frees developers from remembering which objects changed or ordering writes to respect referential integrity. It acts as the controller of the database mapping.

Question 6

Why is an Identity Map important?

Accepted Answer

An Identity Map keeps one in-memory object per database row within a business transaction. Its primary purpose is correctness: if the same row is loaded into two objects and both are modified, you get inconsistent, conflicting writes. Before loading, you check the map and reuse the existing instance. A useful side effect is that it also serves as a read cache, but that is a bonus, not the main reason to use it.

Question 7

What is Lazy Load and what is its main risk?

Accepted Answer

Lazy Load is an object that does not hold all its data yet but knows how to fetch it on demand, avoiding the loading of a huge object graph when only part is needed. It can be implemented with lazy initialization, a virtual proxy, a value holder, or a ghost. Its main risk is ripple loading: a collection of lazy objects examined one at a time triggers one query per object. Avoid it by making the collection itself the lazy load.

Question 8

What are the three ways to map inheritance to relational tables?

Accepted Answer

Single Table Inheritance stores a whole hierarchy in one table; it is simple and join-free but wastes space on empty columns. Class Table Inheritance uses one table per class, matching the model exactly but needing joins to load an object. Concrete Table Inheritance uses one table per concrete class, avoiding joins but becoming brittle to superclass changes. The trade-off is duplication of structure versus speed of access; single table is a common default.

Question 9

What is the difference between optimistic and pessimistic concurrency control?

Accepted Answer

Optimistic control detects conflicts at commit time: everyone works on a copy, and a commit is rejected if the data changed since it was read. Pessimistic control prevents conflicts by locking data before work begins. Choose based on the frequency and cost of conflicts: optimistic gives better liveness and is easier to implement, so it is the usual default; pessimistic suits cases where conflicts are likely or losing work would be too painful.

Question 10

What is the difference between a system transaction and a business transaction?

Accepted Answer

A system transaction is the database transaction, with full ACID support, delimited by begin and end. A business transaction is the user's unit of work, which often spans multiple requests. Wrapping a whole business transaction in one system transaction creates a long transaction that scales poorly, so you break it into short system transactions and enforce ACID-like behavior across them yourself. That cross-transaction problem is called offline concurrency.

Question 11

What are the SQL isolation levels?

Accepted Answer

From strongest to weakest: serializable guarantees a result equivalent to some serial order; repeatable read allows phantoms (newly inserted rows appearing partway through); read committed allows unrepeatable reads (a value changes if reread); and read uncommitted allows dirty reads (seeing uncommitted, possibly rolled-back data). Lower levels buy throughput by permitting more anomalies, so only serializable guarantees correctness; weaker levels are deliberate risk trade-offs.

Question 12

What does 'don't distribute your objects' mean?

Accepted Answer

It warns against scattering fine-grained domain objects across nodes for performance. A call within a process is extremely fast, but a call between processes or machines is orders of magnitude slower, so distributing by class forces many remote calls and awkward coarse-grained interfaces. The better way to use multiple processors is clustering: put all classes in one process and run copies of that process on several nodes, using fast local calls.

Question 13

What is a Remote Facade and a Data Transfer Object?

Accepted Answer

A Remote Facade is a coarse-grained object placed at a distribution boundary whose only job is to provide a remote interface that delegates to fine-grained internal objects, so only callers needing a remote service pay the cost. A Data Transfer Object carries a batch of data across the boundary in one call, since you cannot send an entangled domain object directly. Together they minimize expensive round-trips by keeping methods and data coarse-grained.

Question 14

Where can session state be stored?

Accepted Answer

Session state can live on the client (URL, cookies, hidden fields), on the server (in memory or as a serialized object), or in the database as tables. Client storage lets a server forget abandoned sessions but ships data each request and is tamperable. Server storage is usually least effort with good performance. Database storage copes well with failures and clustering but needs care to isolate session data from record data.

Question 15

What is a Service Layer and how thick should it be?

Accepted Answer

A Service Layer sits over a Domain Model or Table Module and exposes a use-case-oriented API that the presentation calls exclusively, also serving as a natural home for transaction control and security. It can be a thin facade that forwards to richer domain objects, or it can hold most business logic itself. A common recommendation is to keep it as thin as possible, or skip it until the application clearly needs one.

Layering & Application Architecture(5 questions)

What are the three principal layers of an enterprise application, and what is each responsible for?

What is the difference between a layer and a tier?

What are the benefits and the costs of organizing a system into layers?

How do you tell whether a piece of logic belongs in the domain layer or the presentation layer?

When deciding where to run each layer (client vs. server), what trade-offs guide the choice?

Domain Logic Patterns(6 questions)

What is a Transaction Script and when is it a good fit?

What is a Domain Model, and how does it differ from a Transaction Script?

What is a Table Module, and where does it sit between Transaction Script and Domain Model?

How would you choose among Transaction Script, Domain Model, and Table Module?

What is a Service Layer and how thick should it be?

Why does a Service Layer usually accompany a Domain Model or Table Module rather than a Transaction Script?

Data Source Architecture Patterns(6 questions)

Why separate SQL access from domain logic, and what is a Gateway to a table?

What is Active Record and when is it appropriate?

What is a Data Mapper and what problem does it solve that Active Record does not?

How does the choice of data-source pattern depend on the domain-logic pattern you've chosen?

What is the object-relational impedance mismatch, and what are the main ways to deal with it?

What are some practical rules of thumb for reading data efficiently from a relational database?

Object-Relational Behavioral Patterns(5 questions)

What is the Unit of Work pattern and what problem does it solve?

How can a Unit of Work learn which objects it needs to track?

What is an Identity Map and why is its primary purpose correctness rather than performance?

What is Lazy Load, and what are the main ways to implement it?

When is Lazy Load worth the added complexity, and when is it pointless?

Object-Relational Structural & Mapping Patterns(8 questions)

What is an Identity Field and why is a surrogate key often preferred?

How do you map a one-to-many relationship and a many-to-many relationship to relational tables?

What are the three strategies for mapping an inheritance hierarchy to tables, and how do they trade off?

When would you use Embedded Value versus Serialized LOB to persist objects without giving them their own table?

What is Metadata Mapping and what does it enable?

What is a Query Object and what problem does it address?

What is the Repository pattern, and when does it earn its keep?

Why might you introduce a two-step (logical-to-physical) mapping scheme, and what's the cost?

Web Presentation Patterns(6 questions)

What are the three roles in Model-View-Controller as applied to a web application?

What is the difference between a Page Controller and a Front Controller?

What is the difference between a Template View and a Transform View?

What is a Two Step View and when is it worthwhile?

What is an Application Controller and how do you decide whether you need one?

Why did server-side web development converge on separating request interpretation from response formatting?

Concurrency & Offline Locking(7 questions)

What are lost updates and inconsistent reads?

What is the difference between optimistic and pessimistic concurrency control?

What do the ACID properties guarantee for a transaction?

What are the SQL isolation levels and which read anomalies does each permit?

What is the difference between a system transaction and a business transaction, and why does it create the 'offline concurrency' problem?

How do Optimistic Offline Lock and Pessimistic Offline Lock work, and how should they be combined?

What problems do Coarse-Grained Lock and Implicit Lock solve?

Session State & Distribution(6 questions)

What does it mean for a server to be stateless, and why is statelessness valued?

What are the three places to store session state, and how do they trade off?

What is meant by 'the First Law of Distributed Object Design: don't distribute your objects'?

Where are you actually forced to distribute, and how should you handle those boundaries?

What is a Data Transfer Object and why is it needed alongside a Remote Facade?

When should you choose an XML-over-HTTP (web services) interface versus a platform-native remote call?

Base Patterns(5 questions)

What is a Gateway, and how does it differ from Facade, Adapter, and Mapper?

What is a Layer Supertype and why is it useful?

What is a Value Object, and why should value objects usually be immutable?

What is the Special Case (Null Object) pattern and when do you use it?

What is a Registry, and why should it be used as a last resort?