Tutorial notes on. Object relational structural patterns

Transcription

1 Tutorial notes on Object relational structural patterns Dr. C. Constantinides, P.Eng. Computer Science and Software Engineering Concordia University Page 1 of 14

2 Exercise 1. a) Briefly describe what is meant by isomorphism in the context of object-relational mapping, and provide one mapping example where lack of isomorphism occurs. b) Briefly describe filtered mapping and horizontal mapping. c) Use the table below to compare and contrast the two mapping strategies above, on the attributes listed. ATTRIBUTE FILTERED MAPPING HORIZONTAL MAPPING Performance System evolution. Space utilization in the database. d) Deploy filtered- and horizontal mapping to map the class diagram below into two database schemas. For filtered mapping, very briefly describe how you would map an instance of Professor with the field values (01357, C. Constantinides, CSE, Associate Professor ) into a database record. Person id name department Professor rank Student GPA Undergraduate major Graduate researcharea Solution: a) Isomorphism means one-to-one mapping between the elements of two sets. In the context of object-relational mapping, there is no isomorphism between the application domain classes on the one hand and the database tables on the other hand. One example where this is obvious is during filtered (single-table inheritance) mapping. Page 2 of 14

3 b) Filtered mapping: Maps entire system onto a single table. orizontal mapping: Maps each non-abstract class to its own table. c) The table is shown below: ATTRIBUTE FILTERED MAPPING HORIZONTAL MAPPING Performance Performance is very good. Performance is good. System evolution. One large table which can grow (perhaps significantly) with changes such as additional classes. Space utilization in the database. Changes to a class will affect its table and the tables of its subclasses. Space wasted in the database. No unused fields in the database. d) Filtered mapping would result in the following table: <<table>> PERSON Id <<PK>> Name Department Rank GPA Major Research_Area Discriminator For the particular professor record, fields GPA, Major and Research_Area should be set to NULL and the Discriminator should include a code (perhaps of type character) to indicate that this record corresponds to a professor, i.e. (01357, C. Constantinides, CSE, Associate Professor, NULL, NULL, NULL, P). Horizontal mapping on the other hand would result in the following: <<table>> PROFESSOR Id <<PK>> Name Department Rank <<table>> UNDERGRADUATE Id <<PK>> Name Department GPA Page 3 of 14

4 Major <<table>> GRADUATE Id <<PK>> Name Department GPA Research_Area Page 4 of 14

5 Exercise 2. Consider horizontal mapping in the context of object-relational structural patterns. a) How would you capture the attribute visibilities of the object model in the relational schema? b) What possible disadvantage(s) does this mapping provide? c) Consider the following class structure before and after refactoring. In class C, before refactoring, no method other than m needs to access variables x and y. The programmer then decides to adopt shift method up to superclass and shift field to up superclass as shown on the right-hand-side of the table. Briefly describe how this refactoring would affect the tables in the relational schema. How about if class P were concrete? BEFORE REFACTORING abstract class P { protected int x;... } class C extends P { private int y; // method m manipulates x and y;... } AFTER REFACORING abstract class P { protected int x; private int y; // method m manipulates x and y;... } class C extends P {... } Solution: a) We can capture attribute visibilities using foreign keys. b) A possible disadvantage with horizontal mapping (table-per-concrete-class inheritance mapping) is with changes in a class which will have to affect not only the corresponding table, but all tables of all its subclasses. c) No changes are needed. Since only concrete classes are mapped, the table for class C already includes features being inherited from class P. If class P were concrete, tables for both P and C would exist, and both of them would have to be modified (table P to include y, and table C to exclude y). Methods are not mapped to the relational schema. Exercise 3. Consider the following statement: Each class can be easily mapped to a table. Current tools can automate this mapping. It is easy to understand, changes are well localized and the programmer would be expected to produce minimal persistence-related code (if at all). As a Page 5 of 14

6 result, the object-relational conflict is placed to rest. Briefly discuss in no more than ten (10) lines. Solution For complex class structures the mapping must be carefully considered to allow data to be represented and accessed as efficiently as possible. One-to-one (class-to-table) mapping is not always desirable. The case above is one possible solution to a mapping; however, in the case of inheritance there are some problems: The supertype table is accessed very often, leading to join or multiple and queries, resulting in performance issues. Database changes will be required whenever an attribute in the class structure is moved along the class hierarchy. Exercise 4. Consider a system that maintains records of employees and projects, where an employee can be assigned to possibly many projects and where a project may have possibly many participants. An assignment to a project has a start date and an end date. Assume that employees have names and each project has a title. a. Provide a domain model of the system. b. In no more than three lines, describe the rationale behind your model. c. Map the domain model to a relational database schema. d. What mapping are you deploying? e. When we say that a domain class needs to be persisted, what exactly are we referring to? i. How are classes represented in the data model? ii. How are objects represented in the data model? f. Is your mapping isomorphic? Explain. g. Briefly discuss what actions, if any, you need to undertake in order to guarantee consistency of the system (i.e. between domain model and data model). Name any and all applicable patterns that you deploy. h. Name two requirements that a primary key must meet. i. Briefly discuss how you would design and implement the assignment of employees to projects. Describe the rationale of your design with a partial class diagram. Page 6 of 14

7 Solution: 1. A domain model is given below: 2. A many-to-many association calls for an association class (Assignment). Instances of the association class have a life-time dependency on the association. 3. The domain model can be mapped to a relational database schema as follows: 4. We deploy simple mapping where each class maps to a single table. 5. When we say that a class needs persistence, we refer to its instance attributes (variables). 1. Classes are mapped to tables, where each class attribute maps to a column. 2. Objects (individual instances) are represented by table rows (one row per object). 6. As is, the mapping is indeed isomorphic, i.e. there is one-to-one correspondence between elements of the two models (domain model and data model). Page 7 of 14

8 7. To guarantee consistency of the system we must apply identify field into the domain model in order to save the data model primary key, thus maintaining a link between inmemory instances and database rows. 8. A primary key must have two requirements: unique and immutable. 9. Assignment instances must be kept inside a collection class. This calls for a composite aggregation association as shown below: Exercise 5. The collection can be implemented as a Dictionary ADT (either foreign key of the Assignment table can serve as the key to the Dictionary) or any other data structure (static or dynamic) that can serve as a collection. Consider a system that maintains student enrollment information. A student can register to possibly many courses and a course may have possibly many students. A student is identifiable by his or her name and by a (unique) long integer student-id. A course is identifiable by its name, number and term (of offering). For each course, each student obtains a mark. This information is maintained in an enrollment record, one record per student-course pair. 1. Build a domain model of the system. 2. In no more than three lines, describe the rationale behind your model. 3. Map the domain model to a relational database schema ( data model ). Name all pattern(s) that you deploy and in one sentence for each, explain what the pattern involves. 4. When we say that a domain class needs to be persisted, what exactly are we referring to? Page 8 of 14

9 a) How are classes represented in the data model? b) How are associations represented in the data model? (name the pattern) c) How are objects represented in the data model? 5. Is your mapping isomorphic? Explain. 6. Briefly discuss what actions, if any, you need to undertake in order to guarantee consistency of the system (i.e. between the domain model and the data model). Name all patterns that you deploy. 7. The selection of a primary key must meet what criteria? 8. Briefly discuss how you would design and implement the keeping of instances of enrollment records. Clearly describe the rationale of your design with a partial class diagram. 9. Extend the domain model (draw only a partial domain model) by providing a hook for easy extension to support the possible future introduction of professors, administrators etc. (hint: think of reuse). 10. Briefly describe approaches in which such an extension can be mapped into the data model. If performance (e.g. response time) was a critical requirement, which approach would be best suited and why? Page 9 of 14

10 Solution: 1. A domain model is given below: 2. A many-to-many association calls for an association class (Enrollment). Instances of the association class have a life-time dependency on the association. 3. The domain model can be mapped to a relational database schema by deploying the simple mapping pattern (each class maps to a single table) and association table mapping pattern (many-to-many associations are modeled as association classes which, in turn, map into their own table) as follows: Page 10 of 14

11 4. When we say that a class needs persistence, we refer to its instance attributes (variables). a) Classes are modeled as tables where each attribute maps to a column. b) Associations are represented as foreign keys by deploying the foreign key mapping pattern that allows one to navigate from a source table to a destination table. c) Objects (individual instances) are represented by table rows (one row per object). 5. As is, the mapping is indeed isomorphic. Explanation: There is a one-to-one correspondence between the elements of the two models (domain model and data model). 6. To guarantee consistency of the system we must deploy the identify field pattern on the domain model in order to save the data model primary key, thus maintaining a link between in-memory instances and database rows. This need only be done for Course, as in this case the student-id can serve as the primary key in class Student (though this is not absolutely necessary; we can still add a primary key id to the Student class). Page 11 of 14

12 7. The selection of primary key must meet the following criteria: The key must be unique and immutable. The key type must provide a way to compare itself to other keys. The data structure that maintains primary keys must have a way to obtain the next key. 8. Enrollment record instances must be kept inside a collection class. This calls for a composite aggregation association as shown below: The collection can be implemented as a static or dynamic data structure. 9. Reuse can be supported through inheritance. 10. There are three ways to map inheritance: a) Filtered mapping (or single-table inheritance mapping): the entire hierarchy is mapped to a single table, b) Vertical mapping (or table per class inheritance mapping): every class (abstract or concrete) is mapped to a table, Page 12 of 14

13 Exercise 6. and c) Horizontal mapping (or table per concrete class inheritance mapping): every concrete class is mapped to a table. In the case where performance is a critical requirement, filtered mapping would be best suited since there are no multiple tables that would require join or multiple queries. This problem demonstrates the deployment of object-relational structural patterns. a) Consider class Point whose UML representation is shown below. Use this class as an example in order to briefly describe the motivation behind Simple Mapping. b) Use class Point as an example to describe the motivation behind Identity Field. What changes, if any, would be necessary to the class after the deployment of this pattern, and why? c) Deploy single-table-inheritance to map the class diagram below onto a relational database schema. What technique can you adopt to distinguish between instances of Point and those of 3DPoint? Page 13 of 14

14 Solution: a) Simple Mapping implies the representation of a domain class to a database table where the class' attributes become columns in a table and where each table row would correspond to an object. In this example we will have b) The identity Field pattern introduces a primary key to distinguish between rows (objects). Because of this, we need to save a database id field in class Point to maintain a link between an in-memory Point object and its representation in a database row. c) Single-table-inheritance mapping would give us the following table where the id is the primary key. <<table>> POINT id <<PK>> x y z discriminator We can distinguish between rows by setting unused fields to null and by using a discriminator column that would determine the type of object. Page 14 of 14