Implement some performance related properties with TDD
We could add many things to this document! It's a draft version. Don't hesitate to give feedback to jean.bisutti@gmail.com.
Introduction
A first example combining business and performance feedback loops
Evaluate several performance-related properties
Workflow including global performance checks
Conclusion
TDD is about, from @mirjam_diala:
At first, we will implement business, functional, behavior with the help of Test-Driven Development (TDD). In a second step, we will use TDD to implement some persistence performance properties.
Write a failing test on business behavior
Let's imagine that you want to retrieve some data stored in a database.
Let's write a failing test!
@Test
public void should_find_all_players() {
PlayerRepository playerRepository = new PlayerRepository(entityManager);
List<Player> players = playerRepository.findAll();
assertThat(players).hasSize(2);
}java.lang.AssertionError:
Expected size:<2> but was:<0>
The test fails because the implementation returns an empty list:
public class PlayerRepository {
private final EntityManager entityManager;
public PlayerRepository(EntityManager entityManager) {
this.entityManager = entityManager;
}
public List<Player> findAll() {
return Collections.emptyList();
}
}Implement business behavior
Now let's write code to have a green test and to have a working business behavior:
public class PlayerRepository {
private final EntityManager entityManager;
public PlayerRepository(EntityManager entityManager) {
this.entityManager = entityManager;
}
public List<Player> findAll() {
return entityManager.createQuery("FROM Player").getResultList();
}
}Now we have a green test! :)
Refactor the code
Let's clean up a little the code to have something a little more readable:
public class PlayerRepository {
private final EntityManager entityManager;
public PlayerRepository(EntityManager entityManager) {
this.entityManager = entityManager;
}
public List<Player> findAll() {
Query loadPlayerQuery = entityManager.createQuery("FROM Player");
return loadPlayerQuery.getResultList();
}
}The test is still green!
Evaluate a resource usage property: number of JDBC roundtrips
Let's now evaluate a performance-related property! We expect to retrieve the data with one select statement, and so one JDBC roundtrip.
Let's check this with the help of an ExpectSelect(1) QuickPerf annotation:
@QuickPerfTest
public class PlayerRepositoryTest {
@ExpectSelect(1)
@Test
public void should_find_all_players() {
PlayerRepository playerRepository = new PlayerRepository(entityManager);
List<Player> players = playerRepository.findAll();
assertThat(players).hasSize(2);
}Let's restart the test. Now the test is failing!
[PERF] You may think that <1> select statement was sent to the database
But in fact <3>...
💣 You may have even more select statements with production data.
Be careful with the cost of JDBC server roundtrips: https://blog.jooq.org/2017/12/18/the-cost-of-jdbc-server-roundtrips/
Implement a resource usage property: number of JDBC roundtrips
The code should be modified to have only one select sent to the database and get a green test.
Refactor
After, we could check other performance properties, as the number of selected columns:
@QuickPerfTest
public class PlayerRepositoryTest {
@ExpectSelectedColumn(3)
@ExpectSelect(1)
@Test
public void should_find_all_players() {
PlayerRepository playerRepository = new PlayerRepository(entityManager);
List<Player> players = playerRepository.findAll();
assertThat(players).hasSize(2);
}After that, we could for example check the query execution time with a production-like database:
@ExpectMaxQueryExecutionTime(value = 20, unit = TimeUnit.MILLISECONDS)
@ExpectSelectedColumn(3)
@ExpectSelect(1)
@Test
public void should_find_all_players() {
PlayerRepository playerRepository = new PlayerRepository(entityManager);
List<Player> players = playerRepository.findAll();
assertThat(players).hasSize(2);
}It is worth noting that the performance properties do not systematically fail after adding the annotation, unlike the functional properties.
Performance properties are evaluated (and perhaps fixed) one after the other.
By assessing and fixing some performance properties, we can promote performance and scalability at the beginning of application development. These performance properties are like quality attributes added to the feature. In the previous example, we could think that @ExpectSelect(1) and @ExpectMaxQueryExecutionTime are more priority performance quality attributes than @ExpectSelectedColumn. Indeed, @ExpectSelect(1) allows detecting N+1 selects that could lead to many JDBC roundtrips with production data. @ExpectMaxQueryExecutionTime is essential to avoid long queries. Later, we could decrease the maximum permitted query execution time. The number of selected columns can make the query execution time greater and increase the memory consumed and the IO. In some situations, these two last things could be considered first with a low priority even though it is a waste of resources.
If a functional property is broken during an iteration, you can temporarily disable the performance properties' verification by adding @FunctionalIteration on your test method. We try to do one thing at a time; that is to say, we implement the functional behavior, and after that, we check the performance properties.
We could temporarily disable the global performance checks with the addition of @DisableGlobalAnnotations on the test method. We have to explain why in the frame of a TDD workflow... Stay tuned!
