Phase 2: Refactor Service Tests to Use Testcontainers #529

Overview

Refactor BackfillEmbeddingsServiceTest and SearchContentServiceTest to use Testcontainers with real production code instead of Mockito mocks.

Priority: HIGH
Estimated Time: Week 2
Depends On: #528 (Phase 1)

Context

Replace mock-based testing with integration tests using:

• REAL ContentPersistenceAdapter and ArcadeContentRepository
• REAL ArcadeDB with actual vector index
• FakeEmbeddingGenerator for deterministic embeddings
• State-based assertions instead of mock verify() statements

Benefits

• ✅ Tests validate actual database behavior, not mock interactions
• ✅ Catches SQL errors, schema issues, and mapping bugs
• ✅ Tests survive refactoring (no brittle mock setups)
• ✅ More readable with Given/When/Then structure
• ✅ Validates real vector search with LSM_TREE index

Tasks

2.1 Refactor BackfillEmbeddingsServiceTest

File: src/test/java/it/robfrank/linklift/application/domain/service/BackfillEmbeddingsServiceTest.java

Changes:

• Extend ArcadeDbTestBase instead of using @ExtendWith(MockitoExtension.class)
• Replace mocked LoadContentPort and SaveContentPort with real repository
• Replace mocked EmbeddingGenerator with FakeEmbeddingGenerator
• Update all 11 tests to:
  • Use repository.saveContent() to set up test data
  • Verify actual database state with repository.findContentById()
  • Remove all verify() statements
  • Use Awaitility instead of Thread.sleep() where possible

Example transformation:

// Before (with mocks)
when(loadContentPort.findContentsWithoutEmbeddings(100))
    .thenReturn(List.of(content))
    .thenReturn(List.of());
verify(saveContentPort, times(1)).updateContent(argThat(c -> c.embedding() != null));

// After (with Testcontainers)
repository.saveContent(content); // Setup test data in REAL database
backfillEmbeddingsService.backfill();
Thread.sleep(1000);
Content updated = repository.findContentById("id-1").orElseThrow();
assertThat(updated.embedding()).isNotNull().hasSize(384);

2.2 Refactor SearchContentServiceTest

File: src/test/java/it/robfrank/linklift/application/domain/service/SearchContentServiceTest.java

Changes:

• Extend ArcadeDbTestBase
• Replace mocked LoadContentPort with real repository
• Replace mocked EmbeddingGenerator with FakeEmbeddingGenerator
• Update all 21 tests to:
  • Save content with embeddings to real database
  • Verify actual vector search results using ArcadeDB's LSM_TREE index
  • Remove all verify() statements
  • Test actual cosine similarity ranking

Example transformation:

// Before (with mocks)
when(embeddingGenerator.generateEmbedding(query)).thenReturn(queryVector);
when(loadContentPort.findSimilar(queryVector, 10)).thenReturn(List.of(resultContent));

// After (with Testcontainers)
Content ml1 = content.withEmbedding(embeddingGenerator.generateEmbedding("machine learning"));
repository.saveContent(ml1);
List<Content> results = searchContentService.search("machine learning basics", 2);
assertThat(results).extracting(Content::id).containsExactlyInAnyOrder("id-1", "id-2");

2.3 Update Test Structure

For both test classes:

• Add @AfterEach to call super.tearDownDatabase() and clean up ExecutorService
• Use Given/When/Then comments for clarity
• Replace Thread.sleep() with Awaitility where feasible
• Ensure all tests use FIXED_TEST_TIME for deterministic dates

2.4 Validation

• All 32 tests pass (11 from BackfillEmbeddings + 21 from SearchContent)
• No Mockito verify() statements remain in service tests
• Tests use real database state assertions
• Measure total execution time (target: under 40 seconds)
• Verify tests catch SQL/schema errors (try intentional typo)

Success Criteria

• Both test classes extend ArcadeDbTestBase
• All mocks replaced with real implementations
• All 32 tests pass with Testcontainers
• Tests validate actual database state
• Execution time under 40 seconds
• Code coverage maintained or improved

Related

• Depends on: Phase 1: Setup Testcontainers Infrastructure for Vector Search Tests #528 (Phase 1)
• See TEST_REFACTORING_PLAN.md Phase 2 for examples
• Blocks: Phase 3: Optional E2E Tests with Real Ollama (LOW PRIORITY) #530 (Phase 4 - HTTP tests)