tessl/maven-org-springframework-ai--spring-ai-vector-store
Common vector store functionality for Spring AI providing a portable abstraction layer for integrating vector databases with comprehensive filtering, similarity search, and observability support.
edge-cases.mddocs/examples/
Edge Cases and Error Handling
Learn how to handle errors, edge cases, and exceptional scenarios in Spring AI Vector Store.
Error Handling Patterns
Pattern 1: Duplicate Document IDs
Some vector stores throw errors when adding documents with duplicate IDs.
try {
vectorStore.add(documentsWithDuplicateIds);
} catch (RuntimeException e) {
logger.error("Duplicate document IDs detected", e);
// Strategy 1: Generate new IDs
List<Document> withNewIds = documentsWithDuplicateIds.stream()
.map(doc -> new Document(
UUID.randomUUID().toString(),
doc.getContent(),
doc.getMetadata()
))
.toList();
vectorStore.add(withNewIds);
// Strategy 2: Update existing documents
vectorStore.delete(existingIds);
vectorStore.add(documentsWithDuplicateIds);
}Pattern 2: Empty Search Results
Handle cases where no documents match the query or filters.
public List<Document> searchWithFallback(String query, String filter) {
SearchRequest strictRequest = SearchRequest.builder()
.query(query)
.filterExpression(filter)
.similarityThreshold(0.8)
.topK(10)
.build();
List<Document> results = vectorStore.similaritySearch(strictRequest);
if (results.isEmpty()) {
logger.warn("No results found with strict criteria, relaxing...");
// Strategy 1: Lower similarity threshold
SearchRequest relaxedRequest = SearchRequest.builder()
.query(query)
.filterExpression(filter)
.similarityThreshold(0.6) // Lower threshold
.topK(20) // More results
.build();
results = vectorStore.similaritySearch(relaxedRequest);
if (results.isEmpty()) {
// Strategy 2: Remove filters
logger.warn("Still no results, removing filters...");
results = vectorStore.similaritySearch(
SearchRequest.builder()
.query(query)
.topK(10)
.build()
);
}
}
return results;
}Pattern 3: Filter Expression Parse Errors
Handle invalid filter syntax from user input.
import org.springframework.ai.vectorstore.filter.FilterExpressionTextParser;
import org.springframework.ai.vectorstore.filter.FilterExpressionTextParser.FilterExpressionParseException;
public Filter.Expression parseFilterSafely(String filterText) {
if (filterText == null || filterText.isBlank()) {
return null;
}
FilterExpressionTextParser parser = new FilterExpressionTextParser();
try {
return parser.parse(filterText);
} catch (FilterExpressionParseException e) {
logger.error("Invalid filter syntax: {}", filterText, e);
// Strategy 1: Return null (no filter)
return null;
// Strategy 2: Use a default filter
// return new FilterExpressionBuilder().eq("status", "active").build();
// Strategy 3: Throw a user-friendly exception
// throw new InvalidFilterException("Invalid filter syntax: " + e.getMessage());
}
}Pattern 4: Unsupported Operations
Handle operations not supported by all vector stores.
public void deleteByFilter(String filter) {
try {
vectorStore.delete(filter);
} catch (UnsupportedOperationException e) {
logger.warn("Filter-based deletion not supported, falling back to ID-based deletion");
// Fall back to searching and deleting by ID
SearchRequest request = SearchRequest.builder()
.query("*") // Match all
.filterExpression(filter)
.topK(1000) // Batch size
.build();
List<Document> documentsToDelete = vectorStore.similaritySearch(request);
if (!documentsToDelete.isEmpty()) {
List<String> idsToDelete = documentsToDelete.stream()
.map(Document::getId)
.toList();
vectorStore.delete(idsToDelete);
logger.info("Deleted {} documents using ID-based fallback", idsToDelete.size());
}
}
}Pattern 5: I/O Errors (SimpleVectorStore)
Handle file I/O errors when loading/saving state.
public void saveWithRetry(SimpleVectorStore store, File file, int maxRetries) {
int attempts = 0;
while (attempts < maxRetries) {
try {
store.save(file);
logger.info("Vector store saved successfully");
return;
} catch (IOException e) {
attempts++;
logger.error("Failed to save vector store (attempt {}/{})",
attempts, maxRetries, e);
if (attempts >= maxRetries) {
// Last attempt failed
logger.error("All save attempts failed, data may be lost!");
// Strategy 1: Save to backup location
try {
File backup = new File(file.getParent(), "vectorstore-backup.json");
store.save(backup);
logger.info("Saved to backup location: {}", backup);
} catch (IOException backupError) {
logger.error("Backup save also failed", backupError);
}
throw new RuntimeException("Failed to save vector store after " +
maxRetries + " attempts", e);
}
// Wait before retry
try {
Thread.sleep(1000 * attempts); // Exponential backoff
} catch (InterruptedException ie) {
Thread.currentThread().interrupt();
}
}
}
}
public void loadWithFallback(SimpleVectorStore store, File primaryFile, File backupFile) {
try {
store.load(primaryFile);
logger.info("Loaded vector store from primary file");
} catch (IOException e) {
logger.warn("Failed to load from primary file, trying backup", e);
try {
store.load(backupFile);
logger.info("Loaded vector store from backup file");
} catch (IOException backupError) {
logger.error("Failed to load from backup file, starting with empty store", backupError);
// Continue with empty store
}
}
}Edge Cases
Edge Case 1: Null or Empty Metadata
Handle documents with missing or null metadata fields.
public void addDocumentSafely(String content, Map<String, Object> metadata) {
// Ensure metadata is not null
Map<String, Object> safeMetadata = metadata != null ?
new HashMap<>(metadata) : new HashMap<>();
// Add default values
safeMetadata.putIfAbsent("createdAt", Instant.now().toString());
safeMetadata.putIfAbsent("status", "active");
// Remove null values to avoid issues
safeMetadata.entrySet().removeIf(entry -> entry.getValue() == null);
Document doc = new Document(content, safeMetadata);
vectorStore.add(List.of(doc));
}
// When filtering, handle potential nulls
public List<Document> searchWithNullSafeFilter(String query, String category) {
String filter = category != null ?
"category == '" + category + "'" :
"category IS NOT NULL"; // Only return docs with category set
return vectorStore.similaritySearch(
SearchRequest.builder()
.query(query)
.filterExpression(filter)
.build()
);
}Edge Case 2: Empty Document List
Validate inputs before operations.
public void addDocuments(List<Document> documents) {
if (documents == null || documents.isEmpty()) {
logger.warn("Attempted to add null or empty document list");
return; // No-op
}
// Filter out documents with empty content
List<Document> validDocuments = documents.stream()
.filter(doc -> doc.getContent() != null && !doc.getContent().isBlank())
.toList();
if (validDocuments.isEmpty()) {
logger.warn("All documents had empty content, nothing to add");
return;
}
if (validDocuments.size() < documents.size()) {
logger.warn("Filtered out {} documents with empty content",
documents.size() - validDocuments.size());
}
vectorStore.add(validDocuments);
}Edge Case 3: Very Long Content
Handle documents that exceed embedding model limits.
public void addLongDocument(String content, Map<String, Object> metadata) {
final int MAX_TOKENS = 8000; // Example limit
if (estimateTokens(content) > MAX_TOKENS) {
logger.warn("Document exceeds token limit, chunking...");
// Strategy 1: Chunk the document
List<String> chunks = chunkContent(content, MAX_TOKENS);
for (int i = 0; i < chunks.size(); i++) {
Map<String, Object> chunkMetadata = new HashMap<>(metadata);
chunkMetadata.put("chunkIndex", i);
chunkMetadata.put("totalChunks", chunks.size());
Document chunk = new Document(chunks.get(i), chunkMetadata);
vectorStore.add(List.of(chunk));
}
return;
}
// Normal case
Document doc = new Document(content, metadata);
vectorStore.add(List.of(doc));
}
private int estimateTokens(String text) {
// Rough estimation: 1 token ≈ 4 characters
return text.length() / 4;
}
private List<String> chunkContent(String content, int maxTokens) {
List<String> chunks = new ArrayList<>();
int chunkSize = maxTokens * 4; // Convert tokens to characters
for (int i = 0; i < content.length(); i += chunkSize) {
int end = Math.min(i + chunkSize, content.length());
chunks.add(content.substring(i, end));
}
return chunks;
}Edge Case 4: Embedding Generation Failures
Handle failures during embedding generation.
public void addDocumentsWithRetry(List<Document> documents) {
for (Document doc : documents) {
int maxRetries = 3;
int attempt = 0;
while (attempt < maxRetries) {
try {
vectorStore.add(List.of(doc));
break; // Success
} catch (Exception e) {
attempt++;
logger.error("Failed to add document (attempt {}/{}): {}",
attempt, maxRetries, doc.getId(), e);
if (attempt >= maxRetries) {
// Log failure and continue with next document
logger.error("Failed to add document after {} attempts: {}",
maxRetries, doc.getId());
// Optionally: Store failed document for later processing
storeFailedDocument(doc, e);
} else {
// Wait before retry
try {
Thread.sleep(1000 * attempt);
} catch (InterruptedException ie) {
Thread.currentThread().interrupt();
break;
}
}
}
}
}
}
private void storeFailedDocument(Document doc, Exception error) {
// Store for later retry or manual review
logger.error("Storing failed document for manual review: {}", doc.getId());
// Implementation: write to file, database, or queue
}Edge Case 5: Concurrent Modifications
Handle concurrent access to vector stores.
@Service
public class ConcurrentSafeVectorStoreService {
private final VectorStore vectorStore;
private final ReentrantLock lock = new ReentrantLock();
public void addDocumentsSafely(List<Document> documents) {
lock.lock();
try {
vectorStore.add(documents);
} finally {
lock.unlock();
}
}
public List<Document> searchSafely(SearchRequest request) {
// Reads can be concurrent for most vector stores
return vectorStore.similaritySearch(request);
}
// For SimpleVectorStore with file persistence
public void saveStateSafely(SimpleVectorStore store, File file) {
lock.lock();
try {
store.save(file);
} finally {
lock.unlock();
}
}
}Edge Case 6: Type Mismatches in Filters
Ensure metadata types match filter expectations.
public void addDocumentWithTypeSafety(
String content,
String category,
int year,
double rating,
boolean featured) {
// Ensure correct types
Map<String, Object> metadata = Map.of(
"category", category, // String
"year", Integer.valueOf(year), // Integer (not String)
"rating", Double.valueOf(rating), // Double (not String)
"featured", Boolean.valueOf(featured) // Boolean (not String)
);
Document doc = new Document(content, metadata);
vectorStore.add(List.of(doc));
}
// Validate filter will work with metadata types
public boolean validateFilter(String filterExpression) {
try {
FilterExpressionTextParser parser = new FilterExpressionTextParser();
Filter.Expression expr = parser.parse(filterExpression);
// Additional validation: check if fields exist in schema
// Implementation depends on your schema management
return true;
} catch (FilterExpressionParseException e) {
logger.error("Invalid filter expression: {}", filterExpression, e);
return false;
}
}Edge Case 7: Memory Constraints (SimpleVectorStore)
Monitor and handle memory limits for in-memory stores.
@Service
public class MemoryAwareVectorStoreService {
private final SimpleVectorStore vectorStore;
private final int maxDocuments = 10000;
private final AtomicInteger documentCount = new AtomicInteger(0);
public void addDocumentWithCapacityCheck(Document doc) {
if (documentCount.get() >= maxDocuments) {
logger.warn("Vector store at capacity, triggering cleanup");
// Strategy 1: Remove oldest documents
removeOldestDocuments(1000);
// Strategy 2: Archive to disk
archiveAndClear();
// Strategy 3: Switch to persistent store
migrateToPersistentStore();
}
vectorStore.add(List.of(doc));
documentCount.incrementAndGet();
}
private void removeOldestDocuments(int count) {
// Implementation: Query for oldest documents and delete them
SearchRequest request = SearchRequest.builder()
.query("*")
.topK(count)
.filterExpression("1 == 1") // Match all
.build();
List<Document> oldest = vectorStore.similaritySearch(request);
List<String> idsToDelete = oldest.stream()
.sorted((a, b) -> {
String dateA = (String) a.getMetadata().get("createdAt");
String dateB = (String) b.getMetadata().get("createdAt");
return dateA.compareTo(dateB);
})
.limit(count)
.map(Document::getId)
.toList();
vectorStore.delete(idsToDelete);
documentCount.addAndGet(-idsToDelete.size());
}
private void archiveAndClear() {
// Save current state and clear
try {
File archive = new File("archive-" + System.currentTimeMillis() + ".json");
vectorStore.save(archive);
logger.info("Archived to {}", archive);
// Clear by creating new instance
// Note: This depends on your setup
} catch (IOException e) {
logger.error("Failed to archive", e);
}
}
private void migrateToPersistentStore() {
// Switch to production-grade vector store
logger.warn("Memory limit reached, consider migrating to persistent store");
// Implementation: Migrate data to Pinecone, PgVector, etc.
}
}Edge Case 8: Invalid UTF-8 Characters
Handle documents with encoding issues.
public void addDocumentWithSanitization(String content, Map<String, Object> metadata) {
// Remove or replace invalid UTF-8 characters
String sanitizedContent = content
.replaceAll("[^\\x00-\\x7F]", "") // Remove non-ASCII
.replaceAll("\\p{C}", ""); // Remove control characters
if (sanitizedContent.isBlank()) {
logger.warn("Document content became empty after sanitization");
return;
}
// Sanitize metadata values
Map<String, Object> sanitizedMetadata = metadata.entrySet().stream()
.filter(entry -> entry.getValue() != null)
.collect(Collectors.toMap(
Map.Entry::getKey,
entry -> {
Object value = entry.getValue();
if (value instanceof String str) {
return str.replaceAll("[^\\x00-\\x7F]", "");
}
return value;
}
));
Document doc = new Document(sanitizedContent, sanitizedMetadata);
vectorStore.add(List.of(doc));
}Defensive Programming
Input Validation
public class VectorStoreValidator {
public static void validateDocument(Document doc) {
if (doc == null) {
throw new IllegalArgumentException("Document cannot be null");
}
if (doc.getContent() == null || doc.getContent().isBlank()) {
throw new IllegalArgumentException("Document content cannot be empty");
}
if (doc.getMetadata() == null) {
throw new IllegalArgumentException("Document metadata cannot be null");
}
}
public static void validateSearchRequest(SearchRequest request) {
if (request == null) {
throw new IllegalArgumentException("SearchRequest cannot be null");
}
if (request.getQuery() == null || request.getQuery().isBlank()) {
throw new IllegalArgumentException("Query cannot be empty");
}
if (request.getTopK() < 0) {
throw new IllegalArgumentException("TopK must be >= 0");
}
if (request.getSimilarityThreshold() < 0.0 ||
request.getSimilarityThreshold() > 1.0) {
throw new IllegalArgumentException(
"Similarity threshold must be in range [0.0, 1.0]");
}
}
}Circuit Breaker Pattern
@Service
public class ResilientVectorStoreService {
private final VectorStore vectorStore;
private final CircuitBreaker circuitBreaker;
public ResilientVectorStoreService(VectorStore vectorStore) {
this.vectorStore = vectorStore;
this.circuitBreaker = CircuitBreaker.of("vectorStore",
CircuitBreakerConfig.custom()
.failureRateThreshold(50)
.waitDurationInOpenState(Duration.ofSeconds(30))
.build()
);
}
public List<Document> searchWithCircuitBreaker(SearchRequest request) {
return circuitBreaker.executeSupplier(() ->
vectorStore.similaritySearch(request)
);
}
}Testing Edge Cases
@Test
void testEmptyResults() {
VectorStore store = SimpleVectorStore.builder(embeddingModel).build();
SearchRequest request = SearchRequest.builder()
.query("nonexistent query")
.filterExpression("category == 'nonexistent'")
.build();
List<Document> results = store.similaritySearch(request);
assertThat(results).isEmpty();
}
@Test
void testDuplicateIds() {
VectorStore store = SimpleVectorStore.builder(embeddingModel).build();
Document doc1 = new Document("id-1", "content 1", Map.of());
Document doc2 = new Document("id-1", "content 2", Map.of());
store.add(List.of(doc1));
assertThrows(RuntimeException.class, () -> {
store.add(List.of(doc2));
});
}
@Test
void testInvalidFilter() {
FilterExpressionTextParser parser = new FilterExpressionTextParser();
assertThrows(FilterExpressionParseException.class, () -> {
parser.parse("invalid filter syntax <<<");
});
}Reference
- Quick Start Guide - Getting started
- Real-World Scenarios - Common patterns
- API Reference - Complete API
Install with Tessl CLI
npx tessl i tessl/maven-org-springframework-ai--spring-ai-vector-store@1.1.0