RuVector PostgreSQL Bridge Examples

Comprehensive examples demonstrating the RuVector PostgreSQL Bridge plugin features.

Prerequisites

• Node.js 18+
• Docker and Docker Compose
• TypeScript / ts-node

Quick Start

1. Start PostgreSQL with pgvector

cd examples/ruvector
docker compose up -d

This starts:

• PostgreSQL 16 with pgvector extension on port 5432
• Adminer (database UI) on port 8080

2. Install Dependencies

# From the plugins root directory
npm install
npm run build

3. Run Examples

# Basic vector operations
npx ts-node examples/ruvector/basic-usage.ts

# Semantic code search
npx ts-node examples/ruvector/semantic-search.ts

# Attention mechanisms
npx ts-node examples/ruvector/attention-patterns.ts

# Graph neural networks
npx ts-node examples/ruvector/gnn-analysis.ts

# Hyperbolic embeddings
npx ts-node examples/ruvector/hyperbolic-hierarchies.ts

# Self-learning optimization
npx ts-node examples/ruvector/self-learning.ts

# Large-scale streaming
npx ts-node examples/ruvector/streaming-large-data.ts

# Quantization methods
npx ts-node examples/ruvector/quantization.ts

# Transaction patterns
npx ts-node examples/ruvector/transactions.ts

Examples Overview

1. basic-usage.ts

Getting started with RuVector PostgreSQL Bridge:

• Connecting to PostgreSQL
• Creating collections with HNSW indexes
• Inserting and searching vectors
• Batch operations
• Update and delete operations

Expected output:

RuVector PostgreSQL Bridge - Basic Usage Example
================================================

1. Connecting to PostgreSQL...
   Connected successfully!

2. Creating collection "documents"...
   Collection created!

3. Inserting vectors...
   Inserted: doc-1
   ...

2. semantic-search.ts

Semantic code search implementation:

• Embedding code snippets
• Natural language queries
• Hybrid search (semantic + keyword)
• Relevance feedback / re-ranking

Key concepts:

• Code embeddings capture semantic meaning
• Natural language queries find relevant code
• Hybrid scoring combines multiple signals

3. attention-patterns.ts

Using attention mechanisms:

• Multi-head attention
• Self-attention
• Cross-attention (encoder-decoder)
• Causal attention (autoregressive)
• Flash attention simulation
• KV cache for inference

Key concepts:

• Different attention patterns for different use cases
• Memory and compute optimizations
• SQL generation for PostgreSQL execution

4. gnn-analysis.ts

Graph Neural Network analysis:

• Building code dependency graphs
• GCN (Graph Convolutional Network)
• GAT (Graph Attention Network)
• GraphSAGE for inductive learning
• Finding structurally similar modules

Key concepts:

• Code structure as a graph
• Learning from dependencies
• Structural similarity detection

5. hyperbolic-hierarchies.ts

Hyperbolic embeddings for hierarchies:

• File tree embeddings
• Class inheritance hierarchies
• Poincare ball model
• Hierarchy-aware distances

Key concepts:

• Hyperbolic space captures hierarchies better
• Nodes closer to origin = higher in hierarchy
• Distance reflects tree structure

6. self-learning.ts

Self-optimization features:

• Enabling the learning loop
• Query pattern recognition
• Auto-tuning HNSW parameters
• Anomaly detection
• EWC++ for preventing forgetting

Key concepts:

• Continuous learning from query patterns
• Automatic index optimization
• Pattern-based query prediction

7. streaming-large-data.ts

Handling large datasets:

• Streaming millions of vectors
• Backpressure handling
• Progress monitoring
• Memory-efficient processing

Key concepts:

• Async generators for streaming
• Semaphore-based backpressure
• Concurrent batch processing

8. quantization.ts

Memory optimization:

• Int8 scalar quantization (4x compression)
• Int4 scalar quantization (8x compression)
• Binary quantization (32x compression)
• Product Quantization (PQ)
• Recall vs compression trade-offs

Key concepts:

• Quantization reduces memory significantly
• Trade-off between compression and recall
• Different methods for different use cases

9. transactions.ts

ACID operations:

• Multi-vector atomic updates
• Savepoint usage
• Error recovery patterns
• Optimistic locking

Key concepts:

• Transactions ensure consistency
• Savepoints for partial rollbacks
• Retry with exponential backoff

Configuration

Environment Variables

# PostgreSQL connection
export POSTGRES_HOST=localhost
export POSTGRES_PORT=5432
export POSTGRES_DB=vectors
export POSTGRES_USER=postgres
export POSTGRES_PASSWORD=postgres

Docker Compose Services

Access Adminer at http://localhost:8080:

• System: PostgreSQL
• Server: postgres
• Username: postgres
• Password: postgres
• Database: vectors

Troubleshooting

Connection Refused

Ensure PostgreSQL is running:

docker compose ps
docker compose logs postgres

Extension Not Found

The pgvector extension should be auto-created. Verify:

SELECT * FROM pg_extension WHERE extname = 'vector';

Out of Memory

For large datasets, adjust PostgreSQL memory settings in docker-compose.yml:

command: >
  postgres
  -c shared_buffers=512MB
  -c work_mem=32MB

Slow Searches

Ensure HNSW index exists:

SELECT indexname FROM pg_indexes WHERE tablename = 'your_table';

Tune search parameters:

SET hnsw.ef_search = 100;  -- Increase for better recall

Performance Tips

1. Batch inserts for bulk loading (1000+ vectors at a time)
2. Use HNSW indexes for approximate nearest neighbor search
3. Tune ef_search based on recall requirements
4. Consider quantization for large datasets
5. Use connection pooling for concurrent access

Cleanup

Stop and remove containers:

docker compose down

# Remove data volumes too:
docker compose down -v

Resources

• pgvector Documentation
• RuVector Plugin Documentation
• HNSW Algorithm Paper
• Poincare Embeddings Paper