This project demonstrates how to set up and use SQL Server with vector embeddings, including:

• Restoring the StackOverflow sample database
• Generating and storing text embeddings using external models (Ollama)
• Load balancing and performance testing with Nginx

• Docker & Docker Compose
• Ollama

1. Clone the repository

   git clone https://github.com/your/repo.git
   cd ollama-lb-sql

2. Restore the StackOverflow Database

   The restore_stackoverflow.sql script attaches the database files.
   Make sure the MDF/NDF/LDF files are present in /var/opt/mssql/data/ inside the SQL Server container.

   CREATE DATABASE StackOverflow_Embeddings_Small
   ON 
       (FILENAME = N'/var/opt/mssql/data/StackOverflow2013_1.mdf'),
       (FILENAME = N'/var/opt/mssql/data/StackOverflow2013_2.ndf'),
       (FILENAME = N'/var/opt/mssql/data/StackOverflow2013_3.ndf'),
       (FILENAME = N'/var/opt/mssql/data/StackOverflow2013_4.ndf')
   LOG ON
       (FILENAME = N'/var/opt/mssql/data/StackOverflow2013_log.ldf')
   FOR ATTACH ;

3. Start the Services

   docker compose up -d

   This will start:

   • SQL Server
   • Nginx load balancer
   • Certificate generator

4. Configure Nginx for Connection Pooling

   The nginx.conf is set up for high concurrency and connection pooling.

   Key settings:

   • keepalive 256; in the upstream block
   • proxy_set_header Connection ""; in each location block

5. Generate Embeddings

   Use the provided T-SQL scripts (see vector-demos.sql) to generate and store embeddings via the external model endpoints.

6. Load Testing

   Use test.ps1 to simulate concurrent API calls and measure performance.

   pwsh ./test.ps1

   Adjust $totalWork and $threads in the script as needed.