Microsoft.KernelMemory.MemoryDb.SQLServer 0.68.240722.1

Kernel Memory with Microsoft SQL Server

This folder contains tests for the MS SQL Server extension for Kernel Memory.

Please note that the connector should not be confused with a NL2SQL feature, e.g. the ability to query the content of a pre-existing SQL server. If you are looking for a solution that allows to import content from a SQL server and make it searchable please see How to index data from Azure SQL in Azure AI Search

Configuration

Configuration (appsettings.json):

  // ...
    "SqlServer": {
      "ConnectionString": "...",
    }
  // ...

Setup with Kernel Memory Builder / Dependency Injection

Method 1, simple applications:

var sqlServerConfig = cfg.GetSection("Services:SqlServer").Get<SqlServerConfig>()!;
var memory = new KernelMemoryBuilder()
    .WithSqlServerMemoryDb(sqlServerConfig)
    // .WithOpenAI(openAiConfig)
    // .WithAzureOpenAITextGeneration(azureOpenAITextConfiguration)
    // .WithAzureOpenAITextEmbeddingGeneration(azureOpenAIEmbeddingConfiguration)
    // .Build();

Method 2, injecting memory into an ASP.NET app:

builder.Services.AddSingleton<IKernelMemory>(sp =>
{
    KernelMemoryBuilder kernelMemoryBuilder = new()
        .WithSqlServerMemoryDb(builder.Configuration.GetConnectionString("DefaultConnection"))
        //... 

    return kernelMemoryBuilder.Build<MemoryServerless>();
});

KM Service setup

To run Kernel Memory service with SQL Server backend:

1. clone KM repository
2. add KernelMemory.MemoryStorage.SqlServer nuget to Service.csproj
3. edit the part using KernelMemoryBuilder adding the same .WithSqlServerMemoryDb(...) call mentioned in the previous paragraph, e.g.

   IKernelMemory memory = new KernelMemoryBuilder(appBuilder.Services)
   .FromAppSettings()
   .WithSqlServerMemoryDb(...)
   .Build();
   

Local tests with Docker

You can test the connector locally with Docker:

docker pull mcr.microsoft.com/mssql/server:2022-latest

docker run -it -p 1433:1433 --rm -e "MSSQL_SA_PASSWORD=00_CHANGE_ME_00" -e "ACCEPT_EULA=Y" \
    mcr.microsoft.com/mssql/server:2022-latest

...using the following connection string:

Server=tcp:127.0.0.1,1433;Initial Catalog=master;Persist Security Info=False;User ID=sa;Password=00_CHANGE_ME_00;MultipleActiveResultSets=False;TrustServerCertificate=True;Connection Timeout=30;

For more information about the SQL Server Linux container:

• https://learn.microsoft.com/sql/linux/quickstart-install-connect-docker
• https://devblogs.microsoft.com/azure-sql/development-with-sql-in-containers-on-macos/

Batch Upsert Feature

The SQL Server Memory DB now supports batching upsert operations, enhancing performance for bulk data operations. This feature allows for efficient insertion or updating of multiple records in a single operation.

Using Batch Upsert

To use the batch upsert feature, you can utilize the BatchUpsertAsync method. This method accepts an index name and an enumerable of MemoryRecord objects, performing upsert operations for all provided records in a batch.

Example:

var records = new List<MemoryRecord>
{
    new MemoryRecord("id1", new Dictionary<string, object> { { "key", "value1" } }),
    new MemoryRecord("id2", new Dictionary<string, object> { { "key", "value2" } })
};

await memory.BatchUpsertAsync("yourIndexName", records);

This method efficiently handles the insertion or updating of records, significantly improving performance for operations involving large datasets.