sqloom 0.2.0

sqloom

sqloom is a .NET tool for finding slow database work behind API requests in .NET applications. It can read Query Store from SQL Server or Azure SQL, replay API operations through app-specific harnesses, correlate captured SQL with Query Store, and generate tuning advice plus SQL proposal sidecars.

tune runs the full harness flow: replay -> observe -> correlate -> advise.

Install from NuGet.org

dotnet tool install --global sqloom
sqloom --help

To update an existing install:

dotnet tool update --global sqloom

Main commands

• observe: read recent Query Store data with an explicit --read-only-connection-string <connection-string>
• replay: replay API operations through an app-specific harness and capture SQL
• correlate: match replayed SQL back to a Query Store snapshot
• advise: send replay evidence, Query Store matches, and DACPAC-derived schema SQL to OpenAI
• tune: run the full replay -> observe -> correlate -> advise flow

Required inputs

observe and correlate require --read-only-connection-string <connection-string>. tune can use either --read-only-connection-string or a read-only connection string supplied by the harness session.

advise and tune use OpenAI for the advice step. Pass:

• --model-provider openai
• --openai-api-key <key>
• --sqlserver-dacpac-file <path> for advise, unless you pass the expert --sqlserver-schema-file <path> override

For tune, a harness can provide a default DACPAC path in its manifest. When advice runs from a DACPAC, Sqloom extracts model.sql with DacFx and persists the generated schema as sqlserver-schema.sql beside the advice artifacts.

SQL Server-backed replay harnesses can provide default DACPAC, seed script, replay profile, and Query Store profile values. CLI switches such as --sqlserver-dacpac-file <path>, --sqlserver-seed-sql-file <path>, --sqlserver-schema-file <path>, and --read-only-connection-string <connection-string> let you point at different inputs when needed.

Use the global --debug switch when you want stage-owned diagnostics on stderr. In particular, advise --debug prints readable, redacted OpenAI request and response payloads, and tune --debug cascades the same debug mode through observe, replay, correlate, and advise.

Example

sqloom replay .\tests\MyApi.Harness\MyApi.Harness.csproj --target "GET /api/orders/{id}"

Harness projects expose exactly one public non-abstract ISqloomApplication implementation for the app under test. The CLI accepts a harness project, harness assembly, solution, solution filter, or directory containing harness projects.

Install from a local feed

Install the tool from a local folder feed:

dotnet tool install --tool-path <tool-path> sqloom --add-source <local-feed-path> --ignore-failed-sources

The published sqloom tool package is sufficient for local folder-feed installs and public NuGet.org installs.

See the repository README for the full end-to-end sample and maintainer workflow:

Sqloom on GitHub