Agent Skills: SSDT CI/CD Best Practices 2025

🚨 CRITICAL GUIDELINES

Windows File Path Requirements

MANDATORY: Always Use Backslashes on Windows for File Paths

When using Edit or Write tools on Windows, you MUST use backslashes (\) in file paths, NOT forward slashes (/).

Examples:

• ❌ WRONG: D:/repos/project/file.tsx
• ✅ CORRECT: D:\repos\project\file.tsx

This applies to:

• Edit tool file_path parameter
• Write tool file_path parameter
• All file operations on Windows systems

Documentation Guidelines

NEVER create new documentation files unless explicitly requested by the user.

• Priority: Update existing README.md files rather than creating new documentation
• Repository cleanliness: Keep repository root clean - only README.md unless user requests otherwise
• Style: Documentation should be concise, direct, and professional - avoid AI-generated tone
• User preference: Only create additional .md files when user specifically asks for documentation

SSDT CI/CD Best Practices 2025

Overview

This skill provides comprehensive guidance on implementing modern CI/CD pipelines for SQL Server database projects using SSDT, SqlPackage, and contemporary DevOps practices.

Key Principles (2025 Recommended Approach)

1. State-Based Deployment (Recommended)

Definition: Source code represents the current database state, not migration scripts.

How it Works:

• All database objects (tables, procedures, views, functions) stored in separate .sql files
• SqlPackage automatically generates incremental deployment scripts
• Declarative approach: "This is what the database should look like"
• SSDT compares source to target and calculates differences

Advantages:

• Easier to maintain and understand
• No risk of missing migration scripts
• Git history shows complete object definitions
• Branching and merging simplified
• Rollback by redeploying previous version

Implementation:

# GitHub Actions example
- name: Build DACPAC (State-Based)
  run: dotnet build Database.sqlproj -c Release

- name: Deploy State to Target
  run: |
    sqlpackage /Action:Publish \
      /SourceFile:Database.dacpac \
      /TargetConnectionString:"${{ secrets.SQL_CONN }}" \
      /p:BlockOnPossibleDataLoss=True

Contrast with Migration-Based:

• Migration-based: Sequential scripts (001_CreateTable.sql, 002_AddColumn.sql)
• State-based: Object definitions (Tables/Customer.sql contains complete CREATE TABLE)

2. tSQLt Unit Testing (Critical for CI/CD)

Why tSQLt:

• Open-source SQL Server unit testing framework
• Write tests in T-SQL language
• Produces machine-readable XML/JSON results
• Integrates seamlessly with CI/CD pipelines

Key Features:

• Automatic Transactions: Each test runs in a transaction and rolls back
• Schema Grouping: Group related tests in schemas
• Mocking: Fake tables and procedures for isolated testing
• Assertions: Built-in assertion methods (assertEquals, assertEmpty, etc.)

Pipeline Abort on Failure:

# GitHub Actions with tSQLt
- name: Run tSQLt Unit Tests
  run: |
    # Deploy test framework
    sqlpackage /Action:Publish \
      /SourceFile:DatabaseTests.dacpac \
      /TargetConnectionString:"${{ secrets.TEST_SQL_CONN }}"

    # Execute tests and capture results
    sqlcmd -S test-server -d TestDB -Q "EXEC tSQLt.RunAll" -o test-results.txt

    # Parse results and fail pipeline if tests fail
    if grep -q "Failure" test-results.txt; then
      echo "Unit tests failed!"
      exit 1
    fi

    echo "All tests passed!"

- name: Deploy to Production (only runs if tests pass)
  run: |
    sqlpackage /Action:Publish \
      /SourceFile:Database.dacpac \
      /TargetConnectionString:"${{ secrets.PROD_SQL_CONN }}"

Test Structure:

-- tSQLt test example
CREATE SCHEMA CustomerTests;
GO

CREATE PROCEDURE CustomerTests.[test Customer Insert Sets Correct Defaults]
AS
BEGIN
    -- Arrange
    EXEC tSQLt.FakeTable 'dbo.Customers';

    -- Act
    INSERT INTO dbo.Customers (FirstName, LastName, Email)
    VALUES ('John', 'Doe', 'john@example.com');

    -- Assert
    EXEC tSQLt.AssertEquals @Expected = 1,
                             @Actual = (SELECT COUNT(*) FROM dbo.Customers);
    EXEC tSQLt.AssertNotEquals @Expected = NULL,
                                @Actual = (SELECT CreatedDate FROM dbo.Customers);
END;
GO

-- Run all tests
EXEC tSQLt.RunAll;

Azure DevOps Integration:

- task: PowerShell@2
  displayName: 'Run tSQLt Tests'
  inputs:
    targetType: 'inline'
    script: |
      # Execute tSQLt tests
      $results = Invoke-Sqlcmd -ServerInstance $(testServer) `
                                -Database $(testDatabase) `
                                -Query "EXEC tSQLt.RunAll" `
                                -Verbose

      # Check for failures
      $failures = $results | Where-Object { $_.Class -eq 'Failure' }
      if ($failures) {
        Write-Error "Tests failed: $($failures.Count) failures"
        exit 1
      }

3. Windows Authentication Over SQL Authentication

Security Best Practice: Prefer Windows Authentication (Integrated Security) for CI/CD agents.

Why Windows Auth:

• No passwords stored in connection strings
• Leverages existing Active Directory infrastructure
• Service accounts with minimal permissions
• Audit trail via Windows Security logs
• No credential rotation needed

Implementation:

Self-Hosted Agents (Recommended):

# GitHub Actions with self-hosted Windows agent
runs-on: [self-hosted, windows, sql-deploy]

steps:
  - name: Deploy with Windows Auth
    run: |
      sqlpackage /Action:Publish \
        /SourceFile:Database.dacpac \
        /TargetConnectionString:"Server=prod-sql;Database=MyDB;Integrated Security=True;" \
        /p:BlockOnPossibleDataLoss=True

Azure DevOps with Service Connection:

- task: SqlAzureDacpacDeployment@1
  inputs:
    authenticationType: 'integratedAuth'  # Uses Windows Auth
    serverName: 'prod-sql.domain.com'
    databaseName: 'MyDatabase'
    dacpacFile: '$(Build.ArtifactStagingDirectory)/Database.dacpac'

Alternative for Cloud Agents (Azure SQL):

# Use Managed Identity instead of SQL auth
- name: Deploy with Managed Identity
  run: |
    sqlpackage /Action:Publish \
      /SourceFile:Database.dacpac \
      /TargetConnectionString:"Server=tcp:server.database.windows.net;Database=MyDB;Authentication=ActiveDirectoryManagedIdentity;" \
      /p:BlockOnPossibleDataLoss=True

Never Do This:

# BAD: Plain text SQL auth password
TargetConnectionString: "Server=prod;Database=MyDB;User=sa;Password=P@ssw0rd123"

If SQL Auth Required:

# Use secrets/variables (least preferred method)
- name: Deploy with SQL Auth (Not Recommended)
  run: |
    sqlpackage /Action:Publish \
      /SourceFile:Database.dacpac \
      /TargetServerName:"${{ secrets.SQL_SERVER }}" \
      /TargetDatabaseName:"${{ secrets.SQL_DATABASE }}" \
      /TargetUser:"${{ secrets.SQL_USER }}" \
      /TargetPassword:"${{ secrets.SQL_PASSWORD }}" \
      /p:BlockOnPossibleDataLoss=True
  # Still not as secure as Windows Auth!

4. Version Control Everything

What to Version:

DatabaseProject/
├── Tables/
│   ├── Customer.sql
│   └── Order.sql
├── StoredProcedures/
│   └── GetCustomerOrders.sql
├── Tests/
│   ├── CustomerTests/
│   │   └── test_CustomerInsert.sql
│   └── OrderTests/
│       └── test_OrderValidation.sql
├── Scripts/
│   ├── Script.PreDeployment.sql
│   └── Script.PostDeployment.sql
├── Database.sqlproj
├── Database.Dev.publish.xml
├── Database.Prod.publish.xml
└── .gitignore

.gitignore:

# Build outputs
bin/
obj/
*.dacpac

# User-specific files
*.user
*.suo

# Visual Studio folders
.vs/

# Never commit credentials
*.publish.xml.user

Check-in Requirements:

• Require code review for database changes
• Mandate comments on all commits
• Run automated tests before merge
• Enforce naming conventions via branch policies

5. Deployment Reports Always Required

Before Production Deployment:

- name: Generate Deployment Report
  run: |
    sqlpackage /Action:DeployReport \
      /SourceFile:Database.dacpac \
      /TargetConnectionString:"${{ secrets.PROD_SQL_CONN }}" \
      /OutputPath:deploy-report.xml \
      /p:BlockOnPossibleDataLoss=True

- name: Parse and Review Report
  run: |
    # Extract key metrics from XML
    echo "=== DEPLOYMENT REPORT ==="
    # Parse XML for operations count
    # Check for data loss warnings
    # Display to user or post to PR

- name: Require Manual Approval
  uses: trstringer/manual-approval@v1
  with:
    approvers: database-admins
    minimum-approvals: 1
    instructions: "Review deploy-report.xml before approving"

- name: Deploy After Approval
  run: |
    sqlpackage /Action:Publish \
      /SourceFile:Database.dacpac \
      /TargetConnectionString:"${{ secrets.PROD_SQL_CONN }}"

6. Environment Promotion Strategy

Standard Flow: Dev → QA → Staging → Production

Consistent Deployment Options:

# Define environment-specific properties
environments:
  dev:
    blockOnDataLoss: false
    dropObjectsNotInSource: true
    backupBeforeChanges: false

  qa:
    blockOnDataLoss: true
    dropObjectsNotInSource: false
    backupBeforeChanges: true

  staging:
    blockOnDataLoss: true
    dropObjectsNotInSource: false
    backupBeforeChanges: true

  production:
    blockOnDataLoss: true
    dropObjectsNotInSource: false
    backupBeforeChanges: true
    requireApproval: true

Complete CI/CD Pipelines (GitHub Actions + Azure DevOps)

Full working pipeline YAMLs for GitHub Actions and Azure DevOps Pipelines (DACPAC build, drift detection, deploy reports, environment promotion, SqlPackage publish, approval gates) live in references/complete-pipelines.md. Load that reference when wiring CI/CD for an SSDT project.

Best Practices Checklist

Source Control

• [ ] All database objects in source control
• [ ] .gitignore configured for build outputs
• [ ] No credentials committed
• [ ] Test scripts versioned separately
• [ ] Branching strategy defined (gitflow, trunk-based, etc.)

Testing

• [ ] tSQLt framework deployed
• [ ] Unit tests cover critical stored procedures
• [ ] Tests grouped logically in schemas
• [ ] Pipeline aborts on test failure
• [ ] Test results published to dashboard

Security

• [ ] Windows Authentication used for CI/CD
• [ ] Service accounts follow principle of least privilege
• [ ] Secrets stored in Azure Key Vault / GitHub Secrets
• [ ] No plain text passwords
• [ ] Audit logging enabled

Deployment

• [ ] State-based deployment strategy
• [ ] Deployment reports generated before production
• [ ] Manual approval gates for production
• [ ] Backup before changes (production)
• [ ] BlockOnPossibleDataLoss enabled (production)
• [ ] DoNotDropObjectTypes configured
• [ ] Rollback plan documented

Monitoring

• [ ] Deployment logs captured
• [ ] Failed deployments trigger alerts
• [ ] Performance metrics tracked
• [ ] Schema drift detection automated

Resources

• tSQLt Official Site
• Microsoft.Build.Sql Documentation
• SqlPackage Reference
• Azure DevOps SQL Tasks
• GitHub Actions for SQL