Dev

[Chenglong-MS]

This pull request introduces several important improvements and refactors, primarily focused on the SQL Server (MSSQL) data loader, Unicode handling for data export, and error handling for database connections. It also includes minor enhancements to frontend type definitions and UI feedback. The main themes are: migration from pyodbc to mssql-python, improved Unicode data safety, better error classification, and improved image handling for LLM requests.

MSSQL Loader: Migration to mssql-python and Simplification

• Switched the backend SQL Server connector from pyodbc to mssql-python, removing ODBC driver configuration and simplifying authentication logic. The loader now uses direct database connectivity, with no need for manual ODBC driver installation or driver selection. [1] [2] [3] [4] [5] [6]
• Updated the connection parameter forms, marking several options as "advanced" and removing the now-unnecessary driver parameter. [1] [2] [3]
• Simplified the authentication instructions and removed platform-specific ODBC installation steps; only the Azure CLI is now required for Entra ID authentication.

Unicode Handling and Data Export

• Added _repair_invalid_unicode to recursively sanitize and replace malformed Unicode in exported data, preventing serialization errors when exporting dataframes with invalid surrogate code points.

Database Error Handling and Classification

• Improved error classification for database connections: added detection and user guidance for Azure SQL firewall block errors, with a new DB_FIREWALL_BLOCKED error code. [1] [2] [3]

Image Degradation Logic for LLM Requests

• Enhanced image degradation logic: the system now only strips images and retries when the original request actually contained images, and when the upstream error is ambiguous. This avoids unnecessary retries and improves error handling for LLM requests. [1] [2] [3] [4]

Frontend and UI Enhancements

• Updated type definitions for connector parameter forms to support options and advanced fields, reflecting backend changes. [1] [2] [3]
• Improved the load plan UI to show prior-load status alongside actions, making the workflow clearer for users.